Temporal and spatial gradients of Fgf8 and Fgf17 regulate proliferation and differentiation of midline cerebellar structures

Patterning of the midbrain-hindbrain (MHB) anlage depends on an ‘organizer’ activity localized at the MHB junction (Alvarez Otero et al., 1993; Hallonet and Le Douarin, 1993; Hallonet and Alvaradomallart, 1997). Classical transplantation experiments demonstrate that MHB tissue can induce midbrain structures when transplanted into the diencephalon and cerebellar tissue when transplanted to the myelencephalon (Gardner and Barald, 1991; Itasaki et al., 1991; Martinez et al., 1991; Bally-Cuif et al., 1992; Marin and Puelles, 1995). Part of the MHB organizer activity has been attributed to FGF8, which is expressed at the MHB junction and can induce ectopic midbrain and cerebellar tissue formation when applied in the diencephalon or midbrain of chick embryos at early somite stages or when ectopically expressed in the midbrain of the mouse (Crossley et al., 1996; Liu et al., 1999; Martinez et al., 1999). This hypothesis has been further supported by an acerebellar mutant in zebrafish caused by the loss of Fgf8 function (Reifers et al., 1998) and by the loss of midbrain and cerebellar tissue in a mouse carrying a severe hypomorphic allele of Fgf8 (Meyers et al., 1998).

Several families of genes are involved in patterning MHB tissue. Wnt1, Engrailed2 (En2) and Pax5 are essential for specifying midbrain and hindbrain fate before embryonic day 9.5 (E9.5). Wnt1 null mice, Pax2/5 double knockout (KO) mice and Engrailed 1/2 double KO mice all have lost both the mesencephalic vesicle and the cerebellar primordium by embryonic day 10 or earlier (McMahon et al., 1992; Joyner, 1996; Schwarz et al., 1999). Understanding the interaction among these families of genes and their relationship with FGF signaling pathways is essential to delineate the complex genetic program that regulates MHB development. During development, implantation of FGF8 beads into the diencephalon can induce ectopic expression of Fgf8, Wnt1, En1, En2, Pax2 and Pax5 (Crossley et al., 1996; Sheikh and Mason, 1996; Liu et al., 1999; Shamim et al., 1999). These data suggest that FGF8 may mediate the organizer activity of the MHB junction through the activation and/or maintenance of Wnt1, En2, Pax2, Pax5 and itself. Recently, FGF8 has been shown to suppress Otx2 expression at an early somite stage of MHB development (Liu et al., 1999; Martinez et al., 1999), suggesting that FGF8 may also be important for enforcing the midbrain-cerebellar boundary.

In addition to the patterning activity of FGF8, FGF signaling may also regulate the growth and polarity of the midbrain primordium. Overexpression of an Fgf8 transgene, targeted specifically to the midbrain and MHB junction under the control of a Wnt1 enhancer (WEXPZ2-Fgf8), increased the proliferation of midbrain precursor cells and shifted the high proliferation zone at the caudal part of the midbrain primordium more rostrally (Lee et al., 1997). At the molecular level, this transgene promoted the expression of ELF-1 and En2 at the caudal end of the midbrain and shifted their gradients of expression more rostrally. However, unlike the relatively broad expression domain of the WEXPZ2-Fgf8 transgene, endogenous Fgf8 expression is localized primarily in the cerebellar anlage. Thus, although FGF8 may have a significant paracrine effect on the development of adjacent midbrain structures, the role of FGF8 in regulating cerebellar growth and the polarity of cerebellar development remains to be addressed.

FGFs are a large family of at least 19 growth factors sharing 13-71% amino acid identity (pairwise comparisons of mouse or rat FGFs). FGFs are organized into six subfamilies of higher sequence similarity (Coulier et al., 1997; Xu et al., 1999) and members in the same subfamily often share similar biochemical properties. FGFs can activate one of four high-affinity FGF receptor tyrosine kinases, which mediate diverse biological events including ductal branching, cell proliferation, differentiation and migration (Hogan and Yingling, 1998; McKeehan et al., 1998; Naski and Ornitz, 1998; Szebenyi and Fallon, 1999).

Recent expression studies have identified a family of three related Fgfs (Fgf8, Fgf17 and Fgf18) in the MHB junction (Maruoka et al., 1998; Xu et al., 1999). This observation raised the question of whether all three FGFs are equivalent in function or whether they have diverged. Divergence could result from altered biochemical properties, such as receptor specificity, from altered temporal and spatial expression patterns or from an altered repertoire of downstream target genes at different stages of development. In this study, we demonstrate that all three Fgfs have similar biochemical properties in vitro yet have distinct spatial and temporal patterns of expression in vivo. By disrupting alleles of Fgf8 and Fgf17 in mice, we show that these genes have adopted unique yet complementary roles to regulate the size of the medial cerebellar (vermis) anlage and consequently the mature vermis cerebellum. Cellular and genetic evidence supports a functional cooperation between FGF8 and FGF17 and demonstrates that, after E11, these molecules no longer act as an organizer signal but function to regulate cell proliferation and the transition between proliferation and differentiation in the vermis anlage.

A 6.7 kb BamHI-EcoRI genomic clone containing exon 1a, 1b, 1c and 2 was used to construct the targeting vector. Exons 1a and 1b were deleted and replaced with a SalI site and a SacII site, between which a cre gene with human β-actin gene polyadenylation sequence (Lewandoski et al., 1997) and a PGK/neo cassette were inserted (Tybulewicz et al., 1991). The targeting construct was then finished by placing this BamHI and EcoRI fragment into the pTK1TK2 vector (Chisaka and Capecchi, 1991). SM1 ES cells were transfected with the targeting vector and selected with G418 and FIAU. Antibiotic-resistant clones were screened for homologous recombination using both 5′ and 3′ probes as indicated (Fig. 1).

To generate Fgf17^−/−, Fgf8^Δ2,3n/+ mice, Fgf8/17 double heterozygous mice were first obtained by breeding Fgf17⁻/⁻ mice with Fgf8^Δ2,3n/+ mice (Meyers et al., 1998). Double heterozygous mice were then mated to Fgf17⁻/⁻ mice to obtain mutants that are heterozygous for the Fgf8^Δ2,3n/+ allele in an Fgf17⁻/⁻ background. The Fgf8^Δ2,3n/+ allele was identified by Southern blot as previously described (Meyers et al., 1998). All mice were maintained in a 129/sv and C57/B6 mixed genetic background.

Anti-BrdU immunohistochemistry was carried out as previously described (Naski et al., 1998) with minor modifications. Embryos were labeled by IP injection of pregnant females with BrdU (Sigma Inc.) at a dosage of 100 mg/kg. Embryos were fixed in 4% paraformaldehyde overnight at 4°C and embedded in paraffin. After deparaffinization, rehydration and trypsin digestion, 4 μm sections were incubated with an anti-BrdU monoclonal antibody (Becton-Dickson Inc.) overnight at 4°C, followed by peroxidase immunohistochemistry using the ABC kit and the DAB substrate kit (Vector Lab, Inc). Hematoxylin and eosin staining were carried out by our histology core facility. In all cases, serial sagittal sections covering the medial portion of the cerebellum were collected. Midline sections were identified as those with the thinnest cerebellar neuroepithelium and shortest dorsal-ventral dimension in the cerebellar primordium.

For histological analysis, brains from postnatal day 2 and 2-month-old adult mice were fixed in 4% paraformaldehyde overnight at 4°C and embedded in paraffin. 8 μm sagittal sections were collected and stained with Hematoxylin and eosin.

Calbindin expression was detected with a rabbit polyclonal antibody (Chemicon International, Inc.) and visualized by tyramide signal amplification (NEN life science products).

In situ hybridization using ³⁵S-UTP-labeled probes was carried out as previously described (Xu et al., 1999). The plasmids for generating the Wnt1, En2 and Pax5 probes were kindly provided by A. P. McMahon, A. L. Joyner and P. Gruss, respectively. Mouse EST AA199520 (Washington University mouse EST project) was used as the template to generate an Otx2 probe. The template for the Fgf18 probe, consisting 310 bp of 3′ coding sequence and 360 bp of 3′UTR, was subcloned from a cDNA isolated from a mouse E14.5 embryonic cDNA library (Novagen, Cat. No. 69636-1).

To study the in vivo function of the Fgf17 gene, an Fgf17 mutant allele (Fgf17^neo) was generated through homologous recombination in embryonic stem cells (Fig. 1A,B). In the mutant allele, the translation initiation site and two exons encoding the signal peptide were deleted. Embryonic MHB tissue from mice homozygous for this allele expressed no detectable Fgf17 mRNA and heterozygous mice had reduced expression of Fgf17 (Fig. 1C). Therefore, the mutant allele is expected to be a null allele. Homozygous Fgf17^neo(Fgf17^−/−) mice were born alive with the expected Mendelian frequency. They have no gross anatomical abnormalities and appeared healthy. Histological examination of postnatal day 2 (P2) and adult Fgf17^−/− mice revealed a significant tissue loss in the most caudal part of the midbrain, the inferior colliculus and the vermis cerebellum (Fig. 2). Other regions of the CNS appeared normal (data not shown). In the mouse, the most rostral lobe of the vermis is numbered as lobe III (for consistency with other organisms such as rat). In Fgf17^−/− mice, lobe III was roughly one third of normal size. The fissure that separates lobe III and IV failed to form completely, leaving these two lobes partially fused (arrows in Fig. 2B,D). In contrast, the hemispheres of the cerebellum appeared normal in Fgf17^−/− mice (Fig. 2E,F). In addition, the overall size of the mutant vermis cerebellum is about 82% that of control mice (based on measurement of the area of histological sections). Fgf17 is thus an essential regulator of vermis cerebellar size.

Although the MHB tissue loss observed in Fgf17^−/− mice is readily apparent in all mice (n=5) examined, the phenotype is milder than that observed in a previously reported mouse that is severely hypomorphic for Fgf8 (Fgf8^Δ2,3/Fgf8^neo) (Meyers et al., 1998). The difference in phenotype may result from spatial and temporal differences in gene expression of Fgf8, Fgf17 and Fgf18, from different levels of expression or from different biochemical properties of these FGFs. Previous studies have shown that Fgf8, Fgf17 and Fgf18 are all expressed in the MHB junction between E9.5 and E10.5 (Maruoka et al., 1998; Xu et al., 1999). However, existing data did not provide a comparative analysis of all three Fgfs at stages prior to E9.5 or later than E12.5. To investigate the apparent differences in the function of Fgf8, Fgf17 and potentially Fgf18, the spatial and temporal expression patterns of these genes were examined between E8.5 and E14.5. At E8.5 (8-9 somite stage), all three Fgfs became detectable in the MHB junction and examination of serial sections showed that Fgf8 expression clearly covered a broader region than either Fgf17 or Fgf18 (Fig. 3Ba-l). By E10.5, all three Fgfs were expressed with similar intensity, with Fgf17 covering the broadest domain along both the mediolateral (data not shown) and rostrocaudal planes (Fig. 3Bm-q). By E12.5, Fgf8 and Fgf18 expression intensity became weak, while Fgf17 expression intensity remained strong (Fig. 3Br-v). By E13.5 (data not shown) and E14.5 (Fig. 3Bw-aa), Fgf8 and Fgf18 expression were no longer detectable at the MHB junction while Fgf17 expression persisted, although with a narrowed expression domain. This analysis and comparison of the phenotypes of Fgf17^−/− mice and Fgf8^Δ2,3/Fgf8^neo mice suggested that Fgf17 may have a predominant role at later stages of MHB development and that Fgf8 is more important at earlier stages of MHB development (Fig. 3C).

To compare the biochemical properties of FGF8, FGF17 and FGF18 their receptor specificity and biological activity was examined. Human recombinant FGF1, FGF8, FGF17 and FGF18 were assayed for their ability to stimulate DNA synthesis in receptor-expressing BaF3 cells. Consistent with previously published data (Ornitz et al., 1996), FGF1 induced strong activity in all seven cell lines expressing FGFR1b, FGFR1c, FGFR2b, FGFR2c, FGFR3b, FGFR3c or FGFR4. To assess the relative activity of FGF8, FGF17 and FGF18, FGF1 activity was normalized to 100%, and the activity of FGF8, FGF17 and FGF18 at the same concentration was calculated relative to that of FGF1 (Table 1). This analysis showed that FGF8, FGF17 and FGF18 had strong activity towards FGFR3c and FGFR4c, moderate activity towards FGFR2c, minimal activity towards FGFR1c and no detectable activity towards any b isoform receptor. Together these data indicate that the in vitro receptor specificity and activity for FGF8, FGF17 and FGF18 are very similar; although, accessory molecules still could differentially modify relative activity in vivo. Nevertheless, this analysis suggests that phenotypic differences between Fgf8^Δ2,3n/Fgf8^neo mice and Fgf17^−/− mice are due to differences in their spatial and temporal expression patterns in the MHB junction.

Because Fgf8 and Fgf17 overlap in their pattern of expression between E8.5 and E12.5, it was important to address the question of whether alleles of Fgf8 and Fgf17 have equivalent or redundant function. Mice containing null alleles of both genes (Fgf17^−/+, Fgf8^Δ2,3n/+) were generated by mating Fgf17^−/− mice with mice carrying an Fgf8 null allele, Fgf8^Δ2,3n/+, (Meyers et al., 1998). If an allele of Fgf8 could substitute for an allele of Fgf17 then the prediction would be that double heterozygous mice would have a similar phenotype to Fgf17^−/− mice. Histological analysis did not identify any tissue loss in the MHB of P2 or adult double heterozygous mice (compare Fig. 4A and C to Fig. 2A and C). This demonstrates that one allele of Fgf17 is not equivalent to one allele of Fgf8, and that Fgf17 has a unique role during MHB development.

By crossing double heterozygous mice with Fgf17^−/− mice, a compound mutant that is heterozygous-null for Fgf8 in an Fgf17-null background (Fgf17^−/−, Fgf8^Δ2,3n/+) was generated. Histological analysis showed that at both neonatal (P2) and adult stages, vermis lobe III is completely lost in all Fgf17^−/−, Fgf8^Δ2,3n/+ mice analyzed (n=4) (Fig. 4B,D) and that there is a further progressive decrease in the overall size of the vermis cerebellum compared to that of wild-type mice (∼76%). Tissue loss in the inferior colliculus is similar to that in Fgf17^−/− mice (compare Figs 2D and 4D). Therefore, the loss of an additional allele of Fgf8 can significantly increase the cerebellar tissue loss observed in Fgf17^−/− mice demonstrating that these genes do have some redundant function. Interestingly, like in Fgf17^−/− mice, the cerebellar phenotype in Fgf17^−/−, Fgf8^Δ2,3n/+ mice is limited to the midline vermis and the hemispheres of the cerebellum are unaffected (Fig. 4E,F).

Early in MHB development Fgf8 can induce and maintain expression patterns of genes that specify the identity of the MHB region, a property often referred to as organizer activity. Specifically, FGF8 can induce and maintain the expression of Wnt1, En1, En2, Pax2 and Pax5 at the 8-to 12-somite stage of development (E8.5-9.5 in mice) (Crossley et al., 1996; Liu et al., 1999; Shamim et al., 1999). To test whether partial loss of FGF signals between E8.5 and E12.5 affect patterning of the vermis anlage, the expression of Wnt1, En2, Pax5 and Otx2 were compared at E12.5 in wild-type and Fgf17^−/−, Fgf8^Δ2,3n/+ mice. In tissue from Fgf17^−/− , Fgf8Δ2,3n /+ mice, there was no significant change in either the spatial pattern or the level of expression of Wnt1, En2 or Pax5. Otx2, which is negatively regulated by FGF8 at early somite stages, also did not change in MHB tissue from Fgf17^−/−, Fgf8^Δ2,3n/+ mice at this stage (Fig. 5). RNA blot analysis of dissected midbrain and anterior hindbrain tissue showed no quantitative changes in the expression of these genes (data not shown). These data demonstrate that loss of FGF8 and FGF17 activity prior to E12.5 does not cumulatively affect the patterning of the MHB region at E12.5.

The extent of MHB tissue loss observed at postnatal day 2 (P2) was similar to that of adult Fgf17^−/− mice and Fgf17^−/−, Fgf8^Δ2,3n/+ mice suggesting an early developmental origin of the phenotype. To examine whether a defect in cell proliferation or cell death could account for the observed tissue loss, BrdU-labeling indices were examined at embryonic ages E10.5 to E12.5 and TUNEL assays were examined at E12.5 (Fig. 6 and data not shown). Anti-BrdU immunohistochemistry was focused on midline sagittal sections (see methods for details) because it has been documented that the primordium for the anterior vermis is located at the most medial part of the cerebellar plate (Altman and Bayer, 1997). The number of BrdU-positive and BrdU-negative cells in the entire midline primordium were quantified at E10.5, E11.5 and E12.5. At E10.5, no significant difference in cell proliferation indices or cellular morphology of the cerebellar neuroepithelium was detected (Fig. 6Aa-c) and in all genotypes the proliferation index was about 42% (Fig. 6B). By E11.5, the proliferation index of both wild-type and Fgf17^−/− tissue was 40.8±0.2 % (Fig. 6Ad-e,B); however, the proliferation rate in tissue from Fgf17^−/−, Fgf8^Δ2,3n/+ embryos decreased to 31.7±0.3% (Fig. 6Af,B), significantly lower than in tissue from either wild-type or Fgf17^−/− mice. Consistent with the proliferation data, the cerebellar neuroepithelium in Fgf17^−/−, Fgf8^Δ2,3n/+ mice was thinner than in either Fgf17^−/− mice or wild-type mice (Fig. 6Ad-f). By E12.5, the proliferation rate in the wild-type vermis decreased to 30.1±1.5%, in Fgf17^−/− tissue it decreased further to 18.8±1.7%, and in tissue from Fgf17^−/−, Fgf8^Δ2,3n/+ embryos the proliferation index was only 14.1±2.2% (Fig. 6Ag-I,B). These data indicate a progressive decrease in cell proliferation in the Fgf17^−/− and Fgf17^−/−, Fgf8^Δ2,3n/+ mice compared to wild-type mice. A consequence of the decreased proliferation index is that by E12.5 the midline cerebellar neuroepithelium in Fgf17^−/−, Fgf8^Δ2,3n/+ mice and to a lesser extent in Fgf17^−/− mice was thinner than that of wild-type mice (Fig. 6Ag-i). Significantly, the decrease in proliferation was only observed at stages when the expressions of all three Fgfs were declining, suggesting a more significant redundancy at stages prior to E11.5. At E12.5, TUNEL analysis showed that cell death occurred in less than 1% of neuroepithelial cells in the vermis anlage in both wild-type and Fgf17^−/−, Fgf8^Δ2,3n/+ mice (data not shown). In summary, these data demonstrate that, after E11, FGF8, FGF17 and possibly FGF18 act as growth factors that maintain proliferation but not cell survival of precursor cells in the midline of the cerebellar primordium.

Purkinje cells in the mouse cerebellar primordium stop proliferation and begin to differentiate at E13 to E14 (for review see Hatten and Heintz, 1995), coincident with the termination of Fgf8 and Fgf18 expression and with the significant reduction in Fgf17 expression (Fig. 3). These observations suggest a possible role for FGFs in regulating the transition from proliferation to differentiation in the vermis. The onset of Purkinje cell differentiation is highlighted by extensive radial migration from the neuroepithelium and the expression of calbindin, a specific marker of the Purkinje cell lineage in the cerebellar anlage (Altman and Bayer, 1985, 1997; Yuasa et al., 1991). Additionally, thymidine radiographic studies have established that Purkinje cell migration is polarized and proceeds in a caudal-lateral to rostral-medial direction (Altman and Bayer, 1985, 1997), and that migrating Purkinje cells form a morphologically distinct layer of cells adjacent to the neuroepithelium in a domain called the cortical transitory zone (ctz) (Altman and Bayer, 1985) (Fig. 7A-D). Histological analysis of wild-type embryos showed that, at E14, midline sagittal sections showed no evidence of Purkinje cell migration (Fig. 7B). However, on sections 60-80 μm from the midline the ctz began to form (Fig. 7C) and in lateral sagittal sections, Purkinje cells have migrated progressively deeper into the cortical zone (cz) (Fig. 7D). These data indicate that at E14 Purkinje cell differentiation has not extended to the most medial portion of the vermis primordium.

To investigate the relationship between proliferation and differentiation, the timing of the formation of the ctz in wild-type embryos was examined by comparing 1 hour and 48 hours BrdU labeling. In embryos labeled for 1 hour at E14, cells in the neuroepithelium are BrdU-positive whereas all migrating cells in the ctz are BrdU-negative (Fig. 7E). However, if the embryos are labeled at E12.5 and harvested 2 days later, the cells in the ctz, but not those in the cortical zone (cz) at the caudal part of the vermis primordium are labeled with BrdU (Fig. 7F). These data demonstrate that the migratory cells in the ctz exited the cell cycle after E12.5, while migratory cells that have moved into the cz at the caudal and lateral part of the primordium exited the cell cycle prior to E12.5. Therefore, the ctz marks the most recent differentiation field, and the relative position of the ctz along the medial-lateral axis indicates the extent of differentiation.

Histological analysis of serial sagittal sections from E14 Fgf17^−/− and Fgf17^−/−, Fgf8^Δ2,3n/+ embryos showed that a ctz had clearly formed in the midline at the caudal part of the cerebellar primordium indicating that Purkinje cell migration has already begun in the most medial portion of the vermis anlage by E14 (Fig. 7G,H). Variations seen within stage-matched embryos are presented in Table 2.

Collectively, these data show that Purkinje cell differentiation has progressed rostrally in Fgf17^−/− mice and that differentiation was even more advanced in Fgf17^−/−, Fgf8^Δ2,3n/+ mice. In Fgf17^−/−, Fgf8^Δ2,3n/+ mice, the ctz not only formed at the midline in all embryos examined (6 of 6), but also in most cases (5 of 6), had extended into the rostral half of the vermis anlage (Fig. 7H, Table 2).

Calbindin is the earliest known marker for the Purkinje cell lineage and its expression increases as differentiation progresses (Yuasa et al., 1991). To further evaluate the progression of Purkinje cell differentiation, Calbindin expression was examined in the MHB junction at E14.5. In wild-type embryos, the most medial portion of the cerebellar primordium was calbindin-negative (Fig. 8A-C). Calbindin-positive cells were first detectable in sections ∼80 μm away from the midline (Fig. 8D). In contrast, in the most medial portion of the cerebellar primordium of Fgf17^−/−, Fgf8^Δ2,3n/+ mice, there was strong calbindin expression in the ctz and the cz (Fig. 8E-G). In the lateral part of the vermis primordium, there was a much larger and more intense calbindin-positive domain compared to wild type (compare Fig. 8H with 8D). These data, and the histological analysis, suggest that premature differentiation occurred in the medial portion of the vermis primordium as a consequence of decreased FGF signaling.

The vermis cerebellum has been functionally associated with motor coordination and maintenance of body equilibrium (Victor et al., 1959; Brodal, 1998). Gait patterns of 3-to 4-month-old wild-type (n=8) and Fgf17^−/− mice (n=8) showed a normal symmetric gait, characterized by the overlapping of hindlimb and forelimbs paw prints (Fig. 9 left). However, two of ten Fgf17^−/−, Fgf8^Δ2,3n/+ mice exhibited asymmetric gait patterns characterized by lagging of the hindlimbs (Fig. 9 center) or a wider spread between the hindlimbs (Fig. 9 right). The low penetrance of this phenotype indicates that either the functional consequence of tissue loss in the vermis is subtle or that functional compensation by other regions of the CNS occurs by 3-4 months of age. Because such compensation may be less effective in younger mice, further studies will be required to examine cerebellar function throughout postnatal development.

FGF8 is thought to be the signal that mediates MHB organizer function at early somite (6-12 somite) stages, consistent with its expression at that time of development. However, recent expression data (Maruoka et al., 1998; Xu et al., 1999; Fig. 3) showed that two other related Fgfs, Fgf17 and Fgf18, are also expressed in the MHB junction and that the expression of all three Fgfs, especially that of Fgf17, extended into the growth and differentiation phases of cerebellar development. Therefore, it became important to address the function of this family of FGFs at later stages of cerebellar development. BrdU-labeling studies indicated that the peak of cerebellar growth occurred at E10.5-E11.5, because more than 40% of the neuroepithelial cells in the cerebellar primordium were in S-phase at this time, and there is a rapid increase in the total number of cells (more than 5 fold, data not shown) in this region from E10.5 to E12.5. The peak in neuroepithelial cell proliferation 2 days after the onset of Fgf8 expression suggests that the MHB tissue loss observed in mice carrying hypomorphic alleles of Fgf8 is caused primarily by a patterning defect prior to E9.5 (Meyers et al., 1998). In contrast, the MHB tissue loss in the Fgf17^−/− mice and in Fgf17^−/−, Fgf8^Δ2,3n/+ mice reported here are likely the direct result of decreased proliferation and/or a subtle patterning defect affecting only the rostral vermis. A major primary defect in morphogenesis was ruled out by the observation that genes essential for patterning the MHB region were expressed in normal patterns in both types of mutants. In addition, histology of the cerebellar primordium of Fgf-deficient mice did not exhibit any abnormalities until E12.5 when the midline neuroepithelium in Fgf17^−/−, Fgf8^Δ2,3n/+ mice became thinner than that of wild-type neuroepithelium. Moreover, a growth defect is suggested by the significant reduction of BrdU-labeling indices in Fgf17^−/− and Fgf17^−/−, Fgf8^Δ2,3n/+ mice at E11.5 and E12.5. These data also demonstrate that FGF8 and FGF17 act together at these stages because compared to the Fgf17^−/− mouse, the Fgf17^−/−, Fgf8^Δ2,3n/+ mutant is more severely affected both in terms of a decreased cell proliferation index and in terms of the final size of the mature cerebellum. After E11, growth of the vermis anlage is therefore regulated by FGF signals in a dose-dependent manner, which are distinct from their role as an organizer signal 2-3 days earlier in development. It is important to note that the posterior half of the vermis is also reduced in size, although much less than the anterior half. This is most likely due to decreased proliferation at earlier stages of development (before E11.5) in precursor pools of cells that will populate both the anterior and posterior vermis. The more severely affected anterior vermis is likely to result from a more significant proliferation defect at later stages of development (E11.5-14). Because the vermis develops in a caudal to rostral orientation (Altman and Bayer, 1985, 1997) decreased neuroepithelial proliferation late in development should predominantly affect the population of cells that give rise to the rostral vermis.

Although the loss of the rostralmost lobes of the vermis appears to mimic a patterning defect, our model proposes that it is more likely attributable to premature exhaustion of the precursor pool of cells giving rise to this tissue. Nevertheless, it remains possible that part of the midline vermis fails to develop in Fgf17^−/−, Fgf8^Δ2,3n/+ mice and that the tissue that we observe in the midline is derived from more lateral tissue that now occupies a midline position.

As discussed, the Fgf17^−/− mice and the Fgf17^−/−, Fgf8^Δ2,3n/+ mice have a defect specific to the late phase of cerebellar development. We believe that this is caused by a temporal separation of FGF8 and FGF17 function. The expression patterns shown here suggest that FGF8 dominates at early somite stages, while FGF17 dominates after E11.5. Because FGF8, FGF17 and FGF18 have similar biochemical properties, the difference in temporal expression patterns is more likely to account for the phenotypic difference between mice carrying hypomorphic alleles of Fgf8 and mice lacking Fgf17. The expression patterns also predict a strong redundancy among these FGFs at E9.5-E11.5, which was partially revealed by the more severe phenotype of Fgf17^−/−, Fgf8^Δ2,3n/+ mice. In the future, the redundancy may be further examined by generating and analyzing mice lacking additional alleles of Fgf18.

Because Fgf18 expression is not detected during gastrulation, Fgf17^−/−, Fgf8^Δ2,3n/+, Fgf18^+/− or Fgf17^−/−, Fgf8^Δ2,3n/+ Fgf18^−/− embryos may survive long enough to address this issue. There is a strong possibility that the defect in MHB development of these compound mutant mice may occur at stages prior to E11.5. Because the data presented here suggests that FGF signaling no longer regulates the expression of En2, Wnt1, Pax5 and Otx2 at E12.5, mouse mutants lacking alleles of Fgf8, Fgf17 and Fgf18 with deficits earlier in development will help to identify the stage between E9.5 and E12.5, where FGF signaling shifts from a role as an organizer to that of a growth and differentiation factor.

The polarized progression of cerebellar differentiation was reported almost 15 years ago (Altman and Bayer, 1985). Unlike the extensive genetic studies on cerebellar patterning, the factors that control cerebellar growth and differentiation are largely unknown. One mystery, not explained by current models, is why differentiation in the vermis anlage proceeds in a caudolateral-to-rostromedial direction whereas differentiation in the cerebellar hemisphere anlage proceeds in a rostromedial to caudolateral direction. Here we have shown that prematurely decreasing and subsequently terminating FGF signaling in the vermis anlage accelerates the progression of Purkinje cell differentiation. We therefore propose that FGF8 and FGF17 normally function to inhibit differentiation in the vermis primordium. Such an inhibitory effect may be directly associated with the ability of FGF to maintain the precursor cell pool in an undifferentiated proliferating state, as demonstrated here and in other in vitro studies (Gritti et al., 1996; Johe et al., 1996), where FGF has been identified as a growth factor for CNS stem cells. Our observations suggest a model in which the polarized progression of differentiation in the vermis anlage may be achieved in part by the gradual decrease in Fgf expression and the gradual narrowing of the expression domains of Fgf8, Fgf17 and Fgf18 from E11.5 to E14.5. As a result of these dynamic changes in Fgf expression, precursor cells in the caudal part of the cerebellar primordium would first be released from FGF inhibition and subsequently allowed to differentiate, before those at the rostral end of the primordium where Fgf expression persists until E15.5 (Fig. 10). In addition, the polarized differentiation may also be facilitated by a previously reported, rostral-to-caudal, movement of isthmic cells (Hallonet and Le Douarin, 1993), which could push the precursor cells away from the field of FGF expression.

In addition to Fgf8, Fgf17 and Fgf18, Fgf15 is also expressed in the developing cerebellum (McWhirter et al., 1997 and data not shown). Unlike Fgf8, Fgf17 and Fgf18, Fgf15 is expressed in the anlage of the cerebellar hemispheres at E9.5-E10.5, and later is more restricted to the caudal part of the cerebellar hemispheres. Because the differentiation of the cerebellar hemisphere proceeds in a rostral-to-caudal direction, opposite to that in the vermis (Altman and Bayer, 1997), we hypothesize that the onset of differentiation in the cerebellar hemisphere and in the vermis are independently regulated by two different FGF gradients mediated by the Fgf8 family medially and by Fgf15 laterally.

Consistent with the expression of Fgf15 during cerebellar development, it is tempting to speculate that Fgf15 may also regulate the growth of the cerebellar hemispheres. This hypothesis helps to explain why tissue loss in mice lacking alleles of Fgf8 and Fgf17 is restricted to the vermis. Intriguingly, Fgf19, is most closely related to Fgf15 (Nishimura et al., 1999) and preliminary data suggests that Fgf19 is also expressed in the developing CNS. It will therefore be interesting to see if Fgf19 is also expressed in the developing cerebellar hemisphere. Interestingly, the subfamily of Fgf15 and Fgf19 is phylogenetically more related to the Fgf8 subfamily than to other subfamilies of Fgfs (data not shown) and is therefore likely to share similar receptor specificity towards FGFRs expressed in the neuroepithelial primordium. Phylogenetic analysis (Coulier et al., 1997) suggests that the FGF family has grown by gene duplication events throughout evolution, which may have contributed to the disproportional increase in cerebellar size seen in higher animals.

This work was supported by NIH grant CA60673 and a grant from the American Heart Association. We are grateful to E. Spinaio and X. Hua for their technical help, G. Martin and E. Meyers for generously providing Fgf8^Δ2,3n/+ mice and A. P. McMahon, A. L. Joyner and P. Gruss for providing in situ hybridization probes. We thank R. Kopan, I. Boime and R. Cagan for critically reading this manuscript.