Fgf8 is mutated in zebrafish acerebellar (ace) mutants and is required for maintenance of midbrain-hindbrain boundary development and somitogenesisy

Generation of the large number of different cell types in the nervous system requires cell intrinsic programs and coordination between neighbouring cells. Work in recent years has established that designated cell populations exist in the neural plate that influence cell fate in surrounding neural plate cells. One such population is located at the boundary between midbrain and hindbrain (MHB), also referred to as isthmus (Alvarado-Mallart, 1993; Nakamura et al., 1994, Marin and Puelles, 1994; for review: Bally-Cuif and Wassef, 1995; Joyner, 1996; Lumsden and Krumlauf, 1996).

The midbrain derives from the mesencephalic neural plate and includes as major derivatives the optic tectum and the tegmentum. When MHB tissue is transplanted into the caudal forebrain primordium, midbrain-hindbrain markers are not only expressed in the transplanted tissue, but also in the surrounding forebrain tissue. When such transplants are allowed to develop, the induced cells show a midbrain-like character (Gardner and Barald, 1991; Martinez et al., 1991; Bally-Cuif et al., 1992). Substructures in the induced midbrain and MHB tissues are arranged in the normal sequence relative to each other, but are inverted with respect to the endogenous midbrain and cerebellum (Marin and Puelles, 1994). Similarly, transplantation of the MHB cells into the dorsal spinal chord leads to induction of a second cerebellum (Martinez et al., 1995). These experiments identify the MHB region as an important organizing center with a role in midbrain and cerebellar induction and patterning.

At the MHB, several molecules are expressed that have been implicated in cellular interaction processes and could mediate the activity of the MHB organizer. Wnt1 is a cognate of the secreted wingless gene product in Drosophila, and is expressed from the early neural plate stage onwards at the MHB of mouse and other vertebrate embryos (Wilkinson et al., 1987). Targeted inactivation of wnt1 has demonstrated its requirement during maintenance, but not initiation, of midbrain and cerebellar development of mouse embryos (Thomas and Capecchi, 1990; McMahon et al., 1992). In Drosophila, wingless cooperates with the Engrailed transcription factor in several cellular interaction processes, and its vertebrate homologues En1 and En2 are likewise expressed during and required for maintaining early MHB development (Wurst et al., 1994; Millen et al., 1994). Indeed, a major role for wnt1 during MHB development is to maintain expression of En1 (Danielian and McMahon, 1996). Apart from wnt1, engrailed expression also requires the activity of the paired box gene Pax2.1 (formerly pax-b, Pfeffer et al., 1998) in zebrafish and mice (Brand et al., 1996; Favor et al., 1996; Lun and Brand, unpublished data), consistent with the presence of binding sites for paired box proteins in the En2 promoter (Song et al., 1996).

Members of the family of secreted fibroblast growth factors (Fgfs) signal through receptor-tyrosine kinases (Fgfrs 1 to 4) to activate ras signaling (Basilico and Moscatelli, 1992). Fgfs play important roles during growth and patterning in the embryo. For instance, injection of dominant negative Fgf receptor constructs in Xenopus and zebrafish leads to posterior truncation, demonstrating that Fgf signaling is required during gastrulation and mesoderm development (Kroll and Amaya, 1996; Griffin et al., 1995).

In this study, we describe the isolation and expression of zebrafish Fgf8, and its functional requirement in embryonic development. Fgf8 was originally isolated as an androgen-induced growth factor (AIGF, Tanaka et al., 1992). Fgf8 is expressed in chicken and mouse from early somitogenesis onwards at the MHB and in a number of other cell groups with signaling properties (Heikinheimo et al., 1994; Ohuchi et al., 1994; Crossley and Martin, 1995; Mahmood et al., 1995). Implantation of beads soaked in Fgf8 or Fgf4 in chicken induces midbrain or cerebellar tissue in a manner analogous to cells of the MHB organizer, and Fgf8, En2 and wnt1 are activated by the implantation (Crossley et al., 1996a). These experiments strongly suggested that Fgf8 or a similar Fgf is an important component of MHB organizer function. Consistently, the receptors that Fgf8 and Fgf4 bind to in vitro are expressed during MHB development in mouse and zebrafish (MacArthur et al., 1995, and references therein; Thisse et al., 1995; Ornitz et al., 1996; Blunt et al., 1997).

Apart from the MHB, Fgf8 has been suggested to be a key signaling molecule in development of the limb bud (review: Cohn and Tickle, 1996), forebrain (Shimamura and Rubenstein, 1997), tooth (Neubüser et al., 1997), among others. In the limb bud, Fgf8 and Fgf4 are expressed in the apical ectodermal ridge (AER), which directs outgrowth of the limb. Mesenchymal cells of the zone of polarizing activity (ZPA) impose anteroposterior pattern on the limb bud, an activity that is mimicked by Sonic hedgehog (Shh) (Riddle et al., 1993), and Fgf8 and Fgf4 are thought to act in a feedback loop controlling Shh in the ZPA. Fgf beads or Fgf-expressing cells for Fgf1, Fgf2, Fgf4, Fgf8 and Fgf10 are all able to induce an additional limb from the flank of chick embryos (review: Cohn and Tickle, 1996; Ohuchi et al., 1997a), raising questions about the relative role of the various Fgfs. Both Fgf8 and Fgf10 are expressed early enough in the mesenchyme that is thought to induce the limb bud. In two chicken mutants, however, limb buds are established independently of Fgf8 expression (Niswander, 1997), and Fgf8 may therefore mimic the action of Fgf10 in limb induction (Ohuchi et al., 1997a).

Loss-of-function mutations for several Fgfs often display weaker phenotypes than anticipated from their expression patterns or misexpression experiments. Although Fgf3 and Fgf5 are expressed from gastrulation onwards, targeted inactivation of Fgf3 leads only to later defects in morphogenesis and differentiation of the inner ear and somites (Mansour et al., 1993) and Fgf5 mutants have fur alterations (Hebert et al., 1994). Others show very severe phenotypes: Fgf4 mutants die shortly after implantation (Feldman et al., 1995). Likewise, inactivation of Fgfr1 leads to absence of somites and expansion of notochord, suggesting that these embryos cannot respond to an unidentified, organizer-derived signal required to pattern the gastrula embryo (Deng et al., 1994; Yamaguchi et al., 1994). Also, heterozygous mutations in human Fgfr1 to Fgfr3 cause dominant defects in craniofacial development, vertebrae and limbs, indicative of functions in later development (review: Yamaguchi and Rossant, 1995). The effects of loss of Fgf8 function have not been described yet in zebrafish, but a recent study in mice indicates that Fgf8 is required in gastrulation and brain development (Meyers et al., 1998).

As in other species, zebrafish Fgf8 is expressed during gastrulation, in mesodermal tissue, during early MHB development and several other sites in the nervous system. We show here that Fgf8 is mutated in acerebellar (ace). A single recessive acerebellar allele exists, and homozygous mutant embryos lack a MHB and a cerebellum (Brand et al., 1996). We analyze the requirement for Fgf8 in the mutants, compare the effects of misexpressing wild-type and mutant Fgf8 transcripts, and examine Fgf8 dependence in no isthmus mutants which inactivate Pax2.1 (Brand et al., 1996; Lun and Brand, unpublished data).

Zebrafish were raised and kept under standard conditions at about 27°C (Westerfield, 1994). Mutant carriers were identified by random intercrosses. To obtain mutant embryos, heterozygous carriers were intercrossed. Time of development at 28.5°C and morphological features were used to stage the embryos (Kimmel et al., 1995). Occasionally, 0.2 mM phenylthiourea (PTU) was added to prevent melanization. Histology is described in Kuwada et al. (1990).

Digoxigenin- or fluorescein-labelled RNA probes were prepared from linearized templates using an RNA labelling and detection kit (Boehringer). Hybridisation and detection with anti-digoxigenin or anti-fluorescein antibodies coupled to alkaline phosphatase (Boehringer) was modified from Thisse et al. (1994). Hybridisation was at 68.5°C, and Boehringer DIG blocking agent was used during detection as specified by the supplier. 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. Antibodies were preabsorbed against fixed embryo powder. Stained embryos were dissected and thick sections were prepared with sharpened tungsten needles, mounted in glycerol, photographed on a Zeiss Axioskop and assembled using Adobe Photoshop.

Fgf8 was isolated from a λgt11 library (kindly provided by Kai Zinn) using as probe the coding sequence of mouse Fgf8 variant 4 during the initial screen, and a chicken Fgf8 cDNA during rescreen (Crossley et al., 1996b). Candidates were subcloned into pCRII and sequenced (accession number AF051365). One additional zebrafish gene of uncertain relationship resembles Fgf8, but also other Fgfs; in contrast to the Fgf8 gene reported here, this gene is expressed much later in development than Fgf8 (S. Schulte-Merker, personal communication, and F. R. and M.B., unpublished data).

To determine linkage, heterozygous carriers for acerebellar (induced in the Tübingen strain) were crossed to AB wild-type strain. Carriers were identified in F₁and intercrossed. Embryos from such crosses were separated into homozygous acerebellar mutants (n=100 and 108 for two independent experiments) and their siblings (n=100), and DNA and cDNA was prepared from each pool. cDNA synthesis with SuperScriptII reverse transcriptase (GibcoBRL) was according to manufacturers instructions. Intron sequences between exons 1 and 2 (1.6 kb) were amplified from both pools and from Tübingen and AB strains, assuming that exon/intron structure would be conserved relative to mouse Fgf8 (Crossley and Martin, 1995). This assumption was confirmed by our results, and by sequencing of the amplified introns (not shown). Amplified fragments were digested with BglII, which detects a restriction fragment length polymorphism (RFLP) between the Tübingen and AB strains; the resulting gel was blotted and probed with a fragment containing only intron sequences to confirm that the fragments are derived from the Fgf8 locus (not shown). Linkage was also observed for a second RFLP (not shown). Equivalent amplifications were carried out to obtain and sequence the exon 2/3 intron. cDNA was isolated by RT-PCR with nested primers flanking the coding region in two independent amplifications from cDNA pools of homozygous Tübingen wild-type and acerebellar embryos, and was subcloned and sequenced on an ALF sequenator. RT-PCR to detect presence of exon2 was carried out on cDNA from wild-type and acerebellar embryos under standard PCR conditions.

Wild-type and acerebellar mutant versions of Fgf8 were subcloned into pCS2+ (Rupp et al., 1994) and transcribed using the SP6 message machine kit (Ambion). The amount of RNA injected was estimated from the concentration and volume of a sphere of RNA injected into oil at the same pressure settings. Typically, about 25 pg of Fgf8 RNA were injected; higher concentrations cause more severe dorsalizations that lead to rupture of the embryos during somitogenesis (not shown). RNA was dissolved in 0.25 M KCl with 0.2% of phenol red and backloaded into borosilicate capillaries prepared on a Sutter puller. During injection, RNA was deposited into the cytoplasm of 1- to 8-cell-stage embryos; in embryos after the first cleavage, the RNA usually stays in the progeny of the injected blastomere, as judged from the unilateral distribution of control lacZ RNA, as detected with anti-β-gal antibody (Promega, 1:500) after ISH (Dornseifer et al., 1997).

We have isolated zebrafish Fgf8 from an embryonic cDNA library. The aminoacid sequence of zebrafish Fgf8 is 79% identical to mouse and human Fgf8, and 84% identical to chicken Fgf8 (Fig. 1). Amino acids encoded by exon 2 are diagnostic for Fgf8 relative to other Fgf family members (Lorenzi et al., 1995); here the identity is 83% to mouse and human, and 91% to chicken Fgf8, with other similarities being much lower (e.g. Fgf7, 37% and Fgf4, 29%).

To study possible functions of Fgf8, we examined expression in wild-type embryos using whole-mount in situ hybridisation (ISH; Fig. 2). Expression becomes detectable at 30% epiboly in the marginal zone, and develops at 50% epiboly into a gradient with a highpoint in the dorsal embryonic shield, the zebrafish equivalent of Spemann’s organizer (Fig. 2A-C). During gastrulation, dorsoventrally graded expression continues in the marginal zone. At 70% epiboly, expression starts in two transverse stripes in the anterior hindbrain primordium (Fig. 2D,E), and towards the end of epiboly at the anterior margin of the forebrain primordium (Fig. 2F).

During somitogenesis, expression continues in the prospective MHB (see below) and the tailbud, and is initiated in presomitic mesoderm in segmental expression domains (Fig. 2H,J). Expression is found throughout newly formed somites (Fig. 2J), but eventually becomes confined to the anterolateral margin of the maturing somite (Figs 2K, 10R). Transient expression occurs in the floorplate as it emerges from the tailbud (not shown). Posterior to the MHB, three additional stripes are detected in the hindbrain neural keel during early somitogenesis (Fig. 2H). In the forebrain primordium, a dorsomedian stripe is observed in the presumptive telencephalon with an anterior high point of intensity (Fig. 2H,I).

In the brain of pharyngula stage embryos (24-48 hours), expression is still prominent in the MHB, excluding the floorplate (Fig. 2O), and in the optic stalks, retina, a pair of paramedian telencephalic stripes that forms the commissural plate and in the dorsal diencephalon (Fig. 2N-R). Additional expression is seen in the dorsal hypothalamus ventral to the optic recess, in the area where the postoptic commissure will form (Fig. 2P). Around 36 hours, expression is detected in addition near the ventral midline of the hypothalamus in the hypophysis and infundibulum (Fig. 2P,Q), and in the nasal placodes (Fig. 2S). Expression continues in these tissues until 48 hours, the latest stage that we have examined (not shown).

Outside of the brain, expression is found in tissues of the developing head, such as the hyoid, heart, inner ear (Fig. 2L, P-S) and the fin buds (Fig. 11).

Because of its possible patterning function in development of the MHB territory, we have examined the expression in this area in more detail. Fgf8 activation is seen initially as a bilateral stripe at 70% epiboly (Fig. 3A-D). Relative to Krox-20, a marker for rhombomeres 3 and 5 (Oxtoby and Jowett, 1993; Fig. 3A,B), this stripe encompasses by tailbud stage the anterior hindbrain up to and including rhombomere 4. At 5 somites, this domain has become subdivided into several stripes lying at the MHB, in rhombomeres 1 and 4 and ventral rhombomere 2 (Figs 3C-D, 2H). In double stainings with the midbrain marker Pax2.1, Fgf8 expression (Fig. 3E-H) is localized posterior to the domain of Pax2.1 expression at 90% epiboly, with very little, if any, overlap. At 6 somites, however, the MHB stripe is completely contained within the posterior part of the Pax2.1 domain (Fig. 3I,J).

Fgf8 expression occurs in several tissues that are defective in acerebellar mutant embryos. Mutant larvae older than 2 days are retarded and eventually die with severe oedemas (not shown), but develop without significant retardation during the first 48 hours of development. In particular, homozygous acerebellar mutants lack a MHB and a cerebellum (see below). By testing candidate genes, we found that Fgf8 is linked to the acerebellar mutation. In a test cross with two segregating RFLPs of Fgf8, the RFLP characteristic for the ‘Tübingen’ strain (in which acerebellar was induced) is linked to the acerebellar phenotype (Fig. 4). This RFLP was located in intron one and was lost during subsequent generations. However, by RT-PCR with single embryos, we found that the acerebellar phenotype and the lesion in Fgf8 (see below) could not be separated in 101 embryos representing 202 meiotic events (0±0.5 cM; Fig. 5D).

Through characterizing the Fgf8 gene in acerebellar embryos, we found that acerebellar is a mutation that strongly or completely inactivates the Fgf8 gene. We used RT-PCR to amplify Fgf8 cDNAs from homozygous acerebellar embryos and compared them to wild-type cDNA (Fig. 4). In two independent amplifications of the coding region, we find a deletion of 107 bases exactly corresponding to exon 2 in the mutant cDNAs; no other amino acid changes were detected. In order to study how this deletion is generated, we examined genomic DNA of acerebellar embryos and found that exon 2 is both present and of normal sequence. Upon sequencing of the 1.6 kb intron between exons 1 and 2, no conspicuous changes were detected and, in particular, the splicing consensus sequences were found to be intact (not shown). However, in the 5′ splice donor site following exon 2, a G residue was mutated to an A (Fig. 4C). Since this G is 100% conserved in all 5′ splice donor sites (Padgett et al., 1986), this mutation may inactivate this splice site, leading to skipping of exon 2 in the mutants.

As a result of skipping exon 2, the open reading frame runs into a premature stop codon (Fig. 4D). The predicted protein fragment in acerebellar embryos therefore lacks the aminoacids encoded in exons 2 and 3, which are required to activate the receptor and which are conserved between different Fgf8s and other Fgf family members (Lorenzi et al., 1995). The protein fragment in acerebellar mutants is therefore presumably non-functional, a notion that is confirmed by our injection experiments (see below).

Exon 1, but not exon 2, is alternatively spliced in murine Fgf8 (Crossley and Martin, 1995; MacArthur et al., 1995). We find no evidence for differential splicing of exon 2 of Fgf8 in zebrafish : in a timecourse up to day 5, we detect in wild-type embryos a single transcript of the size predicted for transcripts containing exon 2 (Fig. 5). This transcript is also detected at the 4- to 8-cell stage, i.e. prior to activation of zygotic transcription (Kane and Kimmel, 1993), showing that maternal Fgf8 message is present in these embryos (Fig. 5B). We could not, however, detect any maternal RNA by in situ hybridisation (not shown), suggesting that these RNAs are rare.

To assess the strength of the acerebellar allele, it was crucial to determine if any wild-type Fgf8 transcript is present in the mutants. We therefore performed RT-PCR on cDNA from acerebellar embryos. With primers flanking exon2, we detect a single band of the size predicted for transcripts lacking exon 2, but no transcripts of wild-type size (Fig. 5C). With one primer in exon 2 and another in the flanking exons, exon-2-containing transcripts can be detected in acerebellar embryos, but only after two rounds of amplification (Fig. 5C). These transcripts could be of maternal origin (see above), or they could be due to partially spliced mRNA in our cDNA pools; although partially spliced transcripts are usually unstable and confined to the nucleus (Padgett et al., 1986; Khoury et al., 1979). We can at present not distinguish between these possibilities. In either case, wild-type transcripts containing exon 2 must be rare, since they are not detectable with primers flanking exon 2 or by in situ hybridisation. We conclude that acerebellar partially or completely inactivates the Fgf8 gene (see Discussion).

In order to determine whether the Fgf8 protein left in acerebellar mutants has any functional properties, we developed a functional assay for Fgf8 activity. When wild-type Fgf8 RNA is injected into developing embryos, we observe dorsalization and axis duplication (Fig. 6A-F; Table 1). We monitored adaxial and somitic development with myoD probe, and the location of the injected cells by coinjection of lacZ RNA. Embryos with a secondary axis formed in 12% of the cases (Table 1). Invariably, the cells of the secondary axis were the injected cells (Fig. 6B), and failed to form a head, notochord and adaxial cells. Exclusively dorsal location of the injected cells in the notochord primordium has little or no effect, suggesting that this tissue is not competent to respond (Fig. 6C). In dorsolateral levels, Fgf8 misexpression causes severe expansion of the somites to ventral levels (Fig. 6D),which in some cases encircle the embryo (not shown). Consistent with the expansion of dorsal cell fates, Pax2.1 expression in the intermediate mesoderm is suppressed or shifted to more ventral levels (Fig. 6E). Notably, although MHB expression of Pax2.1 is expanded ventrally, it is not expanded along the anteroposterior axis (Fig. 6F), showing that Fgf8 is not sufficient to induce Pax2.1 expression. These findings are summarized in Fig. 6I. Our misexpression studies do not necessarily imply that Fgf8 normally functions in dorsoventral patterning, but they do provide a sensitive assay for functional activity of ectopically expressed Fgf8 transcripts. In contrast to the severe effects of misexpressing wild-type Fgf8, injections of the ace mutant Fgf8 at the same or a tenfold higher concentration causes no effect (Fig. 6G; Table 1). We conclude that the Fgf8 transcript lacking exon 2 is inactive.

Examination of living embryos and of histological sections shows that the MHB fold and the cerebellum are absent in acerebellar embryos (Fig. 7). In living embryos, the MHB fold and the posteriorly adjacent cerebellar primordium are missing (Fig. 7A,B). In histological sections, the MHB tissue is eliminated posterior to the tectum opticum, which itself often appears slightly enlarged (Fig. 7C-F). The MHB can be subdivided into an anterior portion (probably still part of the midbrain) and a posterior portion, which is thought to give rise to cerebellum (Kimmel et al., 1995). Both portions are absent in acerebellar mutants, showing that the defect is not restricted to the cerebellar primordium. We do not know the fate of the prospective fold tissue in acerebellar mutants but, since we have not detected cell death in this region previously (Brand et al., 1996), the apparent increase in the tectal tissue could reflect a transformation to a midbrain fate.

Our above results show that cerebellar and MHB development fail during embryonic stages in acerebellar mutants. We therefore examined the expression of marker genes of this area in a detailed time course (Fig. 8, Table 2). Like Fgf8 itself, wnt1, Pax2.1, Eng1, Eng2, Eng3 and Her5 are expressed during normal development of the MHB primordium (Krauss et al., 1991; Molven et al., 1991; Ekker et al., 1992; Müller et al., 1996). After initially widespread expression in the midbrain primordium, expression of these genes is restricted during mid-somitogenesis towards the posterior midbrain and MHB. Expression of all marker genes that we examined is initiated normally in acerebellar mutant embryos. During early to mid-somitogenesis stages, however, expression gradually fades in acerebellar mutants, but continues in wild-type siblings. The expression domains of all markers gradually narrow and seem to persist longest in dorsoposterior parts of the MHB; eventually, expression is completely eliminated in the mutants (Fig. 8, Table 2). The earliest defect is seen for Her5 at the 5-somite stage (Fig. 8G,J). These results show that maintenance, but not initiation of gene expression at the MHB, is affected in acerebellar mutant embryos.

To examine if establishment of Fgf8 signalling is dependent on Pax2.1, a gene required for early midbrain development (Brand et al., 1996; Lun and Brand, unpublished data), we examined Fgf8 expression in no isthmus (noi) mutant embryos for noitu29a, which lack a functional Pax2.1 gene (Fig. 9). Up until the 10 somites stage, we observe no difference in Fgf8 expression between wild-type and noi embryos (Fig. 9A). At 18 somites, expression is eliminated at the MHB, but not in several other tissues (Fig. 9B,C). Since noi mutants lack the isthmus at this stage (Brand et al., 1996), this is most likely a secondary consequence of elimination of the tissue. We conclude that Fgf8 signalling in the early MHB primordium is activated independently of Pax2.1/Noi.

To study whether Fgf8 functions during development of the mesodermal derivatives that it is expressed in, we examined ace mutants with marker genes for axial, paraxial and intermediate mesoderm, and found defects that are probably due to weakly abnormal dorsoventral patterning (Fig. 10). myoD is expressed in adaxial cells lateral to the notochord from 80% epiboly onwards, and later spreads to the forming somites in the paraxial mesoderm (Weinberg et al., 1996). myoD expression in adaxial cells of acerebellar mutants is strongly reduced at 80% to tailbud stages, and is interrupted during somitogenesis stages (Fig. 10A-F).

Since Fgf8 is not yet expressed in adaxial cells or notochord between 80% and tailbud, the early failure to express myoD could reflect a weak requirement for Fgf8 in dorsoventral patterning. To test a possible requirement in dorsoventral patterning, we examined expression of eve1, fkd3 and BMP4 as markers for ventral and dorsal cell fates (Joly et al., 1993; Chen et al., 1997; J. Odenthal, unpublished) and found no difference at 50%, 80% and tailbud stages (not shown). Likewise, Pax2.1 expression in the intermediate mesoderm is normal at 7 somites (Fig. 10I,J). The tailbud is viewed as a site of continuing gastrulation and patterning (Gont et al., 1993) and expresses Fgf8. We find snail1 expression (Hammerschmidt and Nüsslein-Volhard, 1993) to be absent in the vicinity of the tailbud, possibly reflecting another weak function for Fgf8 in gastrulation or patterning (Fig. 10K,L). We conclude that early defects of myoD and snail1 expression in acerebellar mutants could reflect a weak requirement in dorsoventral patterning. Notably, this requirement is most apparent in the future adaxial and somitic mesoderm, close to the site of highest Fgf8 expression on the dorsal side.

Following activation throughout the somites and adaxial cells, Fgf8 expression is successively confined to anterior-lateral cells of wild-type somites (Fig. 10Q,R). Both somite and adaxial cell development is affected in acerebellar mutants. During midsomitogenesis, myoD and snail1 are expressed in condensing somites of the wild type, but are reduced and patchy in acerebellar mutants (Fig. 10E,F,M,N). Towards the end of the segmentation period, wild-type somites assume a distinct chevron shape and continue to express myoD and snail1 (Fig. 10G,O). In acerebellar mutants, the somites appear more block-shaped, and have strongly reduced levels of myoD and snail1 (Fig. 10H,P).

To examine development of adaxial cells, we studied expression of Eng in muscle pioneers that are derived from a subset of adaxial cells (Devoto et al., 1996). Expression of Eng1 (not shown) and Eng2 is reduced in acerebellar mutants at 24 hours (Fig. 10S,T). Adaxial cells themselves depend on signals from the notochord (Halpern et al., 1993). Brachyury/T expression as a marker for notochord (Schulte-Merker et al., 1992) was however unaffected at 80%, tailbud and 5-somite stages (not shown). In zebrafish you-too mutants, adaxial cells are missing, without affecting overall myoD expression in the early somites (van Eeden et al., 1996). In contrast, acerebellar mutants are defective both in early adaxial cell development (Fig. 10A-D) and in somitic expression of myoD and snail1 (see above). Fgf8 may therefore function independently in development of both cell types; alternatively, the defect could be an indirect consequence of the earlier abnormal expression of paraxial mesoderm genes like myoD and snail1. Given its anteriorly restricted expression pattern at later stages, Fgf8 might also function in polarization of somites along the craniocaudal axis (Hrabe de Angelis et al., 1997). However, in mutant somites, Fgf8 itself is still expressed anteriorly, and snail1 is more highly expressed posteriorly as in wild type, suggesting that the mutant somites are still patterned along the rostrocaudal axis (not shown).

Teleost pectoral fins are homologous to tetrapod forelimbs (Sordino et al., 1995). Fgf8 is discussed as an important regulator of limb development, possibly by maintaining or inducing expression of sonic hedgehog (shh) in the zone of polarizing activity (ZPA; see Cohn and Tickle, 1996, for review). In contrast to chicken and mice, Fgf8 expression starts in the pectoral fin bud ectoderm only after initial fin bud formation, at 36 hours (Fig. 11B). At 48 hours, Fgf8 is confined to the distal-most ridge of the developing fin (Fig. 11C,M), the equivalent of the apical ectodermal ridge of other vertebrates (Wood, 1982). eng1 and shh are activated earlier than Fgf8 in the fin bud of wild-type embryos (Hatta et al., 1991; Krauss et al., 1993), and Fgf8 is therefore probably not involved in the induction and early patterning of the fin bud in zebrafish (Fig. 11A,E,H). To examine if Fgf8 could function in later stages of fin development, we analysed shh expression in the ZPA, and eng1 expression in the ventral-anterior fin bud, but could not detect an effect in acerebellar embryos (Fig. 11F,G,I,J). Furthermore, the overall structure of fins on day 5 of development appears normal in living acerebellar larvae (Fig. 11K,L). We do, however,consistently observe a slight increase in Fgf8 expression in the mutants at 48 hours, possibly reflecting a later function for Fgf8 (Fig. 11D). Although there are no obvious defects, our data do not exclude a late function for Fgf8 in development of the fin bud, since other Fgfs could compensate for missing Fgf8 activity, as reported for other vertebrates (Cohn et al., 1995; Crossley et al., 1996b; Ohuchi et al., 1997a).

We have described the isolation and expression pattern of zebrafish Fgf8 and have shown that acerebellar is a mutation in Fgf8. Unexpectedly, our analysis demonstrates a requirement for Fgf8 in maintenance, but not initiation, of MHB development. In addition, Fgf8 is weakly required for normal dorsoventral patterning and in somite development. Other tissues, like the pectoral fins, apparently require Fgf8 to a lesser degree. Maternally supplied Fgf8 or other Fgfs may compensate for the loss of Fgf8.

Within the Fgf family, zebrafish Fgf8 is most closely related, by sequence, expression pattern and genomic structure, to the Fgf8 subgroup. During development, Fgf8 is expressed in many cell populations that are known to be important signaling centers, such as the shield (Spemann’s organizer), the anterior edge of the neural plate, the MHB and the limb bud. A similar association of Fgf8 expression with signaling centers occurs in other vertebrates, often combined with expression of other Fgfs (Heikinheimo et al., 1994; Crossley and Martin, 1995; Mahmood et al., 1995; Bueno et al., 1996; Neubüser et al., 1997). Our detailed analysis of Fgf8 expression at the MHB shows that even within a given domain expression can be very dynamic.

The combined data of our linkage studies, analysis of the Fgf8 locus in acerebellar mutants, and of our phenotypic studies show that acerebellar is mutated in Fgf8. A key issue is to what extent Fgf8 activity is inactivated in acerebellar. Our analysis shows that mutation of a 100% conserved residue in a splice donor site following exon 2 leads to skipping of exon 2 in mutant embryos (Fig. 4). In a direct comparison of the prevalence of transcripts containing or lacking exon 2 with flanking primers, only transcripts lacking this exon are detectable in acerebellar mutants (Fig. 5). In the more sensitive assay where one primer is located in exon 2, also transcripts containing exon 2 are detectable, but only after two rounds of PCR amplification; such transcripts are therefore probably rare. They could be maternal transcripts, or transcripts resulting from incomplete inactivation of the mutated splice site. Absence of exon 2 results in a frameshift and premature chain termination of the predicted mutant protein. The conserved amino acids encoded by exons 2 and 3 that are thought to be important for Fgf function (Basilico and Moscatelli, 1992) are absent, and the resulting truncated protein is therefore probably inactive. This prediction is confirmed by the results of our injection experiments with RNA encoding the ace mutant version of Fgf8, which even at 10-fold higher RNA concentration does not have a biological effect (Fig. 6G). While these results show that transcripts lacking exon 2 do not produce functional Fgf8 protein, the fact that we do observe a minor amount of wild-type message containing exon 2 means that acerebellar may not cause complete inactivation of Fgf8. Elimination of the maternal component and isolation of further alleles of Fgf8 can be used to address this issue.

In contrast to the drastic effect of misexpressing Fgf8 on patterning of the gastrula, acerebellar mutants display a surprisingly mild phenotype. One possibility is that ace is not a null allele. A stronger phenotype was recently described for mouse Fgf8 mutants: homozygous null Fgf8 mutants fail to gastrulate and have no mesodermal derivatives, whereas weaker alleles display phenotypes more akin to what is seen in acerebellar mutants, including deletions of the posterior midbrain and cerebellum (Meyers et al., 1998).

Other explanations are, however, also possible for the weaker phenotype of acerebellar mutants. The maternal Fgf8 RNA that we have observed (which contains exon 2, Fig. 5) could partially ameliorate the phenotype of acerebellar mutants, thus ‘masking’ a requirement for Fgf8 in zebrafish, but not in mice, which have little maternal cytoplasm. We do not consider this possibility very likely: in contrast to zygotic RNA, maternal RNA is only detectable using the much more sensitive PCR assay, but not by in situ hybridisation, and maternal RNA may not be localized.

A more likely possibility is offered by the observation that Fgf8 is often coexpressed with other members of the Fgf family in gastrulation (reviewed by Yamaguchi and Rossant, 1995). Prior to and during mouse gastrulation, Fgf3, Fgf4, Fgf5 and Fgf8 are expressed in distinct but overlapping patterns in the primitive streak: whereas Fgf5 is found throughout the gastrula ectoderm, Fgf3 is found in future mesodermal cells in the streak, and Fgf4 is at the anterior end of the streak. While the expression patterns of these genes are suggestive, only Fgf4 is required during gastrulation, whereas Fgf3 and Fgf5 are not (Feldman et al.,1995; Mansour et al., 1993; Hebert et al., 1994). An alternative explanation for the absence of severe gastrulation defects in ace mutants is therefore that other Fgfs can compensate for lack of Fgf8, or that Fgf8 has only a weak function.

Inactivation of Fgfr1 causes absence of somites, expanded notochords and primitive streak defects. Fgfr1 was therefore proposed to be the receptor for an organizer-derived signal that patterns paraxial mesoderm (Yamaguchi et al., 1994; Deng et al., 1994). Based on its expression pattern and the phenotypes seen after misexpression and loss of function, Fgf8 or a similar Fgf could be this signal. Similar observations on the effects of Fgf8 misexpression were made by M. Fürthauer, C. Thisse and B. Thisse, who also showed that Fgf8 misexpression alters the distribution of the Bmp4 morphogen (Fürthauer et al., 1997). We note that the mild defect in activation and overall expression of myoD in the acerebellar mutants is consistent with a function of Fgf8 in dorsoventral patterning, in keeping with the stronger gastrulation defect of the mouse mutant (Meyers et al., 1998).

The dynamic pattern of expression of Fgf8 at the MHB is compatible with the functional requirement that we have observed. In chicken, beads containing Fgf8 or Fgf4 protein placed into the posterior forebrain or alar hindbrain primordium are able to induce ectopic isthmic, midbrain and cerebellar structures, strongly suggesting a role for Fgfs in MHB development (Crossley et al., 1996a). Our analysis of Fgf8 requirement is generally compatible with these results. However, the bead experiments have raised the possibility that Fgf8 is the endogenous molecule which induces the midbrain, a notion that is not supported by several observations. (i) At the time Fgf8 is activated at late gastrulation stages, it clearly marks the anterior hindbrain (Fig. 3). Posteriorly, its expression extends to the rhombomere 4/5 boundary and, anteriorly, it abuts the Pax2.1 expression domain. Since the Fgf8 and Pax2.1 domains are largely non-overlapping at this stage, the early expression of Fgf8 is clearly not sufficient to induce MHB markers such as Pax2.1. (ii) Secreted Fgf8 might act on the anteriorly adjacent cells at a distance to induce midbrain fate. We have observed, however, that misexpression of Fgf8 leads to severe expansion of Pax2.1 only along the d/v direction during gastrulation (as a consequence of altered dorsoventral patterning, see Fig. 6), but not to an expansion along the anteroposterior axis as would be expected if anterior cells could respond to Fgf8. Similarly, delocalized Fgf8 expression does not alter early En1 and wnt1 midbrain expression in mice with altered Otx gene dosage; instead, the later restriction of wnt1 and Fgf8 to the posterior midbrain is affected (Acampora et al., 1997), similar to our findings in acerebellar. (iii) We find that Fgf8 and Pax2.1 expression come to gradually depend on each other only during mid-somitogenesis, after the time when the anterior-most Fgf8 subdomain in the MHB region is fully contained within the posterior Pax2.1 domain. We speculate that this time may coincide with the establishment of the isthmus organizer in the region of overlap. (iv) In mice, Fgf8 activation at the MHB occurs only at the 3- to 4-somite stage, and is thus preceded by activation of wnt1 and En1 as midbrain markers (Crossley and Martin, 1995; Mahmood et al., 1995). Taken together, Fgf8 is unlikely to act as the endogenous inducer of midbrain development in zebrafish and mice.

What could be the real inducer? Experimental manipulations in mouse, zebrafish and chicken have provided evidence for a vertical signal in late gastrula stages from mesendoderm to overlying ectoderm to activate expression of some, but not all midbrain markers (Ang and Rossant, 1993; Miyagawa et al., 1996; Darnell and Schoenwolf, 1997). To explain the ability of Fgf8 beads to induce midbrain, Crossley et al. (1996a) proposed that in normal development Fgf8 expression in cardiogenic mesoderm underlying the MHB could provide, by a vertical path, the inductive signal. In zebrafish, cardiogenic precursors have been fate mapped throughout development. They derive from ventrolateral levels of the germ ring and migrate during gastrulation into a longitudinal domain which moves closer to the axis during early somitogenesis (Stainier and Fishman, 1992). Because we have not seen any Fgf8 expression in these cells during gastrulation and because the shape and orientation of the cardiogenic domain (longitudinal) versus the Fgf8 ectodermal domains (transverse) are very different, cardiogenic precursors are unlikely to provide the inductive signal for the ectodermal expression at the MHB in zebrafish gastrulae. We cannot rule out, however, that part of these cell populations are adjacent to each other at some stage during development and that signaling may occur between them.

An alternative source for the midbrain-inducing signal is the germ ring from which mesendodermal tissues derive. Fate-mapping studies in zebrafish have shown that a midbrain primordium is already separately established at late gastrulation stages (Woo and Fraser, 1995). Transplantation studies have suggested that an unknown signal responsible for hindbrain induction is present in the germ ring. This signal is not mimicked by bFgf beads and is absent from the dorsal shield (which has high levels of Fgf8), and may therefore not be a member of the Fgf family (Woo and Fraser, 1997). During gastrulation, Fgf8-expressing and Pax2.1-expressing domains in the neural primordium look quite similar in shape and width. By analogy to the hindbrain, the germ ring could therefore also provide the signal responsible for midbrain induction.

On the basis of the evidence presented here, we suggest that early MHB development occurs in at least two phases. During the establishment phase in late gastrulation, midbrain and hindbrain primordia are set up independently, in a process that does not require Fgf8. Given that Fgf8 and pax2.1 are activated independently of each other, at least two independent signalling pathways must act in parallel during early MHB development. The establishment phase is followed during early somitogenesis by a maintenance phase during which gene expression in the midbrain depends on signal(s) from the MHB. The gain- and loss-of-function experiments in chicken and fish together suggest that Fgf8 is required for the maintenance phase, possibly in combination with wnt1. The beginning of the maintenance phase may be coincident with establishment of the isthmus organizer at the interface between mid- and hindbrain territories.

Why then is Fgf8 on its own, when misexpressed, sufficient to reprogram posterior forebrain to midbrain and/or MHB development? Many of the genes expressed in the maintenance phase in the isthmus (Pax2.1, wnt1, Eng2, Eng3, Fgf8, Her5) are also active earlier during the establishment phase in the midbrain. We have shown here that MHB expression of Eng genes, wnt1, Her5 and Pax2.1, all require Fgf8 activity during the maintenance phase, since they all start to fail in their expression around early to midsomitogenesis and are eventually eliminated in acerebellar mutants. It is thus likely that Fgf8 can impinge on their regulation. Thus, misexpressing Fgf8 probably ectopically activates the complement of genes that also acts during establishment of midbrain development. Indeed, at least Fgf8, wnt1 and En2 are ectopically activated following the Fgf8 bead insertion into neural plate tissue (Crossley et al., 1996a). Moreover, ectopic Fgf8 expression in embryos with altered otx gene dosage recruits En1 and wnt1 to the ectopic position only after some delay (Acampora et al., 1997). Once re-established in an ectopic position, the gene program could then develop accordingly.

What would be the normal function of Fgf8 during the maintenance phase? During this phase, expression of many marker genes is restricted to the posterior part of the midbrain primordium, towards the zone of overlap between Fgf8 and Pax2.1 at the isthmus. A crucial function for Fgf8, and possibly for the isthmus organizer in general, may therefore be to ensure polarized expression of midbrain markers, rather than initial induction. In keeping with this possibility, we find that all posterior midbrain markers we examined are absent from the midbrain of acerebellar mutants at later stages.

A distinction between establishment and maintenance functions for Fgfs has also been made for development of the chick limb bud (for a review, see Cohn and Tickle, 1996; Niswander, 1997). Similar to the situation at the MHB and as with other Fgfs, Fgf8 bead implantation is able to activate the full limb development program ectopically (Crossley et al., 1996b; Vogel et al., 1996). The earliest signal to establish limb development is thought to derived from the mesenchyme of the prospective limb bud. Fgf10 is expressed in the mesenchyme at the right time, preceding Fgf8 expression, and is able to induce complete limbs (Ohuchi et al., 1997a), so Fgf8 could mimic the limb-inducing action of Fgf10. Consistent with this possibility, in two chicken mutants, limbless and wingless, limb buds are established independently of Fgf8 expression (Grieshammer et al., 1996; Ros et al., 1996; Ohuchi et al., 1997b), and Fgf8 is only weakly required in mouse limb bud development (Meyers et al., 1998). In zebrafish, Fgf8 is neither expressed nor required during pectoral fin bud formation (Fig. 11), arguing that also in fish Fgf8 is not involved in fin bud establishment.

We thank M. Fürthauer, B. Thisse, Ch. Thisse, G. Martin, S. Martinez, S. Schulte-Merker and D. Ornitz for sharing unpublished results, Dave Willison for substantial contribution to the cloning of Fgf8, and C. Niehrs, F. Pelegri, P. Sordino, S. Wilson and members of the Brand laboratory for comments. F. R., H. B. and M. B. are funded by the Förderprogramm Neurobiologie, Baden-Württemberg, and the Deutsche Forschungsgemeinschaft (SFB 317). D. Y. R. S. is funded by the NIH and the Packard foundation, E. C. W. is a Howard Hughes Medical Institute predoctoral fellow.