The zebrafish Iroquois gene iro7 positions the r4/r5 boundary and controls neurogenesis in the rostral hindbrain

Anteroposterior (AP) patterning of the vertebrate brain begins during gastrulation in the newly formed neural plate. Signals from adjacent tissues activate the expression of transcription factor genes in distinct domains of the neural plate (reviewed by Lumsden and Krumlauf, 1996). The limits of these expression territories often prefigure morphological boundaries and may be the site of formation of secondary signalling centres that act to organise adjacent tissues(Larsen et al., 2001; Kobayashi et al., 2002)(reviewed by Lumsden and Krumlauf,1996; Wurst and Bally-Cuif,2001).

In the hindbrain, AP patterning involves a segmentation process that leads to the formation of seven transient bulges called rhombomeres (r) and establishes the reiterated organisation of cranial nerves. The rhombomeres are segmental units for neuronal differentiation and gene expression, and constitute cellular compartments (for reviews, see Lumsden and Krumlauf, 1996; Schneider-Maunoury et al.,1998; Moens and Prince,2002). Hindbrain segmentation is also crucial for patterning and migration of the neural crest, thereby influencing face and neck morphogenesis(for a review, see Trainor and Krumlauf,2000), and for the formation of the otic vesicle, the prospective inner ear (for a review, see Torres and Giraldez, 1998).

Rhombomere formation proceeds by successive steps. Before segmentation is morphologically conspicuous, regulatory genes such as the Hox genes hoxb1b (Hoxa1) and hoxb1a (Hoxb1), valentino (mafb – Zebrafish Information Network; the zebrafish orthologue of Mafb), a gene coding for a bZIP transcription factor, and krx20 (egr2b – Zebrafish Information Network; Egr2), a gene coding for a zinc-finger transcription factor,are activated in the hindbrain, in distinct territories but with undefined limits (Wilkinson et al.,1989; Murphy and Hill,1991; Cordes and Barsh,1994) (for reviews, see Schneider-Maunoury et al.,1998; Moens and Prince,2002). These transcription factors are involved both in the formation of different rhombomeres or groups of rhombomeres, and in the specification of their identity(Giudicelli et al., 2001; McClintock et al., 2001; Voiculescu et al., 2001; McClintock et al., 2002; Giudicelli et al., 2003)(reviewed by Morrison, 1998; Schneider-Maunoury et al.,1998). In a second step, the limits of gene expression sharpen and later correspond to morphologically conspicuous rhombomere boundaries(Irving et al., 1996; Moens and Prince, 2002). Boundary formation is triggered by cell segregation at the interfaces between adjacent prerhombomeric territories(Guthrie and Lumsden, 1991; Guthrie et al., 1993; Wizenmann and Lumsden, 1997). This cell-sorting mechanism is mediated by Eph/ephrin interactions(Xu et al., 1995; Xu et al., 1999). Finally, the establishment of a specific pattern of gene expression, including Hox genes,in each rhombomere specifies positional identity along the AP axis and the fate of neuronal derivatives (Bell et al.,1999; Jungbluth et al.,1999) (reviewed by Rijli et al., 1998; Schneider-Maunoury et al., 1998). Inter-rhombomeric signalling is also involved in rhombomere specification, and in this respect r4 plays an important role in adjacent rhombomeres (Graham and Lumsden,1996; Helmbacher et al.,1998; Marin and Charnay,2000; Maves et al.,2002; Walshe et al.,2002).

The mechanisms that lead to the activation of regulatory genes such as krx20 and val at precise positions along the AP axis and thereby to the formation of pre-rhombomeric territories during gastrulation are not well understood. In the posterior hindbrain, the expression of the homeobox gene vhnf1 (tcf2 – Zebrafish Information Network) is activated at the end of gastrulation, with a rostral limit that has been shown to lie within prospective r5(Sun and Hopkins, 2001; Wiellette and Sive, 2003). val and krx20 are activated at the beginning of somitogenesis, in prospective r5 and r6, and r3 and r5, respectively(Wilkinson et al., 1989; Cordes and Barsh, 1994). Recent studies have shown that, in zebrafish embryos, the activation of val in r5 and r6 and of krx20 in r5 depends both on vhnf1 and on FGF3/8 signalling from r4(Sun and Hopkins, 2001; Maves et al., 2002; Walshe et al., 2002; Wiellette and Sive, 2003). However, the mechanisms involved in the establishment of the vhnf1expression domain, and in particular in the positioning of its anterior limit,are not known.

We have investigated the function of a zebrafish gene of the Iroquois (Iro)family, iro7. Iro genes code for homeodomain transcription factors of the TALE (three amino-acid loop extension) superfamily(Burglin, 1997). They are characterised by a highly conserved, 12 amino acid long domain called the Irobox (Cavodeassi et al.,2001). Iro genes were first described in Drosophila,where they perform essential functions in the patterning of the eye/antenna and wing imaginal discs (for reviews, see Cavodeassi et al., 2001; Gomez-Skarmeta and Modolell,2002). Vertebrate Iro genes are involved in various embryonic patterning processes, such as heart, ectoderm and neural tube regionalisation(for reviews, see Cavodeassi et al.,2001; Gomez-Skarmeta and Modolell, 2002). They have also been shown to participate in the activation of proneural gene expression, both in Drosophila and vertebrates (Gomez-Skarmeta et al.,1996; Gomez-Skarmeta et al.,1998; Itoh et al.,2002).

iro7 is a divergent member of the Iro family. Its closest relatives are the members of the irx1/irx3 paralogous group,suggesting that iro7 is an orthologue of these genes that has diverged after duplication of the teleost genome(Lecaudey et al., 2001; Itoh et al., 2002). iro7 shows an AP regionally restricted expression in the neural plate as early as 70% epiboly. At this stage, it is expressed in a large bilateral stripe, covering the neural plate and the future neural crest and placodal regions, and encompassing the prospective midbrain and hindbrain territories along the AP axis (Lecaudey et al.,2001; Itoh et al.,2002). Another zebrafish Iro gene, iro1, shows a similar expression pattern at the same stage, but its caudal limit is anterior to that of iro7 (Itoh et al.,2002). iro7 is necessary for the determination of neurons of the trigeminal placode, and iro1 and iro7 play partially redundant roles in the formation of the midbrain-hindbrain boundary(Itoh et al., 2002).

In this paper, we study the formation of the prospective r4/r5 boundary at the end of gastrulation. We show that the position of this boundary is set up by mutual repression between two transcription factors, Iro7 and vHnf1. In addition, iro7 is required for neurogenesis in the rostral hindbrain. Thus, iro7 is involved in two different aspects of the specification of hindbrain neuronal derivatives: AP patterning and neurogenesis.

Zebrafish (Danio rerio) were raised and staged as previously described (Westerfield, 1994; Kimmel et al., 1995). The vhnf1^hi2169 mutant(Sun and Hopkins, 2001) and the Isl1-GFP transgenic (Higashijima et al., 2000) lines were described previously.

A cDNA encoding full-length vhnf1 was cloned by RT-PCR from total RNA extracted from six- to eight-somite stage embryos. iro7 and vhnf1 cDNAs encoding full-length proteins were subcloned into the CS2+ vector (Rupp et al.,1994). The iro7myc expression vector was made by cloning iro7 cDNA in the CS2+MT vector(Rupp et al., 1994). An inducible form of Iro7 was constructed by fusing the ligand binding domain of the human glucocorticoid receptor (hGR) (from pCS2mcs-hGR, a gift from U. Strähle and P. Blader), to the C-terminal end of Iro7. A mutant form of vhnf1, vhnf1^Q139E, was made by introducing a point mutation in the POU domain using the ExSite PCR-Based Site-Directed Mutagenesis Kit (Stratagene). ΔN-iro7 (a gift from A. Chitnis)codes for a modified Iro7 protein missing the first nine amino acids, thus preventing its hybridisation with Moz7(Itoh et al., 2002).

Capped RNAs were transcribed with SP6 RNA polymerase using the mMessage mMachine Kit (Ambion). An antisense morpholino (Gene-Tools, Inc., Oregon, USA)was designed to target iro7 (Moz7): 5′GGCATCCTTACTCCCTGAGCTCTGG 3′, as well as a control morpholino (Moz7m):5′ GGgATCgTTAgTCCgTGAcCTCaGG 3′, containing six mismatches. Morpholinos were injected at a concentration of 1 mM. In some RNA injections, nls-lacZ (75 ng/μl) or GFP (100 ng/μl) RNAs were added as lineage tracers. The translocation of the Iro7hGR protein into the nucleus was induced by transferring embryos into medium containing 10 μM Dexamethasone (Sigma D-4902) at 40% epiboly.

In situ hybridisation and immunohistochemistry were performed as previously described (Hauptmann and Gerster,1994). iro7 (HindIII, T7), ngn1(neurog1 – Zebrafish Information Network; EcoRI, T7), wnt1 (EcoRI, Sp6) (gifts from A. Chitnis), six3(EcoRI, T3) and vhnf1 (NotI, T7) (gifts from B. Thisse and C. Thisse), hoxb1a, hoxa2 and hoxb3(Prince et al., 1998), krox20 (Oxtoby and Jowett,1993), val (Moens et al., 1998), fgf3(Kudoh et al., 2001), fgf8 (Furthauer et al.,1997), and pax2.1 (pax2a – Zebrafish Information Network) (Krauss et al.,1991) DNAs were used as templates for making RNA probes. For sectioning, embryos were embedded in resin (JB4, Polysciences). For immunohistochemistry, the following antibodies were used: mouse anti-neurofilament 3A10 (DSHB) (Hatta,1992) and RMO44 (Zymed 13-0500)(Popperl et al., 2000)antibodies, mouse Islet 39.4D5 (DSHB)(Ericson et al., 1992), rabbit anti-β-galactosidase (Cappel 55976), rabbit anti-GFP (Molecular Probes A11122) and rabbit anti-Myc epitope (Upstate Biotechnology 06-549).

Capped iro7 RNA (20 ng/μl) was translated in vitro in the presence of increasing concentrations of Moz7 or Moz7m (0.4 μM to 80 μM)using [³⁵S] methionine in a Rabbit reticulocyte lysate(Promega).

In Fig. 4,‘ vhnf1 expression domain AP length’ corresponds to the distance between the anterior and posterior borders of the vhnf1expression domain, including the most anterior domain of weaker expression. P is the probability associated with Student’s t-test. The error bars correspond to the standard deviation.

In previous work, we showed that iro7 is expressed from 70%epiboly onward in a large bilateral stripe in the neural plate, corresponding to the future midbrain and hindbrain, and that at early somite stages its caudal limit is located in the r4/r5 region(Lecaudey et al., 2001). In order to determine more accurately the position of the caudal limit of iro7 expression domain and its relations to pre-rhombomeric territories, we performed double in situ hybridisation with probes for iro7, krx20, val, hoxb3 and vhnf1. vhnf1 is expressed in the caudal neural plate with a rostral limit that has been proposed to lie within r5 at early somite (s) stages (Sun and Hopkins, 2001; Wiellette and Sive, 2003). val is expressed from the tailbud stage onward in prospective r5 and r6 (Moens et al., 1998). krx20 is activated at the tailbud stage in prospective r3 and at the 1 s stage in prospective r5(Oxtoby and Jowett, 1993). hoxb3 is activated at the 1-2 s stage in the neural plate posterior to r5 and at the 3-4 s stage its expression overlaps r5(Prince et al., 1998). We found that vhnf1 expression was activated in the caudal neural plate at 70% epiboly. Up to 90% epiboly, its rostral limit remained slightly caudal to the caudal limit of iro7 expression(Fig. 1A). From the 95% epiboly to the 2 s stages, the rostral limit of vhnf1 expression abutted the caudal limit of iro7 expression(Fig. 1B-E). The caudal limit of iro7 expression also corresponded to the rostral limit of val expression from the tailbud stage onwards(Fig. 1F,G) and of hoxb3 from the 3-4 s stage onwards(Fig. 1H). At the 1 s stage, iro7 was expressed in prospective r3 and r4(Fig. 1I). At this stage, the caudal krx20 expression stripe (future r5) appeared just caudal to the iro7 expression limit (arrowheads in Fig. 1I,J) and within the vhnf1 expression territory (arrowheads in Fig. 1M,N). From the 3 s stage,as krx20 expression expanded in r5, vhnf1 was downregulated in this rhombomere (Fig. 1O)and then in r6 (Fig. 1P). While vhnf1 retracted posteriorly, iro7 expression became restricted to r4 (Fig. 1K,L). Therefore, our data show that, between 95% epiboly and 2 s, the iro7and vhnf1 expression territories are complementary to each other, and that their common limit of expression prefigures the r4/r5 boundary(summarised in Fig. 1Q).

To investigate the function of iro7 in the hindbrain, we used a morpholino antisense oligonucleotide to inhibit mRNA translation(Nasevicius and Ekker, 2000). We designed a morpholino encompassing the initiation codon of iro7mRNA (Moz7) and a derived control morpholino containing mutations at six different positions (Moz7m) to check for specificity. Moz7 was able to inhibit the translation of iro7 capped RNA in an in vitro translation assay in a concentration-dependent manner. Translation was totally abolished in the presence of 8 μM Moz7, while it was not affected by 80 μM Moz7m(Fig. 2A). Injection of Moz7 in zebrafish embryos at the one- to four-cell stage led to defects in brain morphology at 30 hours post fertilisation (hpf): the anteroposterior extent of the midbrain and hindbrain was reduced, and the isthmus was often malformed(Fig. 2B) when compared with controls (Fig. 2C). The otic vesicles were also reduced in size along the AP axis, taking up a round shape(Fig. 2B,C). In the course of our experiments, we never observed any difference between mock-injected, Moz7m injected and uninjected embryos, so they will hereafter be referred to as‘control embryos’.

As iro7 is initially expressed in a large stripe covering the midbrain and anterior hindbrain, we investigated whether these territories were affected in embryos lacking iro7 function, using molecular landmarks. During somitogenesis, pax2.1 is expressed in the otic vesicle, in the optic stalk and at the MHB(Krauss et al., 1991), and wnt1 is expressed at the MHB, in the dorsal neural tube and in the epiphysis (Krauss et al.,1992). In Moz7-injected embryos stained for pax2.1 at the 13 s stage, the domains located between the otic placode and the MHB, and between the MHB and the optic stalk, were shorter along the AP axis(Fig. 2D) when compared with control embryos (Fig. 2E). Similarly, double wnt1/krx20 staining showed that the domains between the MHB and r3, and between the MHB and the epiphysis, were reduced along the AP axis in Moz7-injected embryos (Fig. 2F,G). These data showed that in the absence of iro7function, a large domain comprising the midbrain and the anterior hindbrain was reduced in size.

In order to better characterise the anterior hindbrain defects, we analysed the expression of hoxa2, fgf8 and fgf3 at the end of gastrulation and/or beginning of somitogenesis. hoxa2 is expressed in future r2 and r3 from the 2 s stage onwards(Prince et al., 1998). fgf8 is expressed in the anterior hindbrain, in a large domain that resolves at the 1 s stage into domains at the MHB/r1, ventral r2 and r4(Maves et al., 2002; Walshe et al., 2002). fgf3 is activated in a transverse stripe in the hindbrain at 90%epiboly, and is expressed at a high level in r4 at early somite stages(Maves et al., 2002; Walshe et al., 2002). In Moz7-injected embryos, the expression domains of hoxa2 in r2-r3(Fig. 2H), fgf8 in MHB-r4 (Fig. 2J) and fgf3 in r4 (Fig. 2L,N)were reduced in intensity and AP extent when compared with control embryos(Fig. 2I,K,M,O), but none of them was totally absent.

We then investigated the phenotypes caused by the loss of iro7 function at its posterior expression border. In Moz7 injected embryos stained for krx20, r3 and r5 were differently affected. Whereas the r3 stripe was occasionally reduced, consistent with the reduction in size of the anterior hindbrain, the r5 stripe was always expanded(Fig. 2F-I; Fig 3A-D). In addition, the gap between the r3 and r5 stripes was strongly reduced, suggesting that the r5 domain of krx20 expression [krx20 (r5)] expanded anteriorly into r4. This phenotype was already detectable at the onset of krx20expression: whereas the r3 stripe was slightly thinner, the r5 stripe was stronger and cells expressing krx20 were present in r4(Fig. 3E-H). To evaluate more accurately the anterior expansion of r5 and the reduction of r4, we measured the AP length of r4 and r5 on a batch of flat-mounted 6 s stage embryos. The anterior expansion of the krx20 (r5) stripe was detected in 100% of the injected embryos (n=29), when compared to control embryos(n=11), and covered about half of r4 (mean increase of krx20(r5) AP length: 40%, P<0.01). Isolated krx20-expressing cells were often present in r4 (Fig. 3C,G). These cells were mostly localised medially in the neural plate and then ventrally in the neural tube, often forming a bridge of krx20-expressing cells between r3 and r5. To check if the expansion of krx20 (r5) expression was specifically due to the inhibition of iro7 translation by Moz7, we tried to rescue this phenotype by injecting ΔN-iro7, a modified form of iro7 that does not hybridise with the morpholino. The injection of ΔN-iro7together with the morpholino led to a recovery of the size of r5, which was not statistically different from that of control embryos [mean increase 4.7%(n=19) compared with 40% (n=29) for Moz7-injected embryos]. The number of isolated krx20-expressing cells in r4 was also significantly reduced in rescued embryos (data not shown).

val is expressed from bud stage onwards in r5 and r6, and is necessary for the activation of krx20 in r5(Moens et al., 1998). Moz7 injection led to an expansion of the val expression domain(Fig. 3I,K), when compared with controls (Fig. 3J,L). Double krx20/val in situ hybridisation experiments showed that this expansion occurred anteriorly, was always limited to r4 and coincided with the expansion of the krx20 expression domain(Fig. 3I,K). By contrast, the r6 domain of val expression did not seem affected(Fig. 3K). This confirmed that the expansion of r5 occurs anteriorly, at the expense of r4.

hoxb1a is activated at 90% epiboly in a broad domain with an anterior border at the r3/r4 boundary(Prince et al., 1998). Shortly after its activation, hoxb1a expression is upregulated in r4 and maintained at a high level in this rhombomere(McClintock et al., 2001)(Fig. 3N). In Moz7-injected embryos, hoxb1a expression was activated normally at 90% epiboly in the caudal neural plate (data not shown) but was not reinforced in r4 at the end of gastrulation, as it was in control embryos(Fig. 3M,N). Later, the AP extent of the r4 hoxb1a expression domain was reduced when compared with controls, and coincided with the gap between krx20-expressing cells in r3 and r5 (Fig. 3O,P). In this domain, cells lacking hoxb1a expression were frequently observed ventrally (Fig. 3O,inset in O). These gaps of hoxb1a expression coincided with the bridges of krx20-expressing cells in r4 (arrowheads in Fig. 3O).

In conclusion, the loss of iro7 function leads to an anterior expansion of r5 at the expense of r4: the krx20 (r5) and valexpression domains are expanded anteriorly, and the hoxb1a expression domain in r4 is reduced.

Ectopic expression of vhnf1 leads to ectopic activation of val and krx20 in r4, and to a reduction of hoxb1aexpression in this rhombomere (Sun and Hopkins, 2001; Wiellette and Sive, 2003), a phenotype very similar to that obtained after Moz7 injection. Therefore, we tested whether knocking-down iro7 had any effect on vhnf1 expression, by performing double in situ hybridisation experiments with probes for vhnf1 and iro7. In Moz7-injected embryos at 95% epiboly to tailbud stages, the vhnf1expression territory was expanded anteriorly by about one-third, while the territory expressing iro7 was reduced(Fig. 4A-E). We conclude from these experiments that iro7 is necessary for the repression of vhnf1 in the anterior hindbrain.

As shown above, iro7 is necessary for the repression of vhnf1 and, probably as a consequence, of val and krx20. In order to determine whether iro7 was sufficient to repress vhnf1, val and krx20 in the caudal hindbrain, we performed iro7 gain-of-function experiments. Injection of RNA coding for a Myc-tagged Iro7 protein (Iro7myc) led to frequent gastrulation defects(data not shown), hampering analysis of hindbrain patterning. Therefore, we also injected RNA coding for an inducible form of Iro7, Iro7hGR, which translocates into the nucleus upon dexamethasone (Dex) treatment, and treated the injected embryos with Dex at 40% epiboly. GFP RNA was co-injected as a tracer. In embryos injected with either iro7myc or iro7hGR, we observed a repression of vhnf1(Fig. 5A,B,D), val(Fig. 5F,H) and krx20(Fig. 5J,L,M). However, a high concentration of RNAs (40 ng/μl for iro7myc and 100 ng/μl for iro7hGR) was necessary to obtain these effects. Immunostaining with the anti-GFP or anti-Myc antibodies showed that the domain of repression of vhnf1, val and krx20 always correlated with the injected regions, although not all injected cells showed a phenotype(Fig. 5A,D,H,L,M). In a batch of injected embryos, the proportion of embryos showing a domain of repression relative to the embryos expressing GFP or Myc in the region of interest was 22/53 for vhnf1, 14/16 for val and 11/15 for krx20. None of the GFP-injected control embryos showed repression of vhnf1, val or krx20 expression(Fig. 5C,E,G,I,K,N). In addition, krx20 expression was repressed specifically in r5, even though GFP staining was present in both r3 and r5(Fig. 5L,M). Together, these results show that Iro7 is able to repress vhnf1, val and krx20 expression. However, this repression is not fully penetrant,suggesting that Iro7 requires co-factors that are either regionally restricted or present in limiting amounts.

As vhnf1 and iro7 expression domains are complementary at the end of gastrulation, vhnf1 could also be involved in positioning the r4/r5 boundary by repressing iro7 expression. To test this hypothesis, we analysed iro7 expression in vhnf1^hi2169 mutants, which carry a strong hypomorphic or null mutation in the vhnf1 gene(Sun and Hopkins, 2001). Analysis of krx20 expression allowed us to unambiguously identify homozygous mutants, as the vhnf1 mutation leads to a loss of krx20 expression in r5 (Sun and Hopkins, 2001). In these embryos, the caudal territory of iro7 expression expanded posteriorly when compared with control siblings (Fig. 6A-E). These data showed that vhnf1 is required for iro7 repression in r5. As the caudal expansion was nevertheless limited to about one rhombomere length, other factors present in the caudal neural plate must also be able to repress iro7 expression. Alternatively, iro7 activation may itself be restricted to this region of the hindbrain. We also observed such a limited caudal expansion in vhnf1^hi2169 homozygous embryos for fgf8 (data not shown), whereas a wider caudal expansion was observed for fgf3 (data not shown) and hoxb1a(Sun and Hopkins, 2001) (data not shown).

To further investigate the repressive activity of vhnf1 on iro7, we tested whether vhnf1 is sufficient to repress iro7 anterior to the r4/r5 boundary. We expressed vhnf1ectopically by RNA injection. Escherichia coli lacZ RNA was co-injected as a lineage tracer. vhnf1 ectopic expression resulted in a repression of iro7 (Fig. 6F,G). All the cells that had downregulated iro7expressed β-galactosidase, suggesting that vhnf1 represses iro7 in a cell-autonomous manner(Fig. 6G). To check for the specificity of this repression, we injected RNA coding for a modified form of vHnf1 with a mutation within the POU domain (vHnf1^Q139E). An equivalent mutation in the human HNF1 protein abolishes its DNA-binding ability (S. Cereghini and C. Masdeu, unpublished). Injection of vhnf1^Q139E RNA had no effect on iro7 expression(Fig. 6H). Thus, vhnf1is able to repress iro7 expression. We conclude from these experiments that vhnf1 is required for positioning the future r4-r5 boundary by repressing iro7 expression in r5.

We sought to determine whether the patterning defects observed after Moz7 injection had any consequence on hindbrain neuronal derivatives. Two types of primary neurons are readily identifiable in the hindbrain of early zebrafish embryos and follow a segmental pattern: reticulospinal (RS) neurons and motoneurons (Metcalfe et al.,1986; Hanneman et al.,1988; Chandrasekhar et al.,1997). We analysed the pattern of RS neurons by immunostaining with the anti-neurofilament antibody RMO44(Fig. 7A,B). This antibody identifies the hindbrain RS neurons with characteristic shapes and positions,such as Mauthner neurons in r4, and RoL2 neurons in r2 (arrowhead and arrow,respectively, in Fig. 7B)(Popperl et al., 2000). We found that r4-derived Mauthner neurons were always lost in Moz7-injected embryos (empty arrowhead in Fig. 7A), whereas r2-derived RoL2 neurons were in some cases also affected (empty arrow in Fig. 7A), although with a lower penetrance. We did not detect any modification in the pattern of RS neurons caudally to r4(Fig. 7A,B). To confirm this phenotype, we used two earlier markers of the Mauthner neurons: the anti-neurofilament antibody 3A10, which stains strongly Mauthner neurons at 30 hpf (Furley et al., 1990)(Fig. 7D), and val,which is expressed in Mauthner neurons at 20 hpf in addition to r5, r6 and associated neural crest cells (Moens et al., 1998) (Fig. 7F). These two markers confirmed the total loss of Mauthner neurons in iro7 morphants (empty arrowheads in Fig. 7C,E). In total,r4-derived Mauthner cells were lost in 94% of the Moz7-injected embryos(n=119).

The pattern of motoneurons was analysed using an antibody against the Isl1 protein (Fig. 7G,H) or by injecting Moz7 into Isl1-GFP transgenic embryos(Higashijima et al., 2000)(Fig. 7I-L). In zebrafish embryos, Vth (trigeminal) motoneurons originate in two discrete groups in r2 and r3 and migrate laterally in these two rhombomeres(Chandrasekhar et al., 1997). VIIth (facial) motoneurons originate in r4 and migrate caudally and then laterally to reach their final position in r6 and r7(Chandrasekhar et al., 1997; Higashijima et al., 2000; McClintock et al., 2001). The motor axons of the Vth and VIIth nerves (arrows in Fig. 7J) leave the hindbrain at the level of r2 and r4 to innervate the first and second branchial arches,respectively. VIth (abducens) motoneurons are born in r5 and r6. In contrast to the Isl1 antibody, Isl1-GFP does not label abducens motoneurons(Higashijima et al., 2000). In Moz7-injected embryos, there was an overall reduction in the population of trigeminal and facial branchiomotor neurons(Fig. 7G,I,K) when compared with controls (Fig. 7H,J,L). The motoneurons of the facial nucleus, that are born in r4, were however more affected: the size of this nucleus was reduced by 60 to 80%(Fig. 7K,L). As these neurons migrate caudally to reach their final destination in r6 and r7, this resulted in an overall perturbation of the motoneuronal organisation in the r5-r7 region (Fig. 7, compare G,K with H,L). Motoneurons of the IXth and Xth nerves and spinal motoneurons were not affected (Fig. 7I,K and data not shown). Motor axons could always be detected in the trigeminal nerve in iro7 morphants (arrow in Fig. 7I), whereas the facial motor nerve was missing in 75% of the Moz7-injected embryos (n=8) (Fig. 7I). Together these results showed that iro7 is essential for the specification of primary neurons in the anterior hindbrain, and in particular for r4-derived neurons.

We wondered whether these defects in the pattern of primary neurons could be the result of a downregulation of proneural gene expression. Indeed, recent data showed that iro7 is involved in the formation of the trigeminal ganglion and in the expression of the proneural gene ngn1 in the trigeminal placode; moreover, iro7 ectopic expression is able to activate ngn1 expression ectopically(Itoh et al., 2002). Consistently, we found that Moz7 injection led to a reduction of ngn1expression in the anterior hindbrain region at the onset of somitogenesis(Fig. 7M-P). In r4, the proneural clusters that give rise to RS interneurons (dorsal clusters, r4 RS)and to motoneurons (ventral clusters, r4 Mn) were absent after Moz7 injection(Fig. 7, compare M,O with N,P)(96% of the injected embryos, n=23).

In conclusion, the knockdown of iro7 leads to an overall reduction in the number of primary neurons derived from the r2 to r4 region, with a more pronounced effect in r4. Moz7 injection also leads to a loss or strong downregulation of ngn1 expression, strongly suggesting that the reduction of RS and motoneuronal populations is due to a function of iro7 in the activation of proneural gene expression in the anterior hindbrain.

In this paper, we investigated the function in hindbrain patterning of iro7, a gene coding for a homeodomain-containing transcription factor of the Iroquois family. We showed that the r4/r5 boundary forms at the interface between the expression territories of iro7 rostrally and vhnf1, another homeobox gene, caudally. Gain and loss-of-function experiments demonstrated that these two transcription factors position the boundary by mutual repression. Finally, in addition to its role in early patterning of the hindbrain, iro7 is required for neurogenesis in the r2 to r4 region.

Although hindbrain segmentation has been extensively studied, the mechanisms underlying its early AP patterning and, in particular, those involved in the formation and positioning of early pre-rhombomeric territories and of their boundaries are still poorly understood. In zebrafish, the first boundaries to appear morphologically are the r3/r4 and the r4/r5 boundary(Maves et al., 2002). Until now, the posterior border of the domain expressing hoxb1a at a high level (prospective r4) was the first evidence of the future r4/r5 boundary. iro7 is expressed from 70% epiboly in a transversal stripe in the neural plate (Lecaudey et al.,2001; Itoh et al.,2002) and we demonstrate in this paper that its posterior expression border corresponds to the future r4/r5 boundary. Thus, the posterior border of the iro7 expression domain represents an early limit in the developing hindbrain at the position of the r4/r5 boundary, in a way similar to hoxb1b (Hoxa1), whose anterior expression border prefigures the r3/r4 boundary at mid-gastrulation(Murphy and Hill, 1991; Prince et al., 1998; McClintock et al., 2001). Slightly later, vhnf1 expression is activated in the posterior neural plate and is required for the expression of val in r5 and r6 and of krx20 in r5 (Sun and Hopkins,2001). We show that iro7 and vhnf1 have strictly complementary expression domains between 95% epiboly and 2 s. Thereby, vhnf1 is the earliest gene expressed in the caudal neural plate in a domain adjacent to that of iro7 expression. Altogether, these results make iro7 and vhnf1 good candidates to set up the prospective r4/r5 boundary by mutual repression.

In this paper, we demonstrate that vhnf1, val and krx20(r5) expression territories are expanded anteriorly in iro7morphants, and that ectopic expression of iro7 represses vhnf1,val and krx20 expression in the posterior hindbrain. Thus, iro7 is indeed required to set up the position of the r4/r5 boundary by repressing vhnf1 expression anteriorly. In iro7 ectopic expression experiments, the repression of vhnf1, val and krx20 expression by iro7 is not fully penetrant, suggesting that Iro7 requires co-factors that are either regionally restricted or present in limiting amount. Consistent with this hypothesis, Iroquois proteins belong to the TALE superfamily of transcription factors and other members of this group, such as Meis and Pbx proteins, form multimeric complexes(Ferretti et al., 2000; Choe et al., 2002).

We propose that the prospective r4/r5 boundary is set up at the end of gastrulation/beginning of somitogenesis by mutual repression between two homeodomain transcription factors, Iro7 and vHnf1. However, the expression domains of these two genes are only transiently complementary, between the 95%epiboly and 2 s stages. Therefore, the maintenance of the boundary may later involve mutual repression between Iro7 and transcription factors such as Val,Krx20 and Hoxb3, which are downstream of vhnf1 and remain expressed in r5. According to this hypothesis, iro7 represses val and krx20 more efficiently than vhnf1. In addition, the initial positioning of this boundary is likely to depend mainly on iro7. Indeed, iro7 expression is established before vhnf1expression reaches its definitive anterior limit, and therefore cannot depend totally on vhnf1. Consistently, in vhnf1 mutants, the posterior expansion of iro7 expression domain is limited to the length of one rhombomere. Other factors, such as retinoids, could be involved in positioning the iro7 posterior boundary early on. According to this hypothesis, we observed a repression of iro7 expression after treatment with retinoic acid (data not shown). Therefore, the main function of vhnf1 repressive activity on iro7 could be to refine the r4/r5 boundary.

iro7 is a divergent member of the Iro family, more closely related to the amniote irx1/3 group of paralogues. irx3 is expressed in the mouse neural plate during gastrulation(Bosse et al., 1997; Bellefroid et al., 1998). At the beginning of somitogenesis, the caudal limit of the irx3expression domain corresponds to the anterior limit of vhnf1expression and to the prospective r4/r5 boundary (S.S.M., V.L. and S. Cereghini, unpublished). Thus, mutual repression between Iro and vHnfl transcription factors and its involvement in the establishment of the r4/r5 boundary may constitute a conserved mechanism among vertebrates.

Establishing boundaries by mutual repression between two transcription factors expressed in adjacent territories is a common theme in early brain patterning. In several cases, these boundaries act as secondary signalling centres (Araki and Nakamura,1999; Matsunaga et al.,2000; Kobayashi et al.,2002) (reviewed by Rhinn and Brand, 2001; Wurst and Bally-Cuif, 2001). Iroquois genes have been implicated in boundary formation both in Drosophila and vertebrates: in Drosophila,Iro genes act as dorsal selector genes in the eye/antenna imaginal disc and are involved in the formation of the DV organiser that prefigures the future equator in the adult eye (McNeill et al.,1997; Cavodeassi et al.,1999; Yang et al.,1999; Cavodeassi et al.,2000). In the chick forebrain, Irx3 is involved in the positioning of the zona limitans intrathalamica by mutual repression with Six3, another homeodomain transcription factor(Kobayashi et al., 2002). Our results show that iro7, despite its divergence, has a function similar to that of the other members of the family in positioning the r4/r5 boundary by mutual repression with another transcription factor.

Recent data suggest the presence of a novel signalling centre within the hindbrain, acting across the r4/r5 boundary(Maves et al., 2002; Walshe et al., 2002). In zebrafish embryos, fgf3 and fgf8 are both expressed early in r4 and are required for the expression of krx20 and val in r5 and r5-r6, respectively (Maves et al.,2002; Walshe et al.,2002). Fgf3/8 signalling from r4 is also essential to the formation of the otic vesicle (Kwak et al., 2002; Leger and Brand,2002; Maroon et al.,2002). The knockdown of iro7 leads to a partial mis-specification of r4 but does not result in a reduction of val and krx20 expression level. On the contrary, val expression domain in r5/r6 and krx20 expression domain in r5 are expanded anteriorly. This shows that the r4 signalling centre is still functional,despite the reduction of r4 in iro7 morphants. Examination of fgf3 and fgf8 expression in Moz7-injected embryos showed that the level of expression of these two genes in r4 is reduced, especially at early stages, but that their expression is not abolished. This result suggests that a reduced amount of Fgf3/8 signalling is sufficient to allow val and krx20 expression. It is consistent with the data obtained previously (Wiellette and Sive,2003), showing that vhnf1 ectopic expression leads to ectopic activation of val and krx20, even though it represses fgf8 expression in r4. Although FGF signalling in iro7 morphants is sufficient to activate val and krx20 expression, it may be too low for a correct specification of the otic vesicle, which is reduced and presents an abnormal rounded shape in Moz7-injected embryos.

What are the respective roles of iro7 and the r4 signalling centre in setting up the r4/r5 boundary? FGFs and iro7 have antagonistic functions on val and krx20 activation. However, both signals act at different levels of the molecular hierarchy. FGFs are necessary to activate val and krx20 expression in the vhnf1expressing territory, and are not involved in vhnf1 activation. iro7 is involved in repressing val and krx20expression but, at least in part, as a consequence of vhnf1repression. Thereby, iro7 is involved in an earlier step that is the positioning of the boundary. Nothing is known about the molecular cues involved in vhnf1 activation in the posterior neural plate. An interesting hypothesis would be that these cues are also present in future r4,and that iro7 is required to repress the activation of vhnf1in this rhombomere.

In this paper, we show that the anterior hindbrain is significantly reduced in the absence of Iro7. At the end of gastrulation, the anterior hindbrain markers gbx1 (not shown) and fgf8 are expressed both at a weaker level and in a reduced domain. Slightly later, the expression territories of hoxa2 in r2-r3, hoxb2 in r3-r4 (not shown)and krx20 in r3 are also reduced, but none of them is totally absent. Therefore, the absence of iro7 leads to an overall reduction of the anterior hindbrain but without any obvious mis-specification within this region.

We also demonstrate here that the knockdown of iro7 leads to a strong reduction in the ngn1-expressing proneural clusters in the anterior hindbrain, especially in r4, whereas the more caudal proneural clusters are not affected. Moreover, iro7 knockdown affects specific primary neuronal populations. The nuclei of the facial (VIIth) nerve and, to lesser extent of the trigeminal (Vth) nerve, are reduced, while the r4-derived Mauthner cells are always lost and the r2-derived RoL2 are also occasionally absent. Thus, in the absence of the Iro7 protein, neurogenesis is affected in the rostral hindbrain. Accordingly, iro7 was previously shown to be necessary for the formation of the trigeminal ganglia(Itoh et al., 2002).

The defects we observed in the differentiation of some neuronal subtypes could result from a change in AP specification or from a direct effect of iro7 on neurogenesis. We favour the second hypothesis for several reasons. First, the absence of r4 proneural clusters and of Mauthner neurons is unlikely to result only from the reduction of r4. Mauthner cells are indeed lost in 100% of the Moz7-injected embryos, although the transformation of r4 into r5 is never total and its extent is slightly variable from one embryo to the other. Second, we also observed a reduction of ngn1 expression in r2 clusters, as well as an occasional loss of r2-derived RS neurons. As we mentioned above, the iro7 knock-down leads to a reduction in size but not to a misspecification of the anterior hindbrain, so the neuronal defects in this region are likely to be linked to a role of iro7 in neurogenesis. Third, no ectopic r5-specific RS neurons were observed, a phenotype that would be expected if the loss of Mauthner cells was due to a pure change in AP identity. Finally, ectopic expression of iro7activates ngn1 and, in the neurectoderm, this activation has been proposed to result from a function of Iro7 as a transcriptional activator(Itoh et al., 2002). Thus, the activation of ngn1 expression by Iro7 may be direct.

In conclusion, we propose that iro7 has a dual function in the hindbrain: it is required for the positioning of the r4/r5 boundary by repressing vhnf1, and for neurogenesis in the anterior hindbrain,possibly by direct activation of ngn1 expression. This dual function is consistent with the successive roles found for Iro genes in the Drosophila wing imaginal disc(Gomez-Skarmeta et al., 1996; Leyns et al., 1996; Grillenzoni et al., 1998; Diez del Corral et al., 1999; Calleja et al., 2002). Like its Drosophila cognate genes, iro7 is required for successive steps of the patterning of an embryonic territory.

We are grateful to U. Strähle and P. Blader for providing us with the pCS2-hGR expression construct; to N. Hopkins and H. Okamoto for the gift of the vhnf1^hi2169 and Isl11-GFP fish lines,respectively; to A. Chitnis for the gift of the ngn1 probe and theΔ N-iro7 construct; to V. Prince for the gift of the Hox probes;and to B. Thisse and C. Thisse for providing us with the six3 and vhnf1 probes, and the full-length iro7. We thank F. Bouallague for fish care; S. Cereghini, M. Itoh and A. Chitnis for fruitful discussions and for sharing unpublished material and data; L. Bally-Cuif for fruitful discussions and warm support; and P. Charnay, C. Vesque, S. Cereghini and J. Ghislain for critical reading of the manuscript. This work was supported by funds from the INSERM, FRM, ARC and MENRT. V.L. was supported by a fellowship from MENRT.