A series of no isthmus (noi) alleles of the zebrafish pax2.1 gene reveals multiple signaling events in development of the midbrain-hindbrain boundary

In vertebrate embryos, the fate of progenitor cells in the neural tube is determined by their position with respect to cellular sources of inducing signals (Lumsden and Krumlauf, 1996). Transplantation experiments in chicken suggest that the midbrain-hindbrain boundary (MHB; or mesencephalic- metencephalic boundary, isthmus) may contain such a neuroepithelial organizer (Martinez et al., 1991; Marin and Puelles, 1994; see also reviews by Bally-Cuif and Wassef, 1995; Joyner, 1996; Puelles et al., 1996). Several transcription factors are required during development of the MHB, such as En1 and En2 (Wurst et al., 1994; Millen et al., 1994), Pax2 (Torres et al., 1995; Brand et al., 1996; Favor et al., 1996), Pax5 (Urbanek et al., 1994), Otx1 and Otx2 (Ang et al., 1996; Acampora et al., 1997) and Gbx2 (Wassarman et al., 1997). Among the secreted factors, the vertebrate homologue of Drosophila Wingless, Wnt1 (McMahon et al., 1992) and Fgf8, a member of the fibroblast growth factor family, are required for MHB development in mice (Crossley et al., 1996; Meyers et al., 1998) and zebrafish (Reifers et al., 1998). In spite of the identification of many factors that function in development of the MHB territory, it is unclear which aspects of development of this area are controlled by the various gene products, how this is linked to the generation of the organizer potential in this region, and when and how exactly the organizer is able to act in cell-fate determination of surrounding cells.

In this study, we focus on the function of zebrafish pax2.1 (formerly pax(zf-b); Krauss et al., 1991), a member of the family of transcription factors that includes pax2, pax5 and pax8 (the pax2/5/8 family) in mammals. The proteins in this family share a paired-type domain and a partial homeobox as DNA-binding motifs, an octapeptide for protein-protein interaction, and a transactivating/inhibiting domain at the carboxy terminus (Wehr and Gruss, 1996; Dörfler and Busslinger, 1996; Pfeffer et al., 1998). A targeted and a chemically induced null allele of murine Pax2 have different phenotypes of variable strength, probably due to different genetic backgrounds (Torres et al., 1995; Favor et al., 1996). The phenotype of homozygous mutants ranges from strong defects in development of midbrain, eye, ear and kidney (Favor et al., 1996) to nearly normal development of the MHB (Torres et al., 1995). Pax5 and Pax8, which are expressed in overlapping domains with Pax2 at the MHB (Nornes et al., 1990; Asano and Gruss, 1992; Plachov et al., 1990) may contribute to this variability. Inactivation of murine Pax5 leads, on its own, only to mildly abnormal development of the posterior midbrain and anterior cerebellum (Urbanek et al., 1994). When both Pax2 and Pax5 are inactivated, the resulting phenotype is more severe than in either single mutant, suggesting that the murine members of this family can functionally replace each other (Urbanek et al., 1997; Schwarz et al., 1997).

In zebrafish, lethal mutations in three genes identified in systematic mutagenesis screens affect development of the MHB. Homozygous acerebellar (ace; Brand et al., 1996) or spiel-ohne-grenzen (spg; Schier et al., 1996) embryos lack the MHB and the cerebellum, but retain a midbrain. acerebellar is a mutation in the zebrafish fgf8 gene (Reifers et al., 1998); the gene affected by spg is not known. Embryos homozygous mutant for strong no isthmus (noi) alleles lack the mid- hindbrain boundary and cerebellum, as well as some or all of the dorsal and ventral midbrain (Brand et al., 1996). One noi mutation, noi^th44a, is genetically linked to the pax2.1 gene, and in this allele a stop codon interrupts the pax2.1 reading frame, but leaves a large portion of the molecule intact, suggesting that the noi^th44a mutation is a hypomorphic allele of pax2.1 (Brand et al., 1996, and this paper).

A general property of organizer cell populations is that they control cell fate, via gene expression, at a distance. Because organizers are thought to produce morphogens that determine cell fate in a concentration-dependent manner, organizer activity is often sensitive to the functional level of gene products that are involved in controlling organizer function. The availability of six chemically induced alleles of no isthmus allowed us to clarify the requirement for pax2.1 activity, and suggested they might inactivate pax2.1 to different degrees, thus forming an ‘allelic series’. Generally, functional levels, assayed as gene copy number, are critical for several human and murine Pax genes, including Pax2 (Tassabehji et al., 1992, 1993; Hanson et al., 1994; Sanyanusin et al., 1995; Acampora et al., 1997; Dahl et al., 1997; Schwarz et al., 1997). In this study, we use molecular and phenotypic criteria to establish that the available no isthmus alleles form such an allelic series, ranging from a null allele to weak alleles, and examine the expression of potential target genes. Based on their expression, candidate target genes were the engrailed genes (eng1, eng2 and eng3) (Ekker et al., 1992), wnt1 (Molven et al., 1991), fgf8 (Reifers et al., 1998; Fürthauer et al., 1997) and the bHLH transcription factor her5 (Müller et al., 1996). Our analysis of the various alleles shows that multiple and sequential signalling events must act during development of the zebrafish MHB.

In the accompanying paper, isolation of three new zebrafish pax2/5/8 genes is described: pax5, pax8 and a second, pax2- like gene called pax2.2 (Pfeffer et al., 1998). While pax2.1 is already activated in the late gastrula, these additional genes are activated much later, between the 4- and 9-somite stage. Importantly, pax5 and pax8 expression at the MHB strictly depends on noi/pax2.1 function; in this tissue, inactivation of noi/pax2.1 can therefore be considered as functionally equivalent to triple inactivation of pax2/5/8 genes. These findings clarify the reasons for the apparent phenotypic differences between mice and zebrafish pax2(.1) mutants (Brand et al., 1996; Pfeffer et al., 1998; see also Discussion).

Zebrafish were raised and kept under standard laboratory conditions at about 27°C (Westerfield, 1994). Mutant carriers were identified by random intercrosses. To obtain embryos showing the mutant phenotype, two heterozygous carriers for a mutation were crossed to one another. Typically, the eggs were spawned synchronously at dawn of the next morning, and embryos were collected, sorted, observed and fixed at different times of development at 28.5°C. In addition, morphological features were used to determine the stage of the embryos, as described by Kimmel et al. (1995). In some cases, 0.2 mM phenylthiourea (PTU) was added to prevent melanization. Isolation and initial characterization of no isthmus is described in Brand et al. (1996). For survival tests, embryos were dechorionated on the day of birth, kept at low density (approx. 20 per 9 cm dish) and the E2 medium was frequently changed.

Digoxigenin- or fluorescein-labelled RNA probes were prepared from linearized templates using an RNA labelling and detection kit (Boehringer). Hybridization and detection with anti-digoxygenin or anti-fluorescein antibodies coupled to alkaline phosphatase (Boehringer) is described in Reifers et al. (1998). To determine overlap in double stains with BM purple and FastRed fluorescent substrate (Boehringer), the BM purple reaction was allowed to proceed until it quenched but did not obliterate the fluorescent FastRed signal. Stained embryos were dissected and thick sections prepared with sharpened tungsten needles; these were mounted in glycerol and photographed on a Zeiss axiophot. Composites were assembled with Adobe Photoshop. Probes and wild-type expression patterns are described elsewhere: Eng1-3: Ekker et al. (1992); Pax2.1: Krauss et al. (1991); Her5: Müller et al. (1996); wnt1: Molven et al. (1991); fgf8: Reifers et al. (1998).

Total RNA was prepared by the hot phenol method (Brown and Kafatos, 1988). cDNA was isolated by RT-PCR with nested primers flanking the coding region in at least two independent amplifications from pools of homozygous noi embryos, subcloned into pCRII (Invitrogen) and sequenced using the T7 Sequenase kit (Amersham).

The wild-type, noi^tm243a and noi^tu29a coding regions were cloned into pQE60 or pQE30 (Qiagen). Expression and purification of recombinant proteins was performed according to manufacturer’s protocol. Briefly, recombinant fusion proteins contain a 6×histidine tag at the N (pQE30) and C terminus (pQE60), which allowed binding and purification on Ni²⁺-charged Sepharose resin under denaturing conditions using a stepwise imidazole gradient. Proteins were renaturated by dialysis (8, 6, 4, 2 and 1 M urea) into Z-buffer (25 mM Tris, pH 7.8, 20% glycerol, 12.5 mM MgCl, 0.1 M KCl, 1 mM DTT) at 4°C. Wild-type and mutant proteins were prepared in parallel under identical conditions, and examined on Coomassie blue-stained gels to ensure purity (not shown).

The sequences of the CD19-2(A-ins.) (Kozmik et al., 1992) and BSI+II (Song et al., 1996) Pax2 binding sites have been described. Double-stranded oligonucleotides (100 pmol) were end-labelled with T4 polynucleotide-kinase and [³²P]dATP (6 μCi/μl). Binding reactions were performed in 10 μl at 4°C for 30 minutes and contained an empirically determined amount of affinity-purified Pax2.1 protein, 100 ng of poly(dI·dC), ³²P-labelled probe (40,000- 50,000 cpm), 0.5 mg/ml BSA and Z-buffer. For the competition experiments up to 10-fold excess of unlabelled probe (1 nmol) was added to the reaction. Binding reactions were examined on 6% non- denaturing polyacrylamide gels, in 0.5×TBE at 30 mA for 70 minutes.

Mutants homozygous for the three strong alleles noi^tu29a, noi^th44a and noi^tm243a, lack the midbrain, MHB and cerebellum, whereas in the two weak alleles, noi^ty31a and noi^tb21, some midbrain is still formed (Brand et al., 1996; Fig. 1A-D). In addition, noi mutants show defects in formation of the optic stalk, the inner ear and pronephric duct (Brand et al., 1996; Macdonald et al., 1997), all areas that show pax2.1 expression. The strength of the noi phenotype correlates with the ability to survive under optimal conditions: for the strongest allele, 50% of the embryos died at day 7, whereas embryos mutant for the weak alleles survive about 4 days longer (Fig. 1G). On day 7, the mutants show severe oedema of the pericardium and gut epithelium; the heart is malfunctioning and the embryos never feed (Fig. 1E,F).

To characterize the mutations affecting the noi/pax2.1 gene, we analyzed the genomic organisation of pax2.1. On the basis of the human pax2 gene structure (Sanyanusin et al., 1996), we designed primers to amplify the exon-intron boundaries by PCR from three overlapping genomic phage clones that cover the complete pax2.1 locus (A. Picker and M. Brand, unpublished) and sequenced them (accession numbers AF067530 to AF067541 and AF073442 to AF073445). We find that zebrafish pax2.1 has 12 exons, 10 at the predicted positions and two (5.1 and 7.1) that show no sequence homology to pax2 exons of other species (Fig. 2A). Exon 7.1 consists of 67 bp and contains a stop codon (Fig. 2F). The predicted truncated protein would lack the transactivating and inhibitory domains (Dörfler and Busslinger, 1996) encoded by exons 8, 9 and 10. Exon 7.1, but without a stop codon, has also been found in Xenopus (Heller and Brändli, 1997).

While analysing the no isthmus mutations by RT-PCR (see below), we found a high number of pax2.1 cDNA variants. Exons 5.1, 6, 7 and 7.1 are differentially spliced in different combinations, as shown in Fig. 2B. Alternative splicing has been reported for exon 5.1 in zebrafish (Krauss et al., 1991), mouse and human (Sanyanusin et al., 1996), and for exons 6, 7 and 7.1 in Xenopus (Heller and Brändli, 1997). Furthermore, in two variants an alternative 3′ splice acceptor site is located near the 5′-ends of exons 2 and 9 (Fig. 2C,D). For exon 9, usage of the alternative splice acceptor site leads to a frameshift, with an alternative stop codon in the 3′UTR (Fig. 2E) and a protein lacking part of the inhibitory domain which could therefore be constitutively active. Other cDNAs revealed that additional exon(s) are probably located between exons 7 and 8 (not shown), and additional uncharacterized splice products are seen in the RT-PCR (Fig. 4D, arrowheads). We have observed 12 of 64 theoretically possible splice variants. Among these, the two major splice variants are those either containing or lacking exon 5.1 (Fig. 2B(1),(2); Fig. 4D, arrows). The significance of the multiple splice variants is not clear, but may be functionally relevant given the dynamics and complexity of pax2.1 expression and requirement.

We cloned and sequenced the coding region of pax2.1 for all six known noi alleles, and found various types of mutations in the different alleles, providing further evidence that the noi phenotype results from mutations in the pax2.1 gene. Due to the mutagen used in the zebrafish screen we expected point mutations, which we found for three alleles, predicting the mutant proteins shown schematically in Fig. 3A.

In noi^tu29a, the C→T transition converts codon 139 into a stop codon (Fig. 3A,B). This leads to a C-terminally truncated protein lacking the transactivation and inhibitory domains (T/I, Dörfler and Busslinger, 1996), and six amino acids of the DNA binding paired domain. The same mutation was found in noi^ty22b, which is therefore probably a re-isolate of noi^tu29a (not shown). As shown previously (Brand et al., 1996), in noi^th44a the G→T transversion produces a stop codon (Fig. 3A), leading to a protein with an intact DNA binding domain that lacks the complete C terminus. In both cases the Pax2 protein is severely affected, consistent with the observed strong phenotype in both alleles. The predicted structure is consistent with the failure of homozygous mutants to stain with an antibody directed against the C-terminal part of the protein (Brand et al., 1996).

The third point mutation (G→T transversion) in the weak noi^ty31a allele transforms glycine 75 into valine (Fig. 3A,C). Since this highly conserved residue is located in the turn of the DNA recognition helix in the N-terminal helix-turn-helix (HTH) motif within the paired domain (Xu et al., 1995), loss of this amino acid probably impairs the DNA binding activity of the mutant protein.

In cDNA from homozygous noi^tm243a mutants we find an in- frame deletion of 18 bp at the 5′ end of exon 3 (Fig. 4A). This deletion leads to the loss of six amino acids in helix II of the N-terminal paired domain, but leaves the rest of the protein unaffected. As expected, the mutant protein can be detected with the antibody against the C terminus (Brand et al., 1996). Since helix II of the first HTH motif in the paired domain is crucial for the DNA/protein interaction of paired-type transcription factors (Xu et al., 1995), the noi^tm243a could theoretically cause a complete loss of function, consistent with the observed strong phenotype. Further analysis shows, however, that the deletion is created by splicing that occurs in most, but not all, splicing events. The 18 bp deleted in the cDNA of exon 3 are still present in genomic DNA, which however carries a mutation in the 3′-splice acceptor site preceding exon 3 (Fig. 4B). The highly conserved AG (Padgett et al., 1986) is changed into a TG; this point mutation probably abolishes activity of the original site, and allows usage of a cryptic splice acceptor site in exon 3, thus deleting 18 bp in the cDNA. RT-PCR analysis with primers flanking the region of the deletion shows that the predominant, smaller band diagnostic for transcripts with the deletion is seen only in mutant cDNA and is thus not a normal splice variant. In addition, most, but not all transcripts contain the deletion, explaining the slightly weaker phenotype of this allele (Fig. 4E, arrows).

In homozygous noi^tb21 mutants all pax2.1 transcripts lacked exon 7, both by sequencing of cDNAs and RT-PCR analysis, suggesting that the deletion is also generated by aberrant splicing (Fig. 4C,D). The weak phenotype of noi^tb21 suggests that the mutated Pax2.1 protein retains some activity.

To understand the requirement for pax2.1 in development it is essential to determine its null phenotype, particularly in view of the variable phenotype of mouse pax2 mutants. Critical for the biological function of a transcription factor is its ability to specifically bind DNA and interact with the transcriptional machinery. Our above analysis suggested noi^tu29a and noi^tm243a as the best candidates for null alleles, since both have deletions in the DNA binding paired domain; in particular noi^tu29a protein was likely to be function-less since the T/I domains are deleted as well.

We have used DNA electrophoretic mobility shift assays (EMSA) with purified recombinant wild-type and mutant proteins to test whether noi^tu29a and noi^tm243a could be null alleles. As binding sites we used known Pax2 binding sites, such as BSI and BSII in the murine En2 promoter (Song et al., 1996) and an optimized Pax2 binding site CD19-2(A- ins.) (Kozmik et al., 1992). All sites are bound by wild-type protein (Fig. 5A), but not by the Noi^tm243a and Noi^tu29a mutant proteins (Fig. 5B,C). It remained a formal possibility that the Noi^tu29a protein fragment is able to interfere with binding of wild-type protein, thus exerting a dominant negative effect. We therefore added, in a competition experiment, up to a 100-fold molar excess of noi^tu29a protein to a wild-type Pax2.1 binding reaction to the high-affinity BSI site. Presence of the mutant protein does not interfere with DNA binding of the wild-type protein, showing that the noi^tu29a protein fragment is unlikely to act in a dominant negative way (Fig. 5D). Our results show that in both mutant proteins the DNA binding activity is completely lost. As shown above, however, some functional Pax2.1 transcripts are present in the noi^tm243a mutants, consistent with a weaker phenotype in the marker gene expression analysis than in noi^tu29a (see below).

In conclusion, our molecular data show that noi^tu29a is a null allele (now referred to as noi^−/−); noi^tm243a and noi^th44a are strong alleles, but weaker than the null, and noi^tb21 and noi^ty31a have weaker molecular defects.

Expression of eng2, eng3, wnt1, fgf8 and her5 occurs in the early MHB primordium with a similar time course to pax2.1, suggesting that pax2.1 might regulate their expression. We used the allelic series to examine in detail the onset of expression of these genes in wild-type and in noi mutant embryos, and find that they are differentially regulated. eng3 expression normally starts at the 1-somite stage, and is never activated in noi^−/− mutants (Fig. 6A,B; Table 1). eng2 is expressed from 90% epiboly onwards in the wild type, and expression is detectable at a strongly reduced level in noi^−/− mutants, in a subpopulation of the wild-type domain that fades away with the appearance of the first somite (Fig. 6C). eng1 expression is seen in few cells of the dorsal MHB of wild-type embryos from the 15-somite stage onwards, and is likewise eliminated in noi^−/− (Fig. 8J,K). In mutants homozygous for weak alleles, eng3 and eng2 are both activated normally (not shown). From the 6-somite stage onwards, however, the expression narrows, persisting only in the dorsal part of the normal expression domain (Fig. 6B,E,F). These findings show that noi functions in establishment of eng2 and eng3 gene expression in the midbrain and MHB primordium.

Failure to express eng genes at the MHB of noi^−/− mutants could theoretically be due to absence of these cells in the mutants. We used expression of pax2.1 RNA and double ISH in noi mutant embryos to determine how long MHB cells persist in the various alleles (Fig. 6G,H,O,P). Onset of pax2.1 expression occurs during late gastrulation (80% epiboly) and is unaffected by all noi alleles (not shown); pax2.1 is therefore not initially required for its own expression. The MHB expression in noi^−/− mutants is lost between the 6- and 9-somite stages (Fig. 6G,H). In strong mutants, expression becomes undetectable by the 12-somite stage, and in weak mutants, by the 20-somite stage (Fig. 6H). Within the mutants, pax2.1 RNA levels decrease uniformly, without a particular bias along the dorsoventral axis (Fig. 6G).

In wild-type embryos, her5 and wnt1 are initially expressed throughout the midbrain and MHB primordium, and maintenance, rather than initiation, of MHB expression is affected in noi^−/− mutants. During somitogenesis stages of the wild type, her5 and wnt1 expression becomes gradually restricted from the midbrain towards the MHB; wnt1 retains a dorsal expression stripe in the midbrain. In noi^−/− mutants (identified by absence of eng3 staining), expression is initiated normally, but maintenance of expression becomes abnormal from the 6-somite stage onwards (Fig. 6K,L,P). Similarly, expression of fgf8 in noi^−/− mutants is initiated normally and disappears from the MHB by the 9-somite stage (Fig. 6M,N,O; Reifers et al., 1998). We conclude that noi function is not required for initiating, but for maintaining expression of her5, wnt1, fgf8 and pax2.1 itself at the MHB.

Interestingly, the midbrain and the MHB primordium of the mutants differ in the kinetics with which gene expression disappears (summarized in Table 1). her5 expression in the developing midbrain is still normal at a time when MHB expression is already partially reduced in noi^−/− mutants at the 5-somite stage (Fig. 6I). Homozygotes for weak alleles show the same phenomenon at slightly later stages (Fig. 6J). Likewise, MHB expression of wnt1 is missing in noi^−/− embryos from about the 6-somite stage onwards, whereas the dorsal stripe persists (Fig. 6K, bracket). For intermediate and weak noi alleles, MHB expression of wnt1 persists longer but is also eliminated eventually (Fig. 6L). We conclude that during the maintenance phase, the sensitivity towards missing noi function appears to be higher at the MHB proper than in the adjacent midbrain.

Within the MHB of the mutants, dorsoventral differences in sensitivity can be observed that differ for the marker gene considered. For instance, her5 expression is first lost from the medial, then the dorsal part of the MHB of noi^−/− mutants, and then the expression is reduced to a ventral spot (Fig. 6I) that is eventually eliminated (Fig. 8); for weak alleles, the ventral spot persists to later stages (Fig. 6J). Expression of fgf8 in noi^−/− mutants also persists longest in ventral MHB tissue (Fig. 6M,N). In contrast, eng2 and eng3 expression persists longest in the dorsal MHB (Fig. 6B,E,F; see above). We conclude that gene expression at the MHB is sensitive to the level of noi function during midsomitogenesis stages, and that the requirement differs, unexpectedly, along the dorsoventral axis.

To further understand the differences in genetic requirement for pax2.1, we compared the distribution of pax2.1 with eng2, eng3 and wnt1, using a double ISH procedure that provides nearly cellular resolution (Fig. 7). At its onset in late gastrulation (90% epiboly), eng2 expression coincides with pax2.1 expression (Fig. 7C,D). eng3 expression is initiated at the 1-somite stage within the pax2.1 domain, and slightly later, is expressed coincident with pax2.1 (Fig. 7E,F and not shown). The coexpression in the same cells is consistent with a role for pax2.1 in regulating these genes. In contrast, the domain of wnt1 expression resembles, but is not identical to the pax2.1 domain: the anterior wnt1 border coincides with the anterior pax2.1 border at this stage, but pax2.1 extends further posteriorly than wnt1, and wnt1 extends further laterally than pax2.1 (Fig. 7A,B; brackets denote non-overlapping regions). A separate study showed that the domain of fgf8 expression during gastrulation is located posterior to the pax2.1 domain, with a significant overlap becoming apparent only during mid- somitogenesis stages (Reifers et al., 1998). These findings complement our observations that expression of wnt1, fgf8 and her5 does not depend on noi function. Together, they show that multiple signaling pathways become activated in parallel during early development of the MHB territory.

Up to 48 hours of development, eng expression is not observed in noi^−/− embryos, presumably due to the absence of midbrain and MHB cells at this stage (Fig. 8). Analysis of the weaker alleles of the series suggests, however, that pax2.1 may also be required for later expression of eng2 and eng3. As described above, eng2 and eng3 are not expressed in strong noi^tm243a and noi^th44a mutants at midsomitogenesis. From the 20-somite stage onwards, however, eng2 is detected in a small patch of ventral neural tube cells in strong but not in null mutants (Fig. 8F-I), and this increases in noi^tb21 and noi^ty31a (Fig. 8H,I). Later expression is also seen for eng3: apart from the dorsal patch of cells that persists until 20 hours of development (Fig. 6B), an additional eng3-positive domain is detected in the ventral neural tube (Fig. 8D,E). Again, expression in this ventral domain is stronger in the two weak alleles than in the strong alleles. Mutants for the strong alleles have not expressed eng3 during earlier neuroepithelial development; hence, late eng3 re-expression must occur independently of the earlier neuroepithelial expression. We do not observe re-expression for eng1, her5 or fgf8 in noi mutants at 24 hours of development (Fig. 8J-M, Fig. 6N). Several differentiated neurons later express eng genes in this area in zebrafish and chicken (Hatta et al., 1991; Millet and Alvarado Mallart, 1995) and the expression we observe may be in precursors for these neurons. Notably, this cell population is highly sensitive to the level of functional Pax2.1, which may therefore be required also for the re- expression phase; alternatively, the cells may simply be able to persist to later stages in weak noi mutants.

We have examined the requirement for no isthmus/pax2.1 in development of the zebrafish midbrain hindbrain boundary. In addition to the previously characterized noi^th44a (Brand et al., 1996), we describe four additional mutations in the pax2.1 gene. By molecular and phenotypic criteria, the noi alleles form an ‘allelic series’ in which pax2.1 function is probably inactivated to different degrees, in the following order: null (noi^tu29a) > strong (noi^tm243a, noi^th44a) > weak (noi^ty31a,noi^tb21), and have presented evidence that noi^tu29a is a null allele (now called noi^−/−). Analysis of the noi^−/− mutants allowed subdividing genes expressed during early MHB development into those that require noi function already at the end of gastrulation (eng2, eng3) and those that are activated independently (fgf8, wnt1, her5 and pax2.1 itself), but require it for maintenance during mid-somitogenesis. Together with other data, this argues that multiple signaling pathways operate in early MHB development. Analysis of the intermediate and weak alleles of the allelic series suggests an additional requirement for pax2.1 during dorsoventral patterning, and possibly in a later phase of engrailed gene expression during MHB development.

The pax2.1 genomic structure shows overall conservation of the exon/intron structure between mammals and zebrafish, but also some differences. Of the 12 exons, 10 show sequence homology to other pax2/5/8 sequences, whereas exons 5.1 and 7.1 encode zebrafish-specific sequences. Such species-specific exons have also been reported for murine, human and Xenopus Pax2 (Sanyanusin et al., 1996; Heller and Brändli, 1997; Tavassoli et al., 1997), suggesting that functional specializations have occured between orthologues in different vertebrates. As shown in the accompanying paper, zebrafish pax2.1 and mouse Pax2 differ for instance in their requirement for controlling transcription of pax5 and pax8 at the MHB, which depend on pax2.1 function in zebrafish, but which may act in parallel in mice (Pfeffer et al., 1998). Multiple functions of zebrafish pax2.1 may be reflected in the high number of splice variants we found, which remain to be tested for functional differences.

For functional studies, and given the variable phenotype of mouse Pax2^−/− mutants, it was crucial to determine to what extent pax2.1 function is impaired by noi mutations. In addition to the previously characterized noi^th44a allele, we found molecular aberrations for the remaining five noi alleles in pax2.1. Together with our previous data on genetic linkage and protein expression (Brand et al., 1996), this provides further evidence that the genetically defined no isthmus gene is identical with pax2.1.

The following observations argue that noi^tu29a is a null allele: (1) the mutation creates a stop codon in the paired domain, in an exon common to all splice variants; (2) the predicted mutant protein lacks the C terminus with the transactivating and inhibiting domains, as well as 6 amino acids of the paired domain, which are crucial for DNA recognition (Adams et al., 1992); (3) the proposed structure of the mutant protein is consistent with the absence of a C-terminal Pax2.1 protein epitope in noi^tu29a mutants (Brand et al., 1996), and the protein size when the mutant protein is expressed in bacteria (data not shown); (4) our DNA-binding assays show that the noi^tu29a protein fragment does not bind to known Pax2 binding sites, including two sites required for Pax2-dependent activity of the mouse En2 promoter (Song et al., 1996); (5) we observe no dominant negative effect in DNA-binding assays, even at a 100-fold molar excess to wild-type protein; (6) expression of eng3 as a likely downstream target is abolished in noi^tu29a homozygotes.

Other noi alleles partially reduce pax2.1 activity. The morphologically strong alleles show distinct differences in their molecular phenotype compared to noi^tu29a. The truncated noi^th44a protein contains an intact paired domain and octapeptide (Fig. 3A), and shows a weaker phenotype in our marker gene expression studies; this protein may therefore be able to activate transcription at a low level (see also Nutt et al., 1998). The postulated Noi^tm243a protein lacks six amino acids in the N-terminal region of the paired domain that mediate protein/DNA contacts (Xu et al., 1995) due to a mutated splice acceptor site. Although DNA binding activity of this protein is completely abolished, mutants for this allele display a slightly weaker phenotype in the expression of the marker genes than noi^tu29a, probably because a small amount of normally spliced mRNA is still present. Similar splice acceptor mutations in the globin genes cause thalassemia (Treisman et al., 1983), and the same substitution in a splice acceptor site of Pax3 activates cryptic splice sites within the following exon (Epstein et al., 1993). In these cases, the mutated splice acceptor sites continue to function with low efficiency, and we propose that this is also the case for noi^tm243a.

The deleted exon 7 in mRNA from noi^tb21 homozygotes may also be due to aberrant splicing. The occurence of deletions in two of five noi alleles by aberrant splicing could be chance, but might also reflect easier detectability of deletion phenotypes in the morphological screens. The noi^tb21 mutation effectively forces creation of the naturally occuring major splice form lacking exon 7. Since noi^tb21 retains partial activity, the natural splice variant may also be functional. Alternatively, a small amount of normally spliced transcripts or a minor splice form unaffected in noi^tb21 could provide the residual function. Reconstitution of the various splice forms into a noi^−/− background can now be used to address their function.

The MHB phenotype of noi^−/− mutants is more constantly severe than that of murine Pax2 mutants, and we previously hypothesized that either only a single pax2/5/8 gene exists in zebrafish, or that pax2.1 has become functionally predominant (Brand et al., 1996). The isolation of three additional zebrafish pax2/5/8 genes reported in the accompanying paper (Pfeffer et al., 1998) supports the latter possibility, since pax5 and pax8 are critically dependent on noi/pax2.1 function at the MHB.

Between its onset at 80% of epiboly and the 4-somite stage, pax2.1 is the only known pax2/5/8 gene expressed at the MHB. pax2.2, though expressed in noi mutants from the 5-somite stage, appears unable to replace noi^−/− function, perhaps because its MHB activation occurs too late (see also the discussion in Pfeffer et al., 1998). In other tissues, e.g. in the optic stalk (Macdonald et al., 1997), pax2.2 may partially compensate the missing noi function.

A key finding in our study is that, through the availability of the null allele noi^tu29a, we were able to subdivide genes expressed in the early midbrain-hindbrain primordium according to their requirement for pax2.1. pax2.1 is activated before neural plate formation, prior to and in an overlapping expression pattern with wnt1, eng1, eng2 and eng3, whereas her5 and fgf8 are activated slightly earlier. Expression of eng2 and eng3 already clearly requires pax2.1 function during late gastrulation. In contrast, the normal onset and expression of wnt1, fgf8, her5 and pax2.1 itself up to approximately the 4- to 5-somite stages suggests that additional, pax2.1-independent pathway(s) operate upstream and/or in parallel to induce gene expression in the MHB primordium.

Alternatively, wnt1 or fgf8 could simply be upstream of pax2.1, and would thus not be affected in noi^−/− mutants. For wnt1 this is unlikely, since En1, En2 and Pax2 are activated normally in mouse wnt1^−/− mutants (McMahon et al., 1992; Rowitch and McMahon, 1995). In zebrafish, a wnt1 mutant is not yet known, but since wnt1 and pax2.1 are activated in different, only partially overlapping domains (Fig. 7), and wnt1 misexpression in zebrafish does not activate pax2.1 (Kelly and Moon, 1995), it is very likely that the two genes are activated independently of each other in zebrafish as well. Expression of fgf8 in noi^−/− mutants is unaffected for a different reason: during gastrulation, fgf8 is expressed in the anterior hindbrain, posterior to the domains of pax2.1 and wnt1; moreover, fgf8 misexpression does not alter the anterior-posterior extent of pax2.1 expression (Reifers et al., 1998). Importantly, in the zebrafish fgf8 mutant acerebellar, which lacks MHB and cerebellum but has a midbrain, expression of MHB marker genes is initiated normally but not maintained (Reifers et al., 1998). However, fgf8 inactivation in mice also causes midbrain defects (Meyers et al., 1998), perhaps as a secondary consequence of gastrulation defects which are less severe in acerebellar mutants (see discussion in Reifers et al., 1998).

Together, our findings in zebrafish suggest that pax2.1, wnt1 and fgf8 are initiated independently of each other in late gastrulation, and only later in somitogenesis come to interact (Reifers et al., 1998; this paper; Fig. 9). Evidence for multiple pathways in MHB formation also comes from otx gene dosage studies (Acampora et al., 1997). In otx2^−/−, otx1^+/− animals, ectopic anterior expression of fgf8 and later formation of an ectopic MHB are observed. At the early neural plate stage, however, expression of En2 and wnt1 are initiated at their normal location, and only in midsomitogenesis is the expression recruited to the ectopic, anterior position. As in zebrafish, these studies therefore distinguish at least two pathways, one positioning fgf8 (controlled via otx dosage), and a second pathway activating En and wnt1 expression, independent of otx.

In noi^−/− embryos, eng3 expression is completely abolished and eng2 is expressed transiently at a low level; similarly, the later activation of eng1 is not seen in noi^−/− mutants. noi/pax2.1 is therefore a crucial upstream component in the pathway that activates eng2 and eng3 in the midbrain and MHB primordia during late gastrulation, consistent with the colocalization of the expression domains (Fig. 7C-D). The noi-independent component of eng2 expression demands, however, an additional mechanism for activating eng2. Generally, expression of eng genes is under the control of both pax2(.1) (this paper; Song et al., 1996) and wnt1 (McMahon et al., 1992; Danielian and McMahon, 1996), and can be activated by fgf8 misexpression (Crossley et al., 1996; Lee et al., 1997; Shimamura and Rubenstein, 1997). Importantly, although wnt1 and fgf8 are activated normally in noi^−/− embryos during late gastrulation, they are unable to drive eng3 or normal eng2 expression in the mutants. Thus, although wnt1 and fgf8 are necessary to maintain eng expression (McMahon et al., 1992; Reifers et al., 1998), they are not sufficient, with the possible exception of the weak transient eng2 expression.

In mice, Pax2 directly regulates En2 promotor activity via Pax2/5/8 binding sites (Song et al., 1996). As in mice, Pax2.1 may directly activate eng genes by binding to the eng promoters. These remain to be characterized in zebrafish to explain the differences in regulation of the engrailed genes we have observed, and to distinguish direct from indirect regulation. Given the ongoing morphogenetic movements and the successive restriction of pax2/5/8 and eng2/3 genes towards the MHB constriction, the regulatory relationship between pax2.1 and engrailed genes is likely to be important for later midbrain development, e.g. for retinotectal projection into the midbrain tectum, which is thought to require an Engrailed protein gradient (Rétaux and Harris, 1996).

Organizing centers often establish concentration gradients of signaling molecules that pattern the adjacent tissue, and are thus sensitive to the functional level of components of the signaling pathway. Our comparison of the molecular and phenotypic strength within the noi allelic series suggests that the functional level of pax2.1 is critical for MHB development. By analogy with similar studies in Drosophila (Anderson et al., 1985), we assume that ‘functional level’ means here ‘level of Pax2.1 protein activity’, though this remains to be proved. The failure to maintain expression of her5, fgf8, wnt1 and other markers in noi^−/− mutants may mean that, for continued expression, the MHB organizer needs to be successfully established, producing, for instance, wnt1 and fgf8 as signals important for maintenance. When this occurs in development is currently unknown; our data point to a critical period around the 5-somite stage. Interestingly, the requirement for noi function at this stage is higher at the MHB than in the midbrain primordium, since this is where gene expression is affected first in the mutants. Differential effects on the midbrain and MHB expression of wnt1 are also observed in Gbx2, Otx2 and fgf8 mutants (Acampora et al., 1997; Wassarman et al., 1997; Reifers et al., 1998; Takada et al., 1994). The maintenance requirement we observe in noi mutants is not necessarily direct: it could reflect an inability to form a functional MHB organizer in noi^−/− mutants, or inability to respond to signals from the organizer, or both; further studies are needed to distinguish these possibilities.

Taken together, our data support a model where establishment of the zebrafish MHB occurs in several steps (Fig. 9). During the establishment phase, at least two pathways are activated in the MHB primordium in late gastrulation: one pathway involves Pax2.1 and leads to proper activation of eng genes (with its downstream consequences for development of the retinotectal map). Independently, a second pathway is activated that employs the secreted Wnt1 molecule. Very likely, a third independent pathway employing Fgf8 is activated in the posteriorly abutting anterior hindbrain. What signal(s) in turn activate these pathways is not yet known; they may derive at least in part from the endomesoderm (Ang and Rossant, 1993; Woo and Fraser, 1997; Ang, 1996; Miyagawa et al., 1996).

During the second or maintenance phase in early to mid- somitogenesis, gene expression at the MHB is maintained by reciprocal interactions, as is also suggest by the functional requirement for wnt1 and En1 in mice (McMahon et al., 1992; Wurst et al., 1994). During this period, significant overlap between pax2.1, wnt1 and fgf8 expression is established in the future MHB, which may be crucial in generating its organizing properties. During this period, noi/pax2.1 continues to be required for expression of eng and other genes, and directly or indirectly regulates downstream gene expression in the MHB and around it. Also during the maintenance period, patterning occurs along the dorsoventral axis in the MHB organizer in a manner that still requires pax2.1 activity.

So far, pax2.1 has mainly been considered as a component of the patterning machinery along the anteroposterior (a/p) axis of the brain. Analysis of the allelic series of noi alleles unexpectedly revealed different sensitivity of MHB cells to pax2.1 functional levels along the dorsoventral (d/v) axis, which differs for the gene considered. Expression of pax2.1 itself and wnt1 disappears homogeneously, probably simply reflecting loss of the cells normally destined to form MHB tissue. With increasing strength of the noi allele examined, the MHB loses her5 and fgf8 gene expression from dorsal to ventral, such that the highest level of pax2.1 is required in the dorsal area. In contrast, eng gene expression disappears from ventral to dorsal with increasing allele strength, as has also been observed in for En2 in En1 mutants (Wurst et al., 1994). Thus, these genes respond to d/v positional information, but in a way that is critically dependent on the functional level of pax2.1, and in different ways for the various marker genes. We have no indication that, in wild-type development, pax2.1 itself is distributed in an asymmetric way along the d/v axis of the MHB (although this view does not take the complexity of alternative splice forms into account), and the requirement could be indirect.

Why then are differences observed, depending on the allele and the marker gene studied? Probably, MHB cells continue to require pax2.1 during the maintenance phase. In this view, pax2.1 would serve as an integrator for both a/p and d/v determining signals at sequential stages of development. Sequential assignment of a/p and d/v positional values is also seen in rhombomere 4 of the chick hindbrain (Simon et al., 1995). Likely candidates for d/v signals are signaling molecules like sonic hedgehog (shh) from the ventral side, and BMPs from the dorsal side (Tanabe and Jessel, 1996). Indeed, pax2.1 and shh are functionally linked in patterning of the optic vesicle and optic chiasm in zebrafish (Macdonald et al., 1995, 1997). More recently, a medaka homologue of Drosophila spalt, a target gene for Hedgehog signaling in Drosophila, was found to be expressed in MHB development and to respond to shh RNA injection specifically at the MHB (Köster et al., 1997).

The nervous systems of invertebrates and vertebrates share a common origin, reflected in many conserved interactions within the patterning processes (DeRobertis and Sasai, 1996). In Drosophila, Engrailed, Wingless and Hedgehog are involved in a feedback loop that functions in boundary formation in the embryo and the imaginal discs; a paired-type gene is involved in establishing this loop (Martinez-Arias, 1993). Of their vertebrate homologues, eng genes, pax2(.1) and wnt1 are clearly required for a/p patterning of the MHB, whereas shh is not, and it has been problematic to understand the ‘missing’ shh involvement. Our observations suggest a possible explanation: the circuitry of genes including shh may originally have been maintained for their function within the MHB organizer proper, and may only later in evolution have become adapted for the a/p patterning function of the organizer. Vertebrates have an elaborated midbrain that presumably requires presence of the a/p patterning function, whereas more basal chordates do not (Butler and Hodos, 1996). Interestingly, the protochordate ascidian Halocynthia roretzi has a single archetypal pax2/5/8 gene, which is expressed in the ‘neck’ region, and has therefore been suggested to be related to the midbrain and MHB of vertebrates (Wada et al., 1998). Ascidian pax2/5/8 is expressed posteriorly adjacent to ascidian otx, whereas otx2 and Pax2 expression overlap in the midbrain of vertebrates (Acampora et al., 1997), including zebrafish (unpublished observations). A speculative possibility is therefore that the ascidian pax2/5/8 expression domain is related only to the MHB, but not the midbrain portion of pax2 expression, and that the organizing potential of the MHB for the surrounding midbrain and cerebellar primordia was secondarily acquired in vertebrates.

We thank Monte Westerfield, José Campos-Ortega and Alexander Picker for cDNA and lambda clones, and Alexander Picker, Frank Reifers, Anna Sharman, Sophie Légér and Meinrad Busslinger for comments and vivid discussions. This work was funded by the Deutsche Forschungsgemeinschaft (Br 1746/1-1) and the Förderprogramm Neurobiologie, Baden-Württemberg.