spiel ohne grenzen/pou2 is required during establishment of the zebrafish midbrain-hindbrain boundary organizer

Regionalization of the vertebrate nervous system involves the establishment of local organizing centers that serve as signaling sources for patterning and neuronal differentiation. The MHB, or isthmic organizer, is located at the junction of midbrain and hindbrain. During development, the MHB organizer secretes factors that regulate anteroposterior patterning and specification of the neighboring tectum and cerebellum. The organizing activity of the MHB was initially identified by its ability to induce mes- and metencephalic development when transplanted into the caudal diencephalon (Martinez et al., 1991; Marin and Puelles, 1994). FGF8 was later shown to be capable of mimicking the activities of isthmic transplants in the diencephalon, revealing that FGF8 can induce isthmic organizer formation and mediate some of its patterning activities (Crossley et al., 1996).

Formation of the MHB organizer involves discrete steps, including (1) anteroposterior patterning providing positional information to place the MHB within the neural plate, (2) establishment of the MHB during late gastrulation, and (3) maintenance of its regional identity and signaling activities during further development (Joyner at el., 2000; Rhinn and Brand, 2001). Recent studies have begun to elucidate how these steps are accomplished. Transplantation experiments in chick have shown that juxtaposition of midbrain and hindbrain tissues is sufficient to generate a MHB (Irving and Mason, 1999). At a molecular level, this interface is reflected by the expression boundaries of Otx1/2 and Gbx2, which are the first genes known to be expressed in a restricted manner in the mid-/hindbrain. Analyses of mutant mouse embryos showed that the Otx2/Gbx2 border specifies the position of the MHB organizer (Wassarman et al., 1997; Millet et al., 1999; Broccoli et al., 1999; Joyner et al., 2000; Rhinn and Brand, 2001). How positional information provided by the Otx2/Gbx2 expression border leads to MHB establishment remains unclear. As the anterior notochord of both Xenopus and chick (Hemmati-Brivanlou et al., 1990; Darnell and Schoenwolf, 1997) and the anterior mesendoderm in mouse (Ang and Rossant, 1993) can induce expression of En2, vertical signals from axial mesoderm underlying the mes-metencephalic primordium have been suggested to contribute to MHB-specific gene expression. In zebrafish, however, mutant embryos that completely lack axial mesoderm still form a MHB, arguing that planar signals within the neurectoderm may be sufficient for its induction (Rhinn and Brand, 2001).

From late gastrulation onwards, Fgf8, Wnt1, Pax2/5 and the Engrailed genes are expressed in the MHB region (Wassef and Joyner, 1997; Joyner et al., 2000). Wnt1 is expressed in the midbrain just anterior to the Fgf8 domain in the hindbrain. Pax2/5 and the Engrailed genes are expressed in the mid-/hindbrain region surrounding the isthmus. Gain-of-function studies have shown that misexpression of En1/2 or Pax2/5 in the posterior forebrain of chick or fish results in ectopic expression of MHB genes, including Fgf8, and induction of ectopic MHB development (Araki and Nakamura, 1999; Funahashi et al., 1999; Okafuji et al., 1999; Ristoratore et al., 1999). Loss-of-function studies in mouse and zebrafish have demonstrated that Fgf8, Wnt1, Pax2 and En1 are required for normal mid-/hindbrain development (McMahon and Bradley, 1990; Urbanek et al., 1994; Wurst et al., 1994; Brand et al., 1996; Schwarz et al., 1997; Lun and Brand, 1998; Meyers et al., 1998; Reifers et al., 1998).

Work in mice and zebrafish indicates that at least three parallel pathways are activated during MHB establishment, involving wnt1, pax2.1 and fgf8 (McMahon et al., 1992; Lun and Brand, 1998). wnt1 and pax2.1 appear to be activated independently in partially overlapping domains, and wnt1 expression in noi mutants is unaltered until early somitogenesis (Lun and Brand, 1998). In addition, Pax2 is activated normally in Wnt1 mouse mutants (McMahon et al., 1992; Rowitch and McMahon, 1995). In zebrafish, the initial expression of fgf8 in the anterior hindbrain does not overlap with that of pax2.1 or wnt1 (Reifers et al., 1998), while in mice the initial Fgf8 expression at the MHB, but not that of Wnt1 and En2, depends on Otx1 and Otx2 gene dosage (Acampora et al., 1997). During somitogenesis, Fgf8, Pax2, Wnt1, and En1/2 engage in a regulatory loop that maintains the MHB and its signaling activities during further development (Lun and Brand, 1998; Reifers et al., 1998; Broccoli et al., 1999; Joyner et al., 2000; Rhinn and Brand, 2001).

In this report, we investigate the role of spiel ohne grenzen (spg) (Schier et al., 1996) during zebrafish MHB organizer development. spg mutant embryos develop patterning defects, including a deletion of the isthmic region and cerebellum, and a reduction of the tectum. We have identified pou2, which encodes a POU-domain transcription factor (Takeda et al., 1994; Hauptmann and Gerster, 1995), as the gene affected by spg mutations. pou2 is transiently expressed during late gastrulation and early somitogenesis in the midbrain and anterior hindbrain primordia. We have investigated the effects of a new amorphic spg mutation, spg^m793, on MHB formation. We find that expression of otx2, gbx1 (functional homolog of mouse gbx2 in zebrafish), and fgf8 is initiated normally in spg mutants, while MHB expression of wnt1 and pax2.1 is reduced. We propose that pou2 is required to achieve the pax2.1 and wnt1 expression necessary for MHB organizer establishment during late gastrulation.

The spg alleles spg^m216 and spg^m308 have previously been briefly described (Schier et al., 1996). The new spg^m793 allele was isolated following mutagenesis of AB strain fish with ethyl-nitrosourea (Artinger et al., 1999). Other zebrafish mutations used were acerebellar (ace^ti282a) and no isthmus (noi^tu29a), a null allele of noi (Brand et al., 1996).

To determine genetic linkage, an spg^m793/+ male (G0) was crossed with an India (IN) strain female. Gynogenetic diploid F2 progeny were obtained from spg^m793/+ F1 fish by the ‘early pressure’ technique (Westerfield, 1994). A panel of 40 spg^m793 mutant F₂ embryos was typed with centromere linked SSLP markers (Knapik et al., 1998). spg was localized to linkage group 21. A map panel of 2418 spg^m793/spg^m793 F₂ embryos was generated for high resolution genetic mapping, and the spg locus was placed between the SSLP marker Z12068 and Z21718 in the proximity of an AFLP marker (Vos et al., 1995) named ‘CCA/C’. The ‘CCA/C’ marker was used to isolate PACs from an arrayed PAC library. Three PAC clones were identified that span the entire spg genomic region: PAC A, BUSMP706B14149; PAC B, BUSMP706M22111; and PAC C, BUSMP706E23115. pou2 sequences were identified in both PAC A and PAC B. DNA encoding pou2 was isolated from genomic DNA (AB strain, India strain, spg^m793, spg^m308 and spg^m216) and from cDNA generated from 30% epiboly stage embryos. Sequencing of the PCR products revealed separate point mutations that generate restriction polymorphisms in the pou2 transcribed region of each allele: spg^m793, MseI site absent; spg^m308, BfaI site absent; and spg^m216, XbaI site absent. These polymorphisms were scored on 4% metaphor agarose gels (BioWhittaker, Rockland, MD) to verify the presence of the mutations predicted by sequence analysis (Fig. 2D) and for genotyping experimental embryos.

The following pou2 antisense morpholino (GeneTools, Corvallis USA) was used: 5′-CGCTCTCTCCGT CAT CTTTCCGCTA-3′ (start ATG underlined). A four mismatch control oligo 5′-CGGTCTGTCCGTCATCTATCCCCTA-3′ was used to assess specificity of the knock-down phenotype.

Whole-mount in situ hybridization was performed to visualize gene expression (Hauptmann, 1999). Control embryos denoted wild type in the figures are phenotypically wild-type siblings of mutant embryos shown in the same experiment. The genotype of noi mutant embryos was determined by assaying for reduced expression of eng2 (Brand et al., 1996). cDNAs used have been described (see references in text) except for zebrafish gbx1, which is the functional homolog of mouse Gbx2 (Rhinn and Brand, 2001). While zebrafish gbx1 has greater sequence similarity to mouse Gbx1 than it does to mouse Gbx2 (Bulfone et al., 1993; Rhinn and Brand, 2001) (B. Thisse and C. Thisse, unpublished), the expression of zebrafish gbx1 more closely resembles that of mouse gbx2 during late gastrulation.

The pou2 cDNA (Hauptmann and Gerster, 1995) was subcloned into the pCS2+ vector and transcribed using the Sp6 MessageMachine (Ambion). In vitro synthesized mRNA was dissolved at 2 to 40 ng/μl in water and microinjected into one-cell stage embryos. The actual amounts (2-40 pg) injected into embryos were estimated from the injection volume visualized.

spg mutant embryos are easily identified by the absence of the isthmic constriction which normally demarcates the position of the MHB (Fig. 1) (Schier et al., 1996; Driever et al., 1997). During a recent mutagenesis screen, we identified an allele of spiel ohne grenzen (spg^m793; Fig. 1K) with more severe defects than the previously reported alleles spg^m216 (Fig. 1F) and spg^m308 (not shown; phenotype similar to spg^m216). spg^m793 mutants develop morphological defects in three major parts of the body.

(1) Mid-/hindbrain: The tectum is reduced in anteroposterior extent and develops an atypical globular morphology. The anteroposterior extent of the ventral midbrain is similarly reduced. The isthmic constriction is absent, and a visually discernable cerebellum does not form. Correspondingly, the distance between eye and ear is reduced by about one-third.

(2) Rhombencephalon and ear: Morphological aspects of segmentation are disturbed and the hindbrain is shorter. The otic vesicle is round instead of oval-shaped and often has only a single otolith [hindbrain segmentation abnormalities seen in spg mutants are analyzed in a separate manuscript (Hauptmann et al., unpublished)].

(3) Trunk and tail: Defects are observed during tail development, somite morphology is abnormal, and the number of primary neurons in the spinal cord is reduced (data not shown). We consider these multiple phenotypes to be pleiotropic activities of spg, mediated by independent expression domains of spg/pou2 (see below).

Analysis of gene expression in the MHB region of spg^m793 mutants supports the findings from morphological observation of living embryos. At 24 hpf, otx2 is expressed in parts of the diencephalon and the midbrain including the tectal primordium (Fig. 1B) (Li et al., 1994). In spg^m216 mutants (Fig. 1G), we observed a slight reduction of the midbrain; while in spg^m793 mutants (Fig. 1L), otx2 expression indicates that the caudal tectum lies above the rostral hindbrain. fgf8 is expressed in a thin stripe in the MHB at 24 hpf (Fig. 1C) (Fürthauer et al., 1997; Reifers et al., 1998). In spg^m216 mutants (Fig. 1H), fgf8 expression is reduced to a small patch within the dorsal MHB. Isthmic fgf8 expression is completely absent in spg^m793 mutants (Fig. 1M). The POU-domain transcription factor zp50 is expressed at high levels in the cerebellum (Fig. 1D) (Hauptmann and Gerster, 1996). In spg^m216 mutants, the zp50 expression domain in the cerebellum is still present, albeit reduced in size (Fig. 1I). No cerebellar expression of zp50 is found in spg^m793 mutant embryos (Fig. 1N). Thus, in the absence of a cerebellum, the tectum reaches the hindbrain ventricle in spg^m793 mutants. In agreement with our morphological observations, the expression of otx2, fgf8, and zp50 indicate that spg^m793 is a stronger allele than spg^m216.

In the absence of pronounced morphological landmarks in the ventral MHB region, the gap between the discrete expression domains of the transcription factor zash1a in the posterior tegmentum (ventral midbrain) and the basal rhombomere 1 (r1) is a useful marker of ventral isthmic territories (Fig. 1E) (Allende and Weinberg, 1994). In both spg^m216 and spg^m793 mutants, these zash1a expression domains close in on each other, indicating the absence of the ventral portion of the MHB region (Fig. 1J,O).

To identify the gene affected in spg, we first mapped spg to LG21 and identified molecular markers tightly linked to the locus (Fig. 2B). Fine mapping of the spg genomic region indicated that the pou2 gene (Fig. 2A), previously mapped to LG 21 (Postlethwait et al., 1998), is contained within the spg interval (Fig. 2B). We sequenced genomic DNA from three spg alleles, spg^m793, spg^m308 and spg^m216, and found a single point mutation within the transcribed region of pou2 in each allele (Fig. 2C,D). In spg^m793, the splice acceptor site of exon 2 contains a transition from A to G. Aberrant splicing at this site leads to a shift in the reading frame resulting in a stop codon at the sixth codon downstream of the splice acceptor site. This generates an open reading frame that lacks both the POU-specific domain and the POU-homeodomain. The aberrant splice product was verified by sequencing cDNA from spg^m793 mutant embryos. The mutation in spg^m308 is an A-to-T transversion in the splice acceptor site of exon 5. In these mutants, splicing of pou2 occurs at an alternative splice site located further downstream. An identical naturally occurring splice variant of pou2 has been previously described (Takeda et al., 1994); it disrupts the homeodomain and renders the pou2 protein unable to bind the canonical octamer POU-binding motif (Takeda et al., 1994). The homeobox is also affected in spg^m216, where a transition from T to A replaces a leucine with a proline at position 16 in helix 1 of the homeodomain. This leucine is highly conserved among homeodomain proteins (Scott et al., 1989) and is thought to interact with hydrophobic amino acid residues of helix 3 that contact the DNA target sequence (Qian et al., 1989). As proline residues break the secondary structure of α-helices, this mutation is likely to disrupt the three-dimensional structure of the pou2 homeodomain. Morphologically, spg^m216 and spg^m308 mutants closely resemble each other and display a milder phenotype compared with spg^m793 mutants (Fig. 1; also data not shown). The residual activity of these alleles is likely mediated through the POU-specific domain still present in these proteins.

pou2 is a maternally expressed gene and through mid-gastrulation pou2 mRNA is widely distributed in the blastoderm (Fig. 3A) (Takeda et al., 1994; Hauptmann and Gerster, 1995). The earliest regionalized zygotic expression of pou2 is seen at 80% epiboly adjacent to the midline and in an area corresponding to the midbrain and hindbrain primordia (Fig. 3B). By tailbud stage (Fig. 3C), pou2 expression has condensed into a butterfly shape, with a lower level of expression in the prospective midbrain and a higher level in the anterior hindbrain. As pax2.1 expression in the midbrain (Krauss et al., 1991) co-localizes with the lower level of pou2 expression (Fig. 3D), the border between low and high levels of pou2 expression is located at or near the MHB.

pou2 expression in spg^m216 mutant embryos resembles that of wild-type embryos during gastrulation (data not shown). By the one-somite stage, however, the pattern of pou2 expression is slightly altered in that the chevron shaped expression domains in prospective rhombomere 2 and 4 seen in the wild type (Fig. 3D) appear more diffuse (Fig. 3I). By contrast, pou2 mRNA levels are strongly reduced in spg^m793 mutants from as early as 50% epiboly onwards (compare Fig. 3A with 3F). pou2 expression is further diminished by the end of gastrulation in spg^m793 mutants (Fig. 3C,H) and could not be detected during somitogenesis stages (data not shown). The absence of pou2 RNA in spg^m793 mutant embryos may be due to instability of the aberrant splice product, and the pou2 message detected early may predominantly correspond to maternal transcripts. Both the truncated Pou2 protein predicted in spg^m793 mutants, and the lack of pou2 expression at the end of gastrulation, suggest that spg^m793 acts as a null allele with complete absence of zygotic pou2 activity.

To determine whether the maternal contribution may modify the spg MHB phenotype, we injected a pou2-morpholino antisense oligonucleotide (Nasevicius and Ekker, 2000) to repress translation of maternal message. Injection of about 100 pg per embryo resulted in reduction or absence of pax2.1 expression at three- to six-somite stages, while emx1 (Patarnello et al., 1997) expression in the forebrain appeared normal (data not shown). This is similar to the phenotype observed in zygotic spg^m793 mutants. At high concentrations (approx. 1 ng/embryo), we observed arrest of gastrulation (data not shown). Our data point at separate functions of pou2 during early gastrulation and MHB development. In zygotic spg mutants, maternal contribution of wild-type pou2 message appears to rescue gastrulation defects, while MHB development depends on zygotic gene function.

In wild-type embryos, wnt1 is expressed from 90-100% epiboly in the midbrain and MHB primordium (Fig. 4A,B) (Molven et al., 1991). In spgm793 mutants, wnt1 expression is activated at 90-100% epiboly, but only at a reduced level (Fig. 4E,F). The anteroposterior extent of the bilateral wnt1 expression domains is reduced, with the greatest reductions in medial regions. wnt1 expression in the MHB primordium gradually ceases during somitogenesis, while expression along the dorsal midbrain and neural tube is maintained (Fig. 4G,H).

In wild-type embryos, pax2.1 expression is activated in the MHB primordium at about 80-90% epiboly in two lateral domains (shown at 90-100% epiboly, Fig. 4I) (Krauss et al., 1991). By the end of gastrulation, the bilateral pax2.1 domains have merged at the midline (Fig. 4J). In spgm793 mutants, pax2.1 expression appears nearly absent from the medial neural plate (Fig. 4M,N). The anteroposterior extent of the remaining lateral patches of pax2.1 expression is shorter than normal, and the expression level is strongly reduced. Similar to that of wnt1, pax2.1 expression in the MHB primordium of spgm793 mutants ceases during mid-somitogenesis (Fig. 4O,P).

Changes in the expression of the engrailed genes in spgm793 mutants are similar to those seen in pax2.1 expression. In wild-type embryos, eng2 expression is initiated in bilateral domains at 90-100% epiboly (Fig. 4Q) that merge during early somitogenesis (Fig. 4R,S). At the five-somite stage, eng2 is expressed in the MHB primordium and reaches into the posterior portion of the midbrain (Fig. 4S). In spgm793 mutants, initial eng2 expression is detectable only at reduced levels and in smaller domains corresponding to the reduced pax2.1 domains (Fig. 4U,V). At the five-somite stage, only a small dorsal expression domain of eng2 remains in the midbrain (Fig. 4W). At 1 dpf, eng2 expression can no longer be detected in the midbrain or the MHB region of spgm793 mutants (Fig. 4X). Expression of eng1 and eng3 is similarly affected in spgm793 mutants (data not shown). The reduced size of the engrailed expression domains in spgm793 mutants may account for the smaller size of the tectum observed at 1 dpf (Fig. 1K). Our analysis indicates that spg/pou2 function is required during MHB establishment in late gastrulation. As the initial expression of pax2.1 and wnt1 is reduced in the MHB primordium in spgm793 mutants, spg/pou2 may be involved in the activation of pax2.1 and wnt1 expression. To further analyze the epistatic relationship between pou2 and pax2.1, we examined the expression of pou2 in noi/pax2.1 mutant embryos (Fig. 3E,J), and found that it was normal. These findings suggest that pou2 is required upstream of pax2.1 for MHB establishment.

To test whether wild-type pou2 mRNA can rescue the gene expression defects at the MHB in spg mutants, we injected pou2 mRNA into wild-type and spgm793 mutant embryos at the one-cell stage and assayed the expression of pax2.1 during early somitogenesis (Fig. 5; Table 1). While injection of 2 to 5 pg of pou2 mRNA did not result in significant rescue of gene expression, some pax2.1 expression was seen in the MHB region of spgm793 mutants after injection of 10 pg of pou2 mRNA. Injection of 20 or 40 pg of pou2 mRNA was able to rescue pax2.1 MHB expression, and thus MHB establishment, in most mutant embryos (Fig. 5C; Table 1). Occasionally, we observed malformed embryos (eight out of 117 embryos), that showed various degrees of disorganization, broadening of the neural plate and blisters in the trunk region. These phenotypes are consistent with the interference of high doses of pou2 with gastrulation processes in the early embryo (Takeda et al., 1994). We also assayed injected embryos for the expression of pax2.1 and wnt1 at 24 to 26 hpf (Fig. 5F,I). In five out of seven injected spgm793 mutants, expression of pax2.1 and wnt1 was very similar to or indistinguishable from that of wild-type embryos. Thus, overexpression of pou2 mRNA injected at the one-cell stage can lead to stable expression of MHB-specific genes.

In wild-type zebrafish embryos, pax2.2, pax5, and pax8 are expressed in the MHB onwards from their activation between the 5- and 9-somite stages (Pfeffer et al., 1998). In noi/pax2.1 mutants, pax5 and pax8 are not activated in the MHB region, while pax2.2 is activated, albeit in a smaller domain. In spgm793 mutants, pax5 and pax8 expression is absent from the MHB primordium, although these genes are expressed normally in other regions (e.g. optic stalks, ear; Fig. 6F,H,J,L). This suggests a requirement of spg/pou2 for MHB expression of these genes, which is probably mediated through pax2.1. In contrast to noi/pax2.1 mutants, pax2.2 expression is absent from the MHB region of spgm793 mutants (Fig. 6B,D), suggesting that spg/pou2 is required for the normal activation of both pax2.1 and pax2.2 genes in zebrafish.

As mouse Otx2 and Gbx2 position the MHB, we also examined expression of their homologs in spg mutants. In wild-type mid-gastrula embryos, otx2 is expressed in the forebrain and midbrain primordia (Fig. 7A,C) (Li et al., 1994). Zebrafish gbx1 (the functional equivalent of mouse Gbx2) (Rhinn and Brand, 2001) is expressed directly posterior to the otx2 domain in the anterior hindbrain primordium with the shared otx2/gbx1 expression boundary marking the position of the future MHB (Fig. 7E,G). In mice and chick, FGF8 has been suggested to maintain a metencephalic identity by activating gbx2 and repressing otx2 (Liu et al., 1999; Martinez et al., 1999). Similar to gbx1, fgf8 is also activated in the prospective anterior hindbrain at 70% epiboly (Fig. 7I,K) (Lun and Brand, 1998). During gastrulation in spgm793 mutants, expression of otx2 (Fig. 7B), gbx1 (Fig. 7F) and fgf8 (Fig. 7J) appears normal. Thus, pou2 is not required for the initial expression of otx2, gbx1 and fgf8. Subsequently, however, at the beginning of somitogenesis, fgf8 expression at the MHB fades (Fig. 7L), while expression of otx2 and gbx1 still appears normal (Fig. 7D,H). As early fgf8 expression is independent of spg/pou2, we assayed pou2 expression in ace/fgf8 mutants to see if fgf8 acts upstream of pou2. Comparison of pou2 expression in ace/fgf8 mutants and their wild-type siblings at the tailbud stage (50 embryos examined) and the two-somite stage (26 embryos examined) did not yield visually detectable differences (data not shown). This indicates that pou2 expression does not depend on fgf8.

Experiments in various organisms (Hemmati-Brivanlou et al., 1990; Martinez et al., 1991) as well as the analyses of mutant phenotypes in mice and zebrafish (Rhinn and Brand, 2001) have led to the identification of genes that contribute to the positioning of the MHB in the neural plate during gastrulation. Furthermore, they established the importance of Pax2, Engrailed genes, Fgf8 and Wnt1 in maintenance of the MHB organizer from somitogenesis onwards. By contrast, the mechanisms that activate Pax2, FGF8, and Wnt1 expression during the establishment of the MHB organizer remain unclear. So far, four mutant loci have been identified in zebrafish to affect MHB formation: the phenotypes observed in ace/fgf8 and noi/pax2.1 are consistent with the results obtained from experimental studies in chick and mutations in mice (Brand et al., 1996; Reifers et al., 1998). The molecular basis of the aussicht mutation is so far not understood, but it appears to be involved in the regulation of fgf8 expression (Heisenberg et al., 1999). We demonstrate that spiel ohne grenzen (spg) mutations affect the pou2 gene, and show that pou2 is required for MHB establishment.

MHB development proceeds through three steps: positioning, establishment and maintenance. We discuss roles that spg/pou2 might play during each of these steps.

Anteroposterior positioning of the MHB organizer in the neural plate does not depend on pou2. The expression of otx2 and gbx1 during MHB positioning and establishment appears to be normal in spgm793 mutants, demonstrating that the activation of otx2 and gbx1 does not depend on zygotic pou2 function. In addition, the reduced and transient expression domains of pax2.1 and wnt1 in the MHB primordia of spgm793 mutants appear to be located at the correct anteroposterior position. In mice and chick, FGF8 has been suggested to activate Gbx2 and repress Otx2 (Liu et al., 1999; Martinez et al., 1999). During late gastrulation, we observe that expression of fgf8 in the hindbrain primordium is not affected in spg mutants. Thus, the formation of the initial fgf8 expression domain in the hindbrain primordium is independent of pou2. We found normal pou2 expression in ace/fgf8 mutants, indicating that pou2 is independent of fgf8, and that both genes may function in parallel pathways during late gastrula stages.

In late gastrulation, MHB establishment is characterized by the onset of expression of pax2.1, wnt1, fgf8 and the Engrailed genes in the mid-/hindbrain. Three parallel pathways have been proposed to control MHB establishment, activating pax2.1, wnt1 and fgf8. Mutations in pou2 affect all three pathways (Fig. 8). In spgm793 mutant embryos, the initial expression of pax2.1 and wnt1 in the MHB during late gastrulation is reduced, indicating that pou2 is required for the normal activation of these genes and their continued expression during the establishment phase. Several additional findings suggest that pou2 acts earlier and upstream of pax2.1. First, regionalized expression of pou2 in the MHB primordium begins at 80% epiboly, while we observe that of pax2.1 from 90% epiboly onwards. Second, in strong noi/pax2.1 mutants, expression of genes that act in MHB establishment, with the exception of the Engrailed genes, appears normal until early somitogenesis. By contrast, in spgm793 mutants, the initial expression of MHB genes is reduced in late gastrulation. Third, expression of pou2 appears to be unaltered in noi/pax2.1 mutants until at least the five-somite stage (data not shown).

Although the initial expression of fgf8 in the hindbrain primordium appears normal in spg mutants, it is reduced in the MHB by the two-somite stage and progressively ceases during somitogenesis. Thus, pou2 or pou2-dependent activities appear to be required for maintenance of fgf8 expression, rather than its establishment. Alternatively, the defect in fgf8 maintenance in spgm793 mutants may be secondary to the abnormal expression of other MHB genes, and an early consequence of the failure to establish the regulatory maintenance loop. In noi/pax2.1 mutants, normal levels of fgf8 expression have been reported until the five-somite stage (Lun and Brand, 1998). Thus, pou2 appears to be required before pax2.1 to maintain normal fgf8 expression during MHB establishment.

Pax2 directly regulates the Engrailed genes in the mouse MHB primordium (Song et al., 1996). In noi/pax2.1 zebrafish mutants, MHB expression of eng1/2/3 is strongly reduced, suggesting that pax2.1 is also a direct regulator of the Engrailed genes in zebrafish (Brand et al., 1996; Lun and Brand, 1998). In spgm793 mutants, expression of all engrailed genes (Fig. 4; data not shown) is reduced from its onset. Generally, expression of eng2 and eng3 remains weaker in noi mutants (Lun and Brand, 1998) than in spgm793 mutants. Furthermore, the residual expression domains of eng2 and eng3 in spgm793 mutants correspond to those of pax2.1. These observations suggest that the remaining pax2.1 activity present in spgm793 mutants is responsible for the residual dorsal expression of eng2 and eng3, and that pou2 is not a direct regulator of the engrailed genes.

In zebrafish, MHB maintenance is thought to begin near the five-somite stage, when wnt1 and fgf8 expression become dependent on the function of pax2.1 (Lun and Brand, 1998). By early somitogenesis, pou2 expression in the MHB region disappears. Thus, pou2 is unlikely to be directly involved in maintenance of the MHB organizer. In spgm793 mutants, gene expression during MHB establishment appears to be disrupted in such a way that the regulatory circuit involving wnt1, fgf8, pax2.1, and eng2 is not properly established. Overexpression of pou2 by injection of pou2 mRNA into one-cell stage spgm793 mutants can induce continued expression of pax2.1 and wnt1 in the MHB until at least 28 hpf. These findings suggest a transient requirement for pou2 to contribute to the activation of early-acting genes, including wnt1 and pax2.1, which, together with fgf8 and eng2, later maintain MHB development.

MHB expression of pax2.2/5/8 is activated during early somitogenesis, and, in spgm793 mutants, is affected from its onset. Analysis of noi/pax2.1 mutants showed that pax2.1 is required for the initiation of pax5 and pax8 expression in the MHB, but that pax2.2 MHB expression is independent of pax2.1 (Pfeffer et al., 1998). Although the role of pax2.2 in MHB development is unclear, it does not appear to compensate for pax2.1 function (Pfeffer et al., 1998). As expression of pax2.2/5/8 begins at a time when pou2 is no longer expressed in the MHB primordium, pou2 is not likely to directly regulate these genes. In addition, the defects in pax5 and pax8 expression at the MHB in spgm793 mutants are very similar to those observed in noi/pax2.1 mutants, raising the possibility that the requirement of spg/pou2 for Pax gene expression is mediated entirely through the activity of pax2.1. As pax2.2 expression is dependent on pou2 but not on pax2.1 function, activation of pax2.1 and pax2.2 present distinct requirements for pou2.

Injection of pou2 mRNA into spgm793 mutant embryos can restore expression of pax2.1 and wnt1 at the MHB. As we did not observe ectopic expression of pax2.1 or wnt1 after injecting even large amounts of pou2 mRNA, the ability of pou2 to induce expression of these genes appears to be limited to the MHB region. Expression of pax2.1 and wnt1 in other regions, such as the optic stalk, otic vesicle, midbrain and spinal cord, appeared unaltered by pou2 overexpression. Thus, the requirement for pou2 in the MHB appears permissive: while pou2 is essential for expression of pax2.1 and wnt1 at the MHB, it is insufficient to induce ectopic expression of these genes. This apparent permissive requirement for pou2 suggests either that its function requires the presence of additional factors in the MHB, or suppression of Pou2 activity outside of the MHB region. Other gene products required for MHB establishment may directly interact with Pou2, or indirectly provide competence within the MHB to respond to Pou2. Interaction with co-factors has been shown to determine activity as well as specificity of POU transcription factors (Ryan and Rosenfeld, 1997).

A requirement for interactions between Pou2 and one or more co-factors may serve to integrate different developmental patterning pathways that work during MHB establishment. Interestingly, the spg mutant phenotype shows a dorsoventral gradient in severity, and is most pronounced in the ventral MHB. Thus, pou2 may interact with midline regulatory pathways, although expression of genes required in midline development, such as shh and axial, appears normal in spg mutants (data not shown). Fgf4 is transiently expressed in the chick axial mesoderm underlying the midbrain and hindbrain and may thus contribute a vertical signal to MHB establishment. Accordingly, exogenous FGF4 can induce expression of engrailed in the chick neural plate (Shamim et al., 1999). If so, pou2 activity and FGF signaling may cooperate to establish the ventral MHB primordium. Vertical signals from axial mesoderm do not appear to play a significant role in zebrafish MHB establishment, however, as mutants devoid of axial mesoderm still form a MHB organizer (Rhinn and Brand, 2001). Alternatively, the midline defects observed in spg mutants may be caused by impaired convergence movements. Consistent with this idea, the neural plate and neural keel in the area of the MHB appear slightly wider in spgm793 mutants than in wild-type embryos (data not shown). Other zebrafish mutants that exhibit much more severe convergence defects, such as knypek and trilobite double mutant embryos (Marlow et al., 1998), form a MHB, however, suggesting that the minor impairment in convergence seen in spg mutants cannot explain the extensive MHB defects observed.

In this study, we have presented direct evidence for a novel role of the POU-domain transcription factor Pou2 in the establishment of the isthmic organizer in zebrafish. This raises the question of whether other vertebrate classes possess a pou2 ortholog and whether the function of this ortholog has been conserved during vertebrate evolution. Zebrafish pou2 is most similar to the mammalian transcription factor Pou5f1 (formerly Oct3/4) (Schöler et al., 1989), a Class V POU gene that shares 79% and 74% of amino acid identity within the POU-specific and POU-homeodomain, respectively (Takeda et al., 1994). Two lines of evidence suggest that Pou5f1 may be an ortholog of pou2. First, zebrafish pou2 maps on LG 21 in proximity to the complement factor B gene (BF) (Postlethwait et al., 1998). This synteny has been conserved in both human and mouse: in mice, Pou5f1 and H2-Bf map to chromosome 17, only 0.35 cM apart (Woods et al., 2000; Blake et al., 2000). Second, zebrafish pou2 and mammalian Pou5f1 share important features of their early expression patterns. Both genes are expressed maternally, and early zygotic expression is first found in pluripotent embryonic cells. In contrast to pou2, however, Pou5f1 becomes progressively restricted to the germ line (Pesce et al., 1998) and expression in the neural plate has not been reported. Thus, if pou2 and Pou5f1 are true orthologs, their later functions seem to have diverged between the teleost and mammalian lineages, and the role of pou2 in establishing the isthmic organizer may be unique.

Other POU domain transcription factors appear to regulate development of the mammalian MHB. In mouse, a canonical POU domain binding site has been shown to be essential for the MHB expression of Pax5 (Pfeffer et al., 2000). However, pax5 appears to be controlled by distinct mechanisms in zebrafish and mice (Pfeffer et al., 2000) and pax5 is unlikely to be a direct target of pou2. The phenotypes of Pax2/5 double mutants in mice suggest that the regulatory relationships governing the establishment of the isthmic organizer have undergone various changes during vertebrate evolution (Schwarz et al., 1997). Thus, distinct POU-domain proteins may act in the establishment of the isthmic organizer in fish and mouse.

In conclusion, pou2 appears to be required for MHB establishment in zebrafish as, in the absence of pou2 function, none of the genes implicated in MHB establishment that we have examined is properly activated. By contrast, lack of pou2 does not interfere with the positioning of the MHB primordium. We, thus, surmise that pou2 is involved in the translation of positional information provided by the otx2/gbx1 boundary into successful establishment of the MHB.

We thank Z. Varga, S. Ryu and J. Holzschuh for critical reading of the manuscript; G. Wussler for animal care; E. v. Seydlitz for technical assistance; and M. Brand, A. Molven, M. Ekker, A. Fjose, E. Weinberg, P. Pfeffer and T. Gerster for cDNA constructs and fish strains. This work was supported by grants from NIMH and DFG, a Landesschwerpunktprogramm Baden-Württemberg (to W. D.), an EMBO Fellowship (to D. M.), and a DFG Fellowship (to G. H.).