Differentiation of cerebellar cell identities in absence of Fgf signalling in zebrafish Otx morphants

Following neural induction, local organisers act within the neural plate to refine local identities along the anteroposterior axis(Echevarria et al., 2003; Jessell and Sanes, 2000; Lumsden and Krumlauf, 1996). Understanding how these organisers influence their surrounding is a current challenge. As yet, two such local organisers have been described(Houart et al., 1998; Wilson and Houart, 2004; Wurst and Bally-Cuif, 2001). One of these, the isthmic organiser (IsO), placed at the interface between mesencephalon (or midbrain) and metencephalon (comprising rhombomeres 1 and 2;r1, r2), is necessary and sufficient for the development of these structures[also called mes/met domain (Liu and Joyner, 2001; Raible and Brand, 2004; Wurst and Bally-Cuif, 2001)]. The patterning activity of the isthmus was initially demonstrated in avian embryos by transplantation experiments inducing either transformation of part of the caudal forebrain into an ectopic midbrain (Alvarado-Mallart et al.,1990; Martinez et al.,1991), or re-fating of some of the posterior hindbrain into ectopic cerebellum [a major derivative of the metencephalon(Martinez et al., 1995)]. Genetic deletions and mutations of different IsO mouse genes cause the loss of mes/met derivatives (Chi et al.,2003; Lun and Brand,1998; McMahon and Bradley,1990; Schwarz et al.,1997; Wurst et al.,1994) and demonstrate that the IsO controls proliferation,maintenance and regionalisation of the mes/met domain. Fgf8, a secreted signalling molecule expressed at the IsO, can mimic the inductive effects of IsO grafts, by regulating cell proliferation and controlling gene expression around the isthmus (Crossley et al.,1996). In addition, like the IsO grafts, local misexpression of Fgf8 can induce the development of ectopic or enlarged cerebella(Sato et al., 2001; Sato and Nakamura, 2004; Suzuki-Hirano et al., 2005). However, although transcriptional targets of Fgf signal have been isolated[i.e. feedback-induced antagonists such as Spry4, Pea3, Erm, Sef or Mkp3(Furthauer et al., 2002; Furthauer et al., 2001; Kawakami et al., 2003; Roehl and Nusslein-Volhard,2001)], the cellular events directly requiring Fgf8 signalling are yet to be understood. Fgf8^-/- mouse embryos fail to gastrulate, making it impossible to readily study Fgf8 contribution to the IsO(Meyers et al., 1998; Sun et al., 1999). By contrast, the analysis of a variety of fgf8 conditional inactivations showed the need for Fgf8 in isthmus formation(Meyers et al., 1998). More recently, Fgf8 lack-of-function studies in zebrafish(Jaszai et al., 2003; Reifers et al., 1998) and mice[conditional Fgf8 mutation inside the mes/met domain(Chi et al., 2003)] revealed that the loss of cerebellum in these mutants is preceded by a fate transformation of presumptive r1 cells into otx2-expressing mesencephalic progenitors. In the zebrafish, the lack of gastrulation defect in fgf8-null embryos is explained by the presence of maternal Fgf8 protein, allowing the embryo to gastrulate. The zygotic protein present in ace/fgf8^-/- is fully inactive, and no Fgf activity is detectable in the MHB of ace/fgf8^-/- mutants(Furthauer et al., 2001) (this study).

Thus, a growing body of evidence present the restriction of the otx2 expression domain as one of the early functions of Fgf8 inside the IsO (Martinez et al.,1999; Sato et al.,2001; Sato and Nakamura,2004; Suzuki-Hirano et al.,2005). Molecular studies, using Fgf8 bead implantation rostral to the isthmus, suggest that the formation of an ectopic midbrain is always associated with a very small cerebellum around the Fgf8 source. These tiny cerebella are thought to be formed by local repression of otx2transcripts, creating an artificial boundary of otx2 expression that would allow formation of an IsO able to repolarise the area(Martinez et al., 1999). The local repression of Otx2 is suggested to be necessary for the induction of fgf8 in the host cells surrounding the beads. These fgf8-positive host cells are, in turn, required for the formation of an ectopic cerebellum (Martinez et al.,1999).

We addressed the relationship between Fgf signal and Otx function in the zebrafish. Reduced level of Otx proteins in Otx2^+/-;Otx1^-/- or Otx2^+/-; Otx1^+/-mutant mouse embryos previously unravelled Otx function in positioning the IsO(Acampora et al., 1997; Suda et al., 1997). We obtained hypomorphic Otx lack-of-function (OtxH) embryos using morpholinos against two of the three zebrafish Otx genes. These embryos retain enough Otx function to gastrulate normally but they subsequently show a transformation of the presumptive midbrain territory into an extended r1, resulting in an anterior shift of the isthmus and formation of a bigger cerebellum. In Otx2^+/-; Otx1^-/- mutant mice, the same transformation was attributed to a rostrally expanded fgf8 expression domain (Acampora et al., 1997; Suda et al., 1997). Although such an expansion is also observed in OtxH embryos, we show that zebrafish embryos lacking both Otx and Fgf8 functions maintain a transformation of the midbrain tissue into r1 and retain the ability to differentiate cerebellar cell types. This indicates that both early r1 territory and cerebellar cell types are developing in absence of Fgf8, as long as Otx function is also abrogated. This is not due to persistence or upregulation of other Fgf proteins, as we show that OtxH embryos lacking fgf8 function have no Fgf activity in the presumptive mes/met and that OtxH embryos still develop r1 territory and cerebellar granule cell precursors following complete inhibition of all Fgf signals. Together, our work shows that Fgf8 is not instructive neither in the formation of the r1 territory nor for the differentiation of cerebellar cell identities. Fgf signalling activity is, however, crucial for proliferation, fusion and folding of the cerebellar anlage.

Embryos were staged according to Kimmel et al.(Kimmel et al., 1995). Data presented were obtained from analysis of Kings College London wild-type and of homozygous acerebellar/fgf8^-/- embryos(Reifers et al., 1998)maintained at 28°C on a 14 hour light/10 hour dark cycle.

Whole-mount in situ hybridisation and immunochemistry were performed using standard procedures (Tallafuss and Bally-Cuif, 2003; Westerfield,2000), details available upon request. The Engrailed antibody 4D9 was kindly provided by Simon Hughes.

SU5402 (20 μM, Calbiochem) was added on embryos kept at 31°C from 50% epiboly stage until fixation. Efficiency of the treatment on wild-type embryos is measured by the absence of isthmic constriction similar to the one observed in ace embryos. A typical reduction in the length of the yolk extension and the tail is also seen after SU5402 treatment(Fig. 7E,F) but not in fgf8^-/- embryos, as several other Fgfs are expressed and have redundant functions in tail bud. Loss of spry4 expression after 1 hour of incubation at 31°C was used to confirm treatment efficiency (not shown).

MOs (Gene Tools) were designed against otx1 (GenBank Accession Number BC045290), otx2 (NM_131251) and otx1-like (D26174),diluted in Danieau's media and injected at the one- to four-cell stages. Embryos received 1.2 ng of each MO (0.133 mM, 0.133 pmol/embryo). Morpholinos were as follows: Otx1 MO1-Fluo (complementing bases 417-441),5′-TGAGGTATGACATCATGCTAGAGGC-3′; Otx2 MO-Fluo (complementing bases 261-285), 5′-GTTGCTTGAGATACGACATCATGCT-3′; Otx1-like MO-Fluo(complementing bases 225-249), 5′-GAGGTATGACATCATTGTTGAGCCC-3′;Otx1 MO2-Bare (complementing bases 375-379),5′-GATCTTGCCAATAGTTCGATTTCCC-3′; Otx1MO3-Bare (complementing bases 350-374), 5′-CCGATGTTGCAGTTTGACGAAGGAC-3′.

Most of our morpholinos are labelled with fluorescein to facilitate the selection of equally injected embryos. Under these conditions, we observed the phenotype described here in around 98% of the injected embryos, while we observed a range of milder phenotypes in ∼5% of all injected embryos in the absence of selection. Each experiment was repeated two to six times, with highly similar results.

To control for the specificity and efficiency of our MOs, we cloned all three full-length Otx cDNAs, including 150 bp of their 5′ sequence. As previously reported in Xenopus, overexpression of Otx after RNA injection leads to cell movement defects during gastrulation (otx1RNA 80% affected, n=52; otx2 RNA 88% affected, n=44; otx1-like RNA, 88.2% affected, n=34) (see also Bellipanni et al., 2000). The vast majority of embryos injected with a given Otx RNA and its corresponding MOs showed a rescue [otx1RNA+Otx1-1 MO, 80.6%of wild type (n=31); otx2 RNA+otx2-1 MO, 63.8% of wild type (n=40); otx1-like RNA+ otx1like-1 MO,60.4% of wild type (n=43).

As previously reported, fgf8^-/- mutant larvae are of two different phenotypic classes. Type A larvae (65%, n=78) show a tectum slightly altered in its shape but not affected in its length, whereas type B (35%) larvae are overall retarded most probably as a secondary effect of variably abnormal blood circulation. Some of those embryos (generally 10%)with oedema were removed from our analysis of 5-day-old embryos.

Total extracts were prepared from heads of E10.5 mouse or prim-5 zebrafish(10 embryos), processed for standard western blot assay and probed with Otx antibody (Mallamaci et al.,1996) (1/10000). For detection at bud stage, nuclear extracts were prepared from 50 wild-type or OtxH embryos. Coomassie staining was carried out after transfer to compare proteins levels.

Otx genes, cognates of the Drosophila gap gene Orthodenticle, are highly conserved in all vertebrates. Two closely related family members, Otx1 and Otx2, have been shown to be crucial for early brain development in mammals(Acampora et al., 2003; Acampora et al., 2001). To date, three homologues have been described in zebrafish(Mercier et al., 1995). The third zebrafish gene, previously called otx3 is renamed otx1-like as it has a high percentage of similarity with otx1. otx1 and otx2 are both located on chromosome 17 and otx1-like is on chromosome 1 (Ensembl database). The most likely scenario leading to this zebrafish genomic organisation is that the single ancestral Otx gene found in cephalochordates was first tandemly duplicated(Williams and Holland, 1998). Then the two genes were separated in higher vertebrates while an additional whole genome duplication in teleosts led to four Otx gene copies, with a subsequent loss of an otx2-like gene.

Gene duplication may have been followed by functional diversification of the different paralogues in zebrafish. To define the function of the three paralogues, we used a battery of morpholino antisense oligomers (MOs)(Nasevicius and Ekker, 2000)targeting zebrafish Otx genes. Only the double knock-down of both otx1-like and otx2 led to a severe defect in the embryonic brain. Western blot analyses, using a polyclonal Otx2 antibody recognising vertebrate Otx1 and Otx2, showed that residual Otx proteins were detectable in these embryos at bud stage, but that the proteins were lost by the end of somitogenesis (Fig. 1A). We called these double knock-down embryos OtxH morphant embryos (H standing for`hypomorph'). Brain morphology, first analysed in prim-20 live embryos, showed that the telencephalon, olfactory placodes and hypothalamus are present but the epiphysis is missing in OtxH embryos. The patterning defect is also evident caudally where the isthmic constriction is absent. Only one uniform dorsal structure can be observed (arrow in Fig. 1C) instead of a tectum and cerebellum present in wild type embryos(Fig. 1B). As the single Otx2 and Otx1-like morphants have no visible defect, our data indicate that zebrafish otx1-like and otx2 genes carry a redundant function during early development.

Severe reduction of Otx proteins triggers transformation of the mesencephalic anlage into metencephalic derivatives in mouse. We confirmed, by analyzing mbx and mab21L2, two markers expressed in midbrain and pretectal areas (Kawahara et al.,2002; Kudoh and Dawid,2001), that this area is dramatically reduced as early as the six-somite stage (Fig. 1F,G;data not shown) and absent by the end of somitogenesis(Fig. 1H,I; data not shown) in OtxH embryos. In contrast to this result, genes expressed across the entire mes/met domain, such as engrailed (En), her5 and pax2.1, maintain their expression in OtxH embryos until mid-somitogenesis (data not shown). Among those, En is the only one steadily expressed in the whole mes/met domain during development and is still partially maintained in prim-5 OtxH embryos(Fig. 1J,K).

To study AP patterning caudal to the midbrain, we analysed efnb2a(ephrinb2a) expression, which marks presumptive rhombomere1, 4 and 7(Cooke et al., 2001). From early somite stage, efnb2a r1 domain is reproducibly expanded in OtxH embryos (Fig. 2A,B). In order to assess whether this efnb2a domain is indeed derived from mes/met descendants, we took advantage of the her5pac:egfp line expressing the green fluorescent protein (GFP) under the control of the her5regulatory regions (Tallafuss and Bally-Cuif, 2003). her5 is specifically induced in the entire mes/met domain during late gastrulation (and completely absent in the rest of the hindbrain). As GFP is a very stable protein, it is able to follow temporally the progeny of the mes/met domain. In her5pac:egfpembryos, the metencephalic precursors co-express GFP and efnb2a,while the mesencephalic cells express only GFP(Fig. 2A,C). In OtxH/her5pac:egfp embryos, GFP is always co-expressed with efnb2a (Fig. 2B,D),showing that the OtxH entire presumptive mes/met domain is transformed into an extended metencephalon.

A number of distinct neuronal populations are generated in r1, including the locus coeruleus (LC) (Crossley et al.,1996; Jaszai et al.,2003) and the cerebellar granule cells(Wingate and Hatten, 1999). The LC is a small group of noradrenergic (NA) neurons, known to be induced by coordinated isthmic Fgf and dorsal midline BMP signals(Guo et al., 1999). In agreement with their commitment as NA lineage, LC neurons express two homeobox genes, phox2a and phox2b. In 20-somite stage OtxH embryos,this phox2a-expressing population, located in the dorsal metencephalon, has at least doubled (arrows in Fig. 2E,F). The origin of cerebellar granule neurons has been microsurgically fate-mapped to r1 in chick(Wingate and Hatten, 1999),and specifically to the embryonic rhombic lip, a specialised proliferative epithelium arising at the interface between the neural plate and the roof plate of the IVth ventricle. To further characterise the nature of the expanded r1 area in morphant embryos, we analysed the expression of markers of cerebellar differentiation. We find that the expanded r1 expresses atoh1a and atoh1b, two granule cell markers(Adolf et al., 2004; Koster and Fraser, 2001)(Fig. 2G-J; Fig. 4A,B). Finally, zebrin,marker for cerebellar Purkinje cells also shows an expanded population of these cells in OtxH embryos (see Fig. 5I,J). The OtxH transformed mes/met region is therefore differentiating into a large LC and cerebellum. However, the morphology of the enlarged cerebellum is abnormal, mainly owing to an absence of fusion(Fig. 5F,N; Fig. 6F). Some level of disorganisation is also detected in the Purkinje cell layers(Fig. 5J; see also Fig. S1I in the supplementary material).

Fate transformation of the mes/met area is likely to be triggered by perturbation of the signalling molecules normally expressed in the IsO. Therefore, we analysed the two MHB secreted factors - wnt1 expressed just anterior to the Otx/Gbx boundary and fgf8 just posterior to it(Wurst and Bally-Cuif, 2001) -thereby defining, from early somitogenesis onwards, the midbrain and hindbrain part of the isthmus. In OtxH embryos, an initial residual ventromedial wnt1 expression is rapidly lost (by the six-somite stage, not shown). Expression then reappears at the 20-somite stage in some dorsal cells located anterior to the expanded r1 (likely diencephalic) and within the roof plate of the posterior neural tube (Fig. 3A,B). Conversely, fgf8 is improperly activated in the morphants, from the one-somite stage onwards, in a broad area inside the anterior part of the presumptive mes/met domain, while expression in ventral r2 and r4, and within dorsal forebrain appears unaltered(Fig. 3C-F).

We therefore conclude that, in fish, as in mouse(Acampora et al., 1997; Suda et al., 1997), a decreased level of Otx activity leads to the loss of wnt1 and the expansion of fgf8 signal inside the mes/met area. In mouse, this fgf8 upregulation was proposed as the major cause for the local transformation of the midbrain into r1.

Fgf signal is thought to be necessary and sufficient to instruct r1-derived fates. OtxH embryos provide the unique opportunity to test this hypothesis. Unlike OtxH embryos, the fgf8^-/- mutant [acerebellar,ace (Reifers et al.,1998)] lacks cerebellum because of a progressive transformation of the anterior r1 into mesencephalon (Jaszai et al., 2003). Thus, embryos with lowered level of Otx proteins in an fgf8^-/-/ace background are expected to lack both msencephalon and r1. Zebrafish fgf8^-/-embryos were injected with the combination of otx1-like and otx2morpholinos (referred to as fgf8^-/-; OtxH embryos). As previously reported, atoh1a is consistently missing from the upper rhombic lip in fgf8^-/- mutant embryos(Jaszai et al., 2003) and atoh1a-expressing cells from the lower rhombic lip are not able to contribute to the cerebellar system(Koster and Fraser, 2001). To our surprise, in fgf8^-/-; OtxH embryos, an upper rhombic lip cell population is restored, resembling the one seen in OtxH embryos(Fig. 4B,D), albeit reduced in size.

To test whether cells from the presumptive posterior rhombic lip may possibly migrate anteriorly to regenerate a cerebellar granule cell population in fgf8^-/-; OtxH embryos, we used the stability of GFP protein to follow the origin of the upper rhombic lip cells in fgf8^-/-; her5pac:egfp; OtxH embryos. We consistently observe co-localisation of atoh1a and GFP on parasagittal cryostat sections of fgf8^-/-; her5pac:egfp; OtxH prim5 brains, ruling out the possibility of cell migration from posterior GFP-negative rhombomeres (Fig. 4N,O). In her5pac:egfp wild-type embryos, a subpopulation of GFP-positive cells is co-expressing atoh1a, confirming that, as in higher vertebrates, zebrafish granule cells are born within an r1 territory characterised by her5 expression. This atoh1a and GFP co-expressing population is absent in fgf8^-/- embryos(Fig. 4K,L) and is present in fgf8^-/-; OtxH (Fig. 4N,O), as well as in OtxH embryos(Fig. 4H,I).

Differentiation of cerebellar cells is further assessed using pax6and reelin staining granule cells (Fig. 5A-H) (Costagli et al.,2002) and Zebrin/AldolaseC marking Purkinje cells(Fig. 5I-L; see Fig. S1 in the supplementary material) (Lannoo et al.,1991; Miyamura and Nakayasu,2001) in 5-day-old embryos. These two cell populations are expanded in OtxH embryos (Fig. 5B, n=30 embryos; Fig. 5F, n=16; Fig. 5J, n=36), and are absent in fgf8^-/-(except the rare zebrin-positive cells in 10% of mutants, Fig. 5C, n=12; Fig. 5G, n=2, Fig. 5K, n=8). All pax6, reelin and zebrin-positive cell populations are rescued in fgf8^-/-; OtxH embryos(Fig. 5D, n=10; Fig. 5H, n=4; Fig. 5L, n=12).

Thus, upon reduction of Otx function, Fgf8 signalling is not required for the differentiation of the cerebellar granule and Purkinje cells. However, the rescued cerebellar region is significantly smaller than wild type. Cell death and proliferation were therefore carefully analysed(Fig. 6). Apoptosis, revealed by Acridine Orange staining was enhanced in double-mutant embryos from the 20-somite stage onwards, while no significant difference was observed in ace or OtxH embryos (Fig. 6A-D). Conversely, proliferation rate is decreased in prim-22 ace mutants and this diminution is not worsened by lack of Otx proteins (Fig. 6E-H). Thus,although Fgf activity is not required for cerebellar cell type specification,it is necessary for cell proliferation. Moreover, Otx and Fgf8 are both required for cell survival.

Finally, we analysed development of the r1-derived LC neurons. In fgf8^-/-; OtxH embryos [analysed at both the 15-somite stage (n=85) and prim-5 stage (n=52); data not shown], phox2a expression is never found, showing that LC fate requires fgf8, whether or not Otx function is lost.

The presence of cerebellar differentiation in absence of Fgf8 in our OtxH morphants may mean that Fgf activity is required to maintain Otx expression away from r1 territory but is not necessary for any other step in cerebellar cell fate determination. However, a more trivial possibility is that some Fgf activity, carried out by other Fgfs in the MHB, may rescue cerebellar differentiation in the fgf8^-/-; OtxH embryos. To address this possibility, we measured Fgf activity in the MHB region. As the best indicator of Fgf gene activity is the expression of transcriptional targets,three of them were analysed in the four genetic contexts studied. As illustrated in Fig. 6I-P (and in Fig. S2 in the supplementary material), erm, spry4 and pea3 are highly expressed in wild-type MHB from bud stage onwards,but are never induced there in fgf8^-/- and fgf8^-/-; OtxH embryos, indicating a complete absence of Fgf activity in this region. Finally, treating batches of fgf8^-/- embryos with the Fgfr inhibitor SU5402(Mohammadi et al., 1997) does not worsen the fgf8^-/- midbrain phenotype (data not shown), further supporting a complete absence of Fgf signalling in the MHB of fgf8^-/- mutants. The differentiation of cerebellar cell fates in our double loss-of-function embryos therefore occurs in absence of any Fgf activity in the MHB from at least bud stage onwards.

To assess whether midbrain to r1 transformation can be observed in OtxH embryos totally depleted from earlier Fgf activity, we analysed the development of the mes/met area in OtxH embryos treated with SU5402 from the onset of gastrulation. Wild-type embryos treated with SU5402 present a marked reduction of efnb2a in presumptive r1, while, in SU treated OtxH morphant embryos, the anteroposterior extent of efnb2a-expressing r1 is similar to the one observed in untreated OtxH embryos(Fig. 7A-D). Early midbrain to r1 transformation is therefore taking place in the absence of Fgf activity in OtxH embryos. We then analysed atoh1a expression in the upper rhombic lip in wild-type and OtxH embryos, with or without SU5402 respectively. Results are summarized in Fig. S1L-O (see supplementary material). Inhibition of FgfR in wild-type embryos leads to the absence of prospective granule cells. By contrast, OtxH embryos subjected to the same treatment show formation of atoh1a-expressing granule cells. This experiment confirms that decreased level of Otx proteins allows for the development of cerebellar granule cell precursors in the complete absence of Fgf activity.

In higher vertebrates, two major transcription factors establish the presumptive mesencephalon and metencephalon [respectively, Otx2(Martinez-Barbera et al.,2001) and Gbx2 (Wassarman et al., 1997)]. In fgf8^-/- embryos, metencephalon precursors erroneously express otx2, indicating a posterior shift of mesencephalon identity (Jaszai et al.,2003; Tallafuss and Bally-Cuif, 2003), which we found can be revealed as early as the eight-somite stage (Fig. 7E,F). Complementarily, rostral expansion of r1 territory in OtxH embryos also starts at early somitogenesis, which may possibly be induced by rostral expansion of Gbx2 expression. Testing this possibility, we found that formation of cerebellar tissue, in fgf8^-/-; OtxH embryos, occurs independently of presence of gbx2 (arrows in Fig. 7G-J). Zebrafish gbx1 and gbx2 are sequentially expressed in r1; and it has been suggested that gbx1 is the functional homologue of mouse Gbx2(Rhinn et al., 2003). Zebrafish gbx1 expression is switched off in the wild-type mes/met area at a stage preceding the onset of mispatterning in OtxH embryos and is not altered in these embryos, precluding its involvement in the process studied here. Conversely, gbx2 expression is expanded anteriorly in OtxH embryos, as expected following the anterior shift of the r1 territory. As previously reported in fish and mouse (Chi et al., 2003; Rhinn et al.,2003), we confirm that maintenance of gbx2 expression is dependent upon Fgf activity. However, we observed that loss of Otx function in fgf8^-/-; OtxH embryos is unable to rescue gbx2expression (Fig. 7J). As these embryos develop cerebellar fates, this implies that from the six-somite stage onwards, gbx2 is not required for cerebellum cell differentiation in the absence of Otx function. Supporting this, a mouse conditional knockout showed that cerebellum still develops after gbx2 inactivation at 9.5 dpc (Li et al., 2002).

We show that, in zebrafish, double lack of otx1-like and otx2 function leads to the transformation of the midbrain territory into an expanded presumptive r1, characterized by an enlarged cerebellum,highly reminiscent of the phenotype of Otx2^+/-;Otx1^-/- mouse embryos(Acampora et al., 1997; Suda et al., 1997). It is worth noting that OtxH zebrafish embryos, like Otx2^+/-;Otx1^-/- mice, present a transformation of retinal pigment epithelium into an enlarged retina (Fig. 2C and data not shown)(Martinez-Morales et al.,2001). This confirms a conserved role of Otx homeobox proteins across different species.

Taking advantage of these results, we analysed, for the first time, the mes-metencephalic phenotype of embryos lacking both Otx and Fgf functions in the MHB. We show that these embryos develop r1 identity and differentiate several cerebellar cell fates but lack the locus coeruleus. Thus, early r1 territory and cerebellar cell identities are induced in absence of Fgf signalling as long as Otx function is also removed. Our results therefore challenge the current model in which Fgf activity from the IsO is providing the information required for induction of cerebellar cell identities. If Fgf signalling is not instructive for cerebellar identity; it is, however,required to maintain the r1 compartment free of Otx protein, stabilizing compartment boundaries; it is also required not only for the proliferation of cerebellar precursor cells but also for cerebellar dorsoventral organisation and dorsal fusion.

Fgf8 gain- and loss-of-function (Sato et al., 2001; Sato and Nakamura, 2004; Suzuki-Hirano et al., 2005; Trokovic et al.,2003) have led to a prevalent model in which Fgf8 signalling instructs cerebellar development. More recently, studies in fish, chick and ascidians propose a primary role for Fgf8 in repressing Otx expression, in vertebrates Otx2 in turn represses r1 and cerebellar identity(Hudson et al., 2003; Jaszai et al., 2003; Martinez et al., 1999; Sato et al., 2001; Sato and Nakamura, 2004; Suzuki-Hirano et al., 2005). A mouse Fgf8 conditional null mutation producing a truncated protein similar to the one present in zebrafish ace has been reported(Chi et al., 2003). This shortened protein is also completely inactive in mouse, and embryos that carry this conditional knockout allele (which removes Fgf8 function inside the En1-expressing territory) show, like their zebrafish counterparts, a posterior shift of otx2 expression at the 12-somite stage(Chi et al., 2003; Jaszai et al., 2003).

Here, we report that Fgf signalling is dispensable for cerebellar cell differentiation in the absence of Otx function. As we argue above, Otx2 is never expressed in presumptive r1. In chick and mouse, the posterior boundary of otx2 expression defines the caudal limit of the presumptive midbrain and, hence, the anterior limit of the cerebellar anlage(Millet et al., 1996; Zervas et al., 2004). In fish too, three arguments strongly suggest that otx2 is never expressed in prospective r1 territory. First, the posterior limit of otx2expression at late gastrulation is always anterior to that of her5,which maps to the posterior part of r1(Tallafuss and Bally-Cuif,2003; Tallafuss et al.,2001). Second, the gap between the posterior boundary of otx2 expression and the anterior limit of the hoxb1b domain(earliest marker defining r3/r4 boundary)(Prince et al., 1998) is constant between 60% epiboly and the end of gastrulation (see Fig. S1 in the supplementary material). And third, gbx1 expression only overlaps with otx2 at 60% epiboly, well before both the sharpening of the r1 boundary and the onset of fgf8 expression in r4(Reifers et al., 1998; Rhinn et al., 2003). Finally,the onset of fgf8 expression in r1 is taking place during late gastrulation, long after the formation of the Otx expression domain in fish(Maves et al., 2002; Reifers et al., 1998; Rhinn et al., 2003) and mouse(Martinez-Barbera et al.,2001; Garda et al.,2001). All together, the existing data all strongly suggest that Fgf8 has no role in defining otx2 posterior boundary initially but maintains it during somitogenesis. Hence, we propose that the function of Fgf signalling from the IsO is neither to induce r1 nor cerebellar cell identities but to prevent a later propagation of Otx expression inside the r1 territory,thereby maintaining a set of developmental decisions required to maintain r1,and preventing repression of cerebellar identities by Otx.

If it is not Fgf, which is the key molecule responsible for the switch to a cerebellar fate? Targeted mutations of either otx2 or hoxa2resulted in either a rostral or a caudal expansion of the cerebellum,respectively (Acampora et al.,1997; Gavalas et al.,1997). Moreover, hoxa2 mis-expression experiments in r1 showed cell-autonomous inhibition of granule cell fate(Eddison et al., 2004). Therefore, in Fgf misexpression experiments, the formation of an ectopic cerebellum would rather occur by local downregulation of the cerebellum repressors otx2 and/or hoxa2(Irving and Mason, 1999; Martinez et al., 1999), rather than by direct induction. A member of the iroquois homeodomain gene family, irx2 has been recently suggested as a specific inducer of chick cerebellar identity downstream of Fgf8 signal(Matsumoto et al., 2004). Exclusively in presence of Fgf activity, activated phosphorylated Irx2 forces the rostral hindbrain towards cerebellar fate. Although not discussed by the authors, the possibility of irx2 attributes cerebellar identity independently from any Fgf activity cannot be excluded. Indeed, iro1band iro2a are both expressed in r1 during early to mid-somitogenesis(Lecaudey et al., 2005). In our four conditions, both genes are still expressed in r1 of fgf8^-/-; OtxH embryos, showing that iro1b and iro2a are responding like other r1 markers studied (see Fig. S3 in the supplementary material). However, presence of these genes transcripts in both the mesencephalic and metencephalic compartment confirms that any r1-specific function of Iroquois should be studied at the protein level.

There are great variations in cerebellum morphology among teleosts and the zebrafish cerebellum (mainly lateral and vestibular) is one of the simplest. The corpus cerebelli and valvula cerebelli are thought to be related to the mouse vermis and flocculi, respectively; and the corpus cerebelli, which received inputs from the vestibular apparatus, is believed to be the most primitive part of the cerebellum(Wullimann et al., 1996).

In our experiments, lack of Otx function affects the dorsal midline fusion that normally occurs during the early stage of cerebellar development. Based on an unexpected increase in isthmic structure in En1^+/Otx2LacZ mouse embryos, it has been suggested that otx2-positive cells are crossing the otx2 boundary to form a substratum region (called velum) required for midline fusion(Louvi et al., 2003). Even if,in zebrafish, the valvula cerebelli forms directly adjacent to the mesencephalon - without the presence of a recognizable velum - our results are compatible with a function of Otx in cerebellar dorsal fusion. Indeed, the absence of fusion in OtxH embryos suggests that Otx-expressing cells may be involved in midline fusion in fish.

This absence of fusion is exacerbated in absence of Fgf signalling. Phenotypic analysis of fgf8^-/-; OtxH zebrafish embryos shed some light on the Fgf function in cerebellar morphogenesis. fgf8^-/-; OtxH cerebellar tissue undergoes a severe lateral displacement. In particular, the granule cells of the eminentia granularis normally spread dorsoventrally in wild-type and OtxH embryos(Fig. 5A,B,E,F) but are only seen in the ventral area in fgf8^-/-; OtxH embryos(arrowheads in Fig. 5D,H). Finally, in fgf8^-/-; OtxH, the Purkinje cells loose their layered organisation and the ventral axonal projections are much reduced. Such projection has not yet been described in zebrafish and the only ventral projection described in adult teleost is thought to come from a specific subpopulation of caudal lobe Purkinje cells(Lannoo et al., 1991). Absence or disturbed axonal projections in the fgf8^-/-; OtxH cerebellum could therefore be due to either a defect in pathfinding, a delay in neuronal maturation or loss of a specific subpopulation of the Purkinje cell population. We therefore conclude that Otx activity is primarily required for cerebellar fusion, while Fgf8 activity is necessary for cerebellum folding and cellular organisation (but not differentiation) of the cerebellum.

The mechanisms that underlie cerebellar dorsoventral organisation and late cerebellar morphogenesis are also starting to be uncovered in mouse. Recent mouse studies reveal the importance of Hedgehog signalling for the cerebellar morphogenesis and its dorsoventral organisation(Sotelo, 2004; Wechsler-Reya and Scott,1999). Our next challenge is therefore to understand how Fgf and Hh signals coordinate these events and what the downstream effectors are that they regulate in these processes.

Finally, our findings confirm the central role of Fgf in proliferation. In mouse, mis-expression of fgf8 under the wnt1 promoter revealed its mitogenic activity on mesencephalic cells(Lee et al., 1997) and partial fgf8 lack of function reduces proliferation of the medial cerebellar anlage (Xu et al., 2000). This effect on proliferation is detected from prim-22 in fgf8^-/- embryos, and reducing Otx levels in this context does not affect the proliferation rate, indicating that Otx is not involved in this process.

Conversely, cell death is not more frequent in absence of Fgf function,although the dying cells are more often found at the roof plate of the enlarged midbrain. By contrast, cell death is increased in OtxH mes/met area starting at 20-somite stage, most probably owing to the lack of Wnt1 shown to be crucial for cell survival. This phenotype is exacerbated in absence of Fgf,probably revealing cooperation between these two signals. This result explains the reduced size (52% the size of OtxH; Fig. 4B,D) of the rescued granule cells population in fgf8^-/-; OtxH. At earlier stages, the overall size of the Her5-GFP territory is not dramatically different in double mutant embryos, neither is the expanded efnb2aterritory in the SU-treated morphants, showing that there is no significant apoptosis during the first half of somitogenesis. Together, reduced proliferation and increased cell death explain the relatively variable number of rescued differentiated cells observed at 5 dpf in fgf8^-/-; OtxH.

Our study supports previous studies(Guo et al., 1999; Lam, 2003)showing that the IsO Fgf signalling is strictly required for the induction and/or early survival of LC tyrosin hydroxylase-positive precursors. Our data also indicate that loss of Otx function is not able to rescue LC neurons in fgf8^-/-mutant. This result suggests that, together with Bmp, Fgf is directly required to induce LC identity.

We are grateful to R. Wingate, for insightful discussion and helpful comments, to S. W. Wilson for providing probes, and to P. Herbomel and C. Petit for their support on the revised manuscript. We thank all members of the laboratory for useful comments on the manuscript, Amy Cording for excellent fish and technical assistance, and Alexandra Tallafuss for providing ace,her5pac:egfp fish. This work was supported by research grants to C.H. from the MRC and the Wellcome Trust, and by a Junior Group grant from the VolkswagenStiftung to L.B.C.