Math1 controls cerebellar granule cell differentiation by regulating multiple components of the Notch signaling pathway

The highly ordered cytoarchitecture and the relative simplicity of cerebellar development make it one of the best studied systems for neurogenesis. Most of the cerebellar neurons (e.g. Purkinje cells, deep cerebellar nuclei and interneurons) arise at a ventricular zone located at the edge of the fourth ventricle (Hatten and Heintz, 1995). Precursors of the cerebellar granule cells (CGC)are born in a second proliferative zone, the rhombic lip, where they proliferate and later migrate via a rostral movement over the surface of the embryonic cerebellum (Altman and Bayer,1997; Gilthorpe et al.,2002; Wingate,2001). Consequently, these CGC precursors yield the external granule/germinate layer (EGL) of the cerebellum, a displaced germinal zone,where proliferation continues and peaks at postnatal day 7 (P7) in mouse(Altman and Bayer, 1997; Hatten et al., 1997; Hatten and Heintz, 1995). Postmitotic cells congregate in the inner EGL, and then migrate into the cerebellar cortex along Bergman radial glia towards their final destination:the cerebellar internal granule layer (IGL)(Edmondson et al., 1988; Fishman and Hatten, 1993; Hatten and Heintz, 1995). The later stages of CGC development - EGL formation and migration towards the IGL- have been extensively studied (reviewed by Goldowitz and Hamre, 1998; Hatten and Heintz, 1995; Millen et al., 1999; Wang and Zoghbi, 2001), in contrast to the earlier stages of precursor specification and differentiation,which are less characterized.

Math1 (Atoh1 - Mouse Genome Informatics) encodes a murine basic helix-loop-helix (bHLH) transcription activator(Akazawa et al., 1995; Ben-Arie et al., 1996),orthologous to the Drosophila atonal. In the developing cerebellum, Math1 is expressed in mitotic CGC at the rhombic lip and in the outer EGL (Akazawa et al., 1995; Ben-Arie et al., 1997; Ben-Arie et al., 2000; Ben-Arie et al., 1996; Helms et al., 2000). Genomic disruption has proven that Math1 is essential for proper development of CGC, as Math1 null mice lack the EGL(Ben-Arie et al., 1997; Ben-Arie et al., 2000). However, overexpression of Math1 resulted in cerebellar abnormalities without extra neurogenesis (Helms et al.,2001; Isaka et al.,1999), arguing against a proneural role for Math1 in the developing nervous system of the mouse.

The Notch signaling pathway is a crucial mechanism for controlling cell specification and differentiation in both invertebrates and vertebrates(Artavanis-Tsakonas et al.,1999; Beatus and Lendahl,1998; de la Pompa et al.,1997; Frisen and Lendahl,2001; Gaiano and Fishell,2002; Justice and Jan,2002). Notch signaling components, such as the receptors Notch1 and Notch2, the ligands Delta1 (Dll1 - Mouse Genome Informatics), Dll3, Jag1 and Jag2, the DNA-binding protein interactor Cbf1 (Rbpsuh -Mouse Genome Informatics), and the effectors Hes1 and Hes5were found to be expressed in the EGL of neonatal mice(Irvin et al., 2001; Kusumi et al., 2001; Solecki et al., 2001; Tanaka et al., 1999). Moreover, activation of Notch and overexpression of its effector Hes1, maintained the proliferation of CGC EGL precursors(Solecki et al., 2001). Loss of Notch1 was shown to result in a premature onset of neurogenesis,which resulted in a reduced number of neurons in the adult cerebellum(Lutolf et al., 2002). Similarly, the importance of Hes1 and Hes3 in cerebellar development was identified in knockout mice(Hirata et al., 2001).

Links between Notch signaling pathway and Math1 were identified in various tissues. Math1 was shown to be essential for the generation of inner ear hair cells (Bermingham et al., 1999; Chen et al.,2002; Kawamoto et al.,2003; Shou et al.,2003; Zheng and Gao,2000). Moreover, activation of Notch via Jag2 was shown to inhibit expression of Math1 in cochlear progenitor cells, possibly through the activity of Hes5(Lanford et al., 2000). Indeed, upregulation of Math1 in Hes1 and Hes5mutant cochleae suggested that Hes genes regulate hair cell differentiation by antagonizing Math1 expression(Zine and de Ribaupierre,2002). Notch pathway components were similarly found to be variably expressed in the mouse small intestine(Schroder and Gossler, 2002). Notably, in the small intestine of Math1-null mice, which lack secretory cells, the expression of Dll3 was halved, while Dll1,Hes1, Notch1, Notch2, Notch3 and Notch4 expression was unaffected (Yang et al.,2001).

In this study we aimed to deepen our insight into CGC neurogenesis, by taking advantage of Math1-null mice, in which this process is arrested. The development of CGC precursors in Math1-null mice was followed by examination of Math1 promoter activity. Rhombic lip cells were then cultured and analyzed for their survival, specification and differentiation in vitro. Our data show that lack of Math1 did not affect the viability of CGC or their specification. Rather, CGC progenitors were abnormal in their differentiation, as evident molecularly (by the continuous activation of Math1 promoter) and morphologically (by their inability to extend processes in culture). Among all Notch receptors and ligands expressed in the rhombic lip, Notch4 and Dll1 showed the most pronounced downregulation in Math1-null mice. Moreover, by testing two Notch effectors we have discovered that the expression of Hes5, but not Hes1, is Math1 dependent, and that MATH1 can bind directly Hes5, thus demonstrating a novel negative autoregulatory loop of Math1 expression. The feedback mechanism requires an accumulation of MATH1, and therefore provides an explanation for the delayed downregulation of Math1 in cultured cells. Taken together, our data reveal that Math1 controls cerebellar granule cell differentiation as well as its own expression, at least in part, through the Notch signaling pathway.

The generation of Math1-null allele mice has been previously described (Ben-Arie et al.,2000). In this line, the entire coding region of Math1has been removed, and replaced by a pSAβgal/PGK-neo cassette, such that lacZ expression is driven by the endogenous control elements of Math1. As Math1^{β-gal/β-gal} mice are not viable, heterozygous mice were mated to obtain all Math1 genotypes. The morning of vaginal plug appearance was considered as embryonic day (E)0.5. Experiments were conducted according to an ethical approval from the Hebrew University of Jerusalem, according to the Israeli laws.

Whole embryos or tissue staining was previously described(Ben-Arie et al., 2000). To stain cultured cells the wells were washed twice in PBS, fixed by 0.05%gluteraldehyde in PBS for 5 minutes at room temperature, and washed three times in PBS. Staining was performed at 37°C for about 10 hours, in solution of 1 mg/ml X-Gal, 5 mM potassium ferricyanide, 5 mM potassium ferrocyanide and 1 mM MgCl₂ in PBS. After postfixation in 4%paraformaldehyde in PBS, cells were counterstained by Nuclear Fast Red(Aldrich) and clarified in 75% Glycerol in PBS.

Culturing of cerebellar granule cells is based on a previously described procedure (Alder et al., 1996; Hatten et al., 1998). Briefly,embryos were collected in ice-cold CMF-PBS(Hatten et al., 1998), and the cerebellum isolated under a dissecting microscope by two incisions across the mesencephalon/metencephalon border and across the fourth ventricle. The rhombic lip tissue was pealed off with fine forceps, placed in CMF-PBS and stored on ice. Dissociation was performed by incubation of the tissue in 0.08%Trypsin (Biological Industries, Beit-Haemek, Israel), 0.02% EGTA, 0.05 mg/ml DNaseI (Sigma) in CMF-PBS, for 15 minutes at 37°C; which was then changed to 0.05 mg/ml DNaseI, 0.45% Glucose in ice cold Eagle's basal medium (BME). The tissue was triturated by passing through a pipettor tip,centrifuged at 700 g at 4°C for 5 minutes, and pellets resuspended in 50 μl granule cell medium(Hatten et al., 1998)supplemented by 5% fetal calf serum and 10% horse serum (Biological Industries, Beit-Haemek, Israel). Cells were diluted to 1200-1300 cell/μl before plating into Terasaki Micro Plate (#1006-01-3, Robbins, Sunnyvale, CA). Normally, four or five wells were plated from each embryo(22×10³ cells/well). Cultures were grown in 95% air/5%CO₂ humidified incubator, at 37°C. Half the medium was changed on the next day after plating and every other day thereafter.

Liquid assay for the lacZ reporter activity was performed using the All-in-One Mammalian β-Galactosidase Assay Kit (Pierce, Rockford,IL). Cultured rhombic lip cells grown in Terasaki plates were washed with PBS,lysed by the addition of 29 μl M-PER (Pierce, Rockford, IL) per well and incubated for 5 minutes. An aliquot of 20 μl was transferred into a 96-well plate, and 180 μl All-in-One reagent added. Reaction was carried out at 37°C for 6 hours and color development was measured every hour at 405 nm. A second aliquot of 8 μl was used for protein quantification; using Protein-Assay Reagent (BioRad, Hercules, CA).

Cultured cells were fixed by 4% paraformaldehyde in PBS for 15 minutes at room temperature, washed three times with PBS, and blocked by 5% normal goat serum, 2% BSA, 0.1% Triton X-100 in PBS for 1 hour at room temperature. Primary antibodies were diluted in blocking buffer and incubated overnight at 4°C, then for 1 hour at room temperature. The antibodies used were: mouse anti-β-tubulin (1:10, DSHB, E7), rabbit anti-NF160 (1:100, Sigma, N4142),mouse anti-phosphorylated neurofilaments (1:5, DSHB, RT97) and mouse anti-NCAM(1:5, DSHB, 5B8). Cells were washed four times with 0.1% Triton X-100 in PBS;before the addition of secondary antibodies conjugated to FITC or Biotin(Sigma), and incubated for 2 hours at room temperature, after which they were washed three times with PBS. For Biotin-conjugated antibodies StreptAvidin-TexasRed (Vector Laboratories, Burlingame, CA) was used for visualization. Counterstaining by DAPI was performed before mounting with 1%n-propyl-galate (Sigma) in 90% glycerol. Pictures were taken under an Axioskop2 microscope (Zeiss, Germany), using a DP10 digital camera (Olympus,Germany). Images were assembled using NIH ImageJ software(http://rsb.info.nih.gov/nih-image/index.html).

For quantification of processes the cultures were grown for 6 days, fixed,blocked and stained with mouse anti-β-tubulin as above. Then, cells were washed, incubated for 2 hours at room temperature with a secondary antibody conjugated to peroxidase (Jackson ImmunoResearch, West Grove, PA) and washed. The cells were then lysed by CytoBuster (Novagene, Milwaukee, WI) and the content of each two wells combined. A colorimetric reactions was initiated by the addition of 1mg/ml ABTS(2,2′-Azino-bis(3-ethylbenzothiazoline-6-sulfonic acid, diammonium salt)), 0.003% H₂O₂ (Sigma), 28 mM citric acid and 44 mM Na₂HPO₄. The O.D (405 nm) was measured every 15 minutes to ensure that the values are within the linear range.

RNA was extracted as described(Chirgwin et al., 1979). Cultured cells were lysed with 25 μl/well of lysis buffer (4 M guanidine thiocyanate, 25 mM sodium citrate, 17 mM N-laurylsarcosine) for 5 minutes at room temperature and kept at -70°C. After genotyping lysates were thawed and mixed with 1 μl β-mercaptoethanol, 12.5 μl 2M sodium acetate pH 4.0, 125 μl acidic phenol, 25 μl chloroform-isoamyl alcohol (49:1). The aqueous phase was extracted twice using chloroform-isoamyl alcohol,precipitated by isopropanol with glycogen as a carrier, washed by 70% ethanol,dried, dissolved in 25 μl water, and DNaseI treated using the DNA-free kit (Ambion, Austin, Texas). Reverse transcription was carried out by RevertAid H Minus First Strand cDNA Synthesis Kit (Fermentas MBI, Vilnius,Lithuania).

PCR amplifications were performed using FastStart Taq DNA polymerase(Roche, Germany), 0.2 mM dNTPs, 1.5 mM MgCl₂ and 1 μM each primer. The thermocycling parameters for Zic1, Zipro1 andβ-actin (set A) were: 94°C/4 minutes, 40 cycles of 94°C for 30 seconds, 55°C for 30 seconds, 72°C for 30 seconds, and 72°C for 3 minutes; and for Hes1, Hes5 and β-actin (set B): 94°C for 4 minutes; 34 cycles of 94°C for 30 seconds, 68°C for 120 seconds and 68°C for 5 minutes.

Real-time amplifications were performed on Rotor-Gene machine (Corbett Research, Australia) using 2 mM MgCl₂ and ×0.3 SYBR I Green. Thermocycling conditions were 94°C for 4 minutes, then 45 cycles of 96°C for 25 seconds, 60°C for 20 seconds, 72°C for 30 seconds;72°C for 1 minute. Amplification of a single product was verified by melting curves, and the correct product size by gel separation. For quantification, calibration curves were run simultaneously with experimental samples and C_t calculations were performed by the Rotor-Gene software.

The primers used were as follows: Zic1, (F) GGCCAACCCCAAAAAGTC,(R) CGTTAAAATTCGAAGAGAGCG; Zipro1, (F) CCAGACTCCAAAGCGGTTCTGAG, (R)AGTGTCATGGTACCCAAATTG; β-actin (A), (F) TGTTACCAACTGGGACGACA, (R)TGTTACCAACTGGGACGACA; β-actin (B), (F) GTGGGCCGCTCTAGGCACCAA, (R)CTCTTTGATGTCACGCACGATTTC; Hes1, (F) AGCTGGAGAGGCTGCCAAGGTTT, (R)ACATGGAGTCCGAAGTGAGCGAG; Hes5, (F) TTAAGCAAGTGACTTCTGCGAAGTTC, (R)GGCCATGTGGACCTTGAGGTGAG; Notch1, (F) AGAGATGTGGGATGCAGGAC, (R)CACACAGGGAACTTCACCCT; Notch2, (F) TGTACCAGATCCCAGAGATGC, (R)GTCAGATGCAGAGTGTGGTGA; Notch3, (F) AATCCTGTAGCTGTTCCCCTC, (R)CTGGGCTAGGTGTTGAGTCAG; Notch4, (F) ATCACAGGATGACTGGCCTC, (R)ACTCGTACGTGTCGCTTCCT; Dll1, (F) CTGAGGTGTAAGATGGAAGCG, (R)CAACTGTCCATAGTGCAATGG; Dll3, (F) CACCAGTAGCTGCCTGAACTC, (R)GTTAGAGCCTTGGAAACCAAG; Dll4, (F) CCTCTAGGCAAGAGTTGGTCC, (R)TAGAAAGGCCAGTGCTTCTGA; Jag1, (F) TGACATGGATAAACACCAGCA, (R)GCAGCCCACTGTCTGCTATAC; Jag2, (F) ATTGTAGCAAGGTATGGTGCG, (R)GCACAGTTGTTGTCCAAATGA.

Full length Math1 and E47 cDNAs were cloned into pGEX-3X and pET28(a)expression vectors, respectively. MATH1/GST and E47/6xHIS fusion proteins were purified from IPTG-induced BL21 bacteria by agarose-Glutathione (Sigma, USA)or Co Talon Affinity Resin (Clontech, USA), respectively.

For EMSA, two oligonucleotides CAGGAGCCCTGCCAGGCAGCTGGTGGCATTCTCCA and GTGGAGAATGCCACCAGCTGCCTGGCAGGGCTCCTG were annealed and labeled by Klenow enzyme in the presence of [α-³²P]dCTP. A positive control probe was E1 according to (Akazawa et al.,1995). EMSA was carried out as previously described(Ben-Porath et al., 1999).

Targeted deletion of Math1 (Math1^-/-) or a total replacement of the coding region by a reporter gene(Math1^{β-gal/β-gal}) was shown to cause lack of the EGL at the time of birth (Ben-Arie et al., 1997; Ben-Arie et al.,2000). Here, we further examined Math1^β-gal/+ (which displayed a normal phenotype and could serve as controls) and Math1^{β-gal/β-gal}mice by whole-mount X-Gal staining of the brain.

As seen in Fig. 1A-D, by E14.5 CGC precursors occupy the cerebellar rhombic lip, as revealed by Math1/lacZ activity (lacZ expression under Math1endogenous control elements). Similar staining pattern in Math1^β-gal/+ (Fig. 1A,C) and Math1^{β-gal/β-gal}(Fig. 1B,D) indicated that in Math1-null mice CGC precursors were born and reached a state of differentiation that required Math1 expression. At E16.5, Math1^β-gal/+ displayed staining all over the surface of the developing cerebellum (Fig. 1E), consistent with the formation of EGL by a rostromedial migration of CGC progenitors from the rhombic lip(Altman and Bayer, 1997; Gilthorpe et al., 2002; Hatten and Heintz, 1995). By contrast, in Math1^{β-gal/β-gal} there were less Math1/lacZ-positive cells at the cerebellar surface, although the rhombic lip continued to include surviving progenitors(Fig. 1F). At both stages, the rhombic lip was smaller in Math1-null embryos when compared with the heterozygous littermate. This was in agreement with the previous histological analysis of sectioned cerebella and proliferation rate measured by BrdU incorporation (Ben-Arie et al.,1997; Ben-Arie et al.,2000) and suggested that CGC progenitors were viable even without Math1 expression. Moreover, examination of the entire brain revealed no ectopic migration in Math1-null mice, excluding such an explanation for the lack of EGL.

To investigate the origin of the EGL agenesis in Math1-null mice,we attempted to separate the complex processes they normally undergo in vivo,by examining the CGC progenitors in vitro. In addition, culturing allowed us to follow cells isolated from the rhombic lip, which is a transient structure that disappears during normal embryogenesis.

Based on the spatiotemporal expression pattern of Math1/lacZ(Fig. 1), we chose to examine CGC precursors at E14.5, as an advanced stage in which the rhombic lip progenitors are present in both Math1^{β-gal/β-gal}and Math1^β-gal/+, and the abnormal phenotype is only emerging. A typical example of a dissected rhombic lip from Math1^β-gal/+ cerebellum, which was subsequently stained by X-Gal, showed that an enriched source of Math1/lacZ-expressing cells could be obtained(Fig. 1G). Isolation of a totally pure CGC population from individual embryos was impractical, but not essential, as similar proportions of Math1/lacZ-negative cells were present in the different cultures compared, regardless of Math1genotype. As the isolated tissues may contain CGC precursors as well as other cell types, we use the term `rhombic lip cells'. Further confirmation for the enrichment of the isolated rhombic lip tissue by CGC was obtained by RT-PCR. Isolated rhombic lips from Math1^β-gal/+ expressed lacZ and Math1, as expected(Fig. 1H). An independent verification was provided by the expression of Zipro1 (RU49/Zfp38), a zinc-finger transcription factor specifically expressed in CGC from early stages (Yang et al., 1996)(Fig. 1H).

We followed the expression of Math1/lacZ over time in cultures obtained from individual embryos of the three Math1 genotypes. No notable differences, such as density of cells or increased number of dead cells, were observed in cultures from controls and Math1^{β-gal/β-gal} (data not shown). Staining for lacZ after 3 days in culture (Fig. 2A-F) revealed no background in cultures from Math1^+/+, although most cells from Math1^β-gal/+ (Fig. 2B,E) and Math1^{β-gal/β-gal}(Fig. 2C,F) appeared blue. Comparison of cell density, proportion of stained cells and staining intensity did not imply any major difference between Math1 null and control cells at this stage. Moreover, Math1/lacZ expression indicated that CGC precursors lacking Math1 survived after 3 days in vitro and continuously maintained Math1 promoter activity. Hence, it was concluded that Math1 was not essential for the survival of CGC precursors.

As no differences were visible between Math1-null and control cells after 3 days in vitro, we challenged the cells with a longer culturing period (Fig. 2G-L). After 6 days in vitro, Math1/lacZ expression was dramatically decreased in cultures from Math1^β-gal/+(Fig. 2H,K). This observation was consistent with the expression of Math1 in the outer EGL, and its downregulation in differentiating cells at the inner EGL(Helms and Johnson, 1998). Surprisingly, Math1/lacZ activity in Math1^{β-gal/β-gal} was still strong after 6 days in culture (Fig. 2I,L). Thus, Math1 promoter activity remained high in cells derived from Math1^{β-gal/β-gal}, while downregulated in cells from Math1^β-gal/+ littermates.

To refine this observation, we used a quantitative colorimetric assay forβ-galactosidase activity in the cultured cells. After 3 days in culture, Math1/lacZ activity was very similar in Math1^β-gal/+ and Math1^{β-gal/β-gal} cultures, much above the background measured in Math1^+/+(Fig. 2M). However, after 6 days in culture a significantly higher level of β-galactosidase activity remained in Math1^{β-gal/β-gal} cells, in contrast to the significant reduction of activity in Math1^β-gal/+ cultures (P<0.001, t-test). Math1 was shown before to act as a positive autoregulator (Helms et al.,2000), and our data demonstrated for the first time a role for Math1 also in a negative autoregulation of its own expression.

The absence of an essential transcription factor may change the fate of neural precursor cells (Guillemot,1999; Hassan and Bellen,2000). Moreover, culturing of normal neural precursors may lead to alteration of the cellular identity that may result in a fate switch, by accelerating differentiation or causing de-differentiation(Anderson, 2001). Therefore, we studied cell fate specification of the rhombic lip cells in Math1-null mice.

As Math1 is expressed in a limited time window during CGC development, both accelerated differentiation and de-differentiation of the progenitors may silence Math1 promoter, resulting in a decreased Math1/lacZ expression. Therefore, we examined the expression of two CGC-specific transcription factors Zic1 and Zipro1, which are expressed in rhombic lip precursors, as well as in mature CGC in the IGL(Aruga et al., 1994; Nagai et al., 1997; Yang et al., 1996). RT-PCR analysis was performed on cells cultured for 3 and 6 days from all Math1 genotypes (Fig. 3). Similar levels of Zic1 and Zipro1transcripts were detected in Math1^{β-gal/β-gal},when compared with Math1^+/+ and Math1^β-gal/+ littermates at both time points. These data revealed that the initiation and maintenance of the correct fate of rhombic lip cells destined to become CGC, was independent of Math1,and was not lost upon prolonged growth in vitro.

As specification was not altered in Math1-null CGC, we examined the in vitro differentiation capability of the cells. Embryonic CGC precursors have been shown before to be able to differentiate in culture(Alder et al., 1996). We chose to examine process extension as a pronounced phenotype of neuronal maturation.

Immunofluorescent detection of β-tubulin, which is known to be expressed in CGC processes (Alder et al.,1999; Helms et al.,2001), showed that a large number of processes have developed from Math1^+/+ and Math1^β-gal/+ rhombic lip cells, when cultured for 6 days, but not in cultures from Math1^{β-gal/β-gal}(Fig. 4A-C). Quantification of the processes evaluation was achieved by β-tubulin staining of similarly grown cultures followed by a colorimetric assay. The absorbance of control cultures from Math1^+/+ and Math1^β-gal/+ (n=15) was 0.30 (±0.04),and was reduced to 0.13 (±0.02) in Math1^{β-gal/β-gal} (n=11), which is significantly lower (P<0.001, t-test). The difference is smaller than visualized by immunostaining, as staining of both the soma and processes were measured. Staining against phosphorylated neurofilaments(Fig. 4D-F), NF160(Fig. 4G-I) and NCAM(Fig. 4J-L) illustrated long processes in Math1^+/+ and Math1^β-gal/+, but not in Math1^{β-gal/β-gal}. Control nuclear staining by DAPI showed a uniform cell density in all genotypes(Fig. 4M-O, and Fig. 4A-L as counterstaining),indicating a similar survival of cells after 6 days in culture. The neural phenotype displayed by only a fraction of the cultured cells was consistent with previous reports that only some of rhombic lip precursors are competent to differentiate in vitro (Alder et al.,1996).

The molecular and phenotypic manifestation of neural differentiation was detected in cultured rhombic lip cells from wild-type and heterozygous, but not Math1-null cultures, although the specification and survival of cells appeared similar. These finding are compatible with the hypothesis that Math1 is essential for neural differentiation of CGC progenitors. The molecular mechanisms underlying this ability should be further pursued.

Accumulating data support the involvement of the Notch signaling pathway in cerebellar development, and connect Math1 to this pathway in various organs during embryogenesis. Therefore, we first analyzed the expression of various receptors (Notch1 to Notch4) and ligands (Dll1,Dll3, Dll4, Jag1 and Jag2) in the rhombic lip at E14 by quantitative real-time RT-PCR. We assumed that analyzing the absolute level of each transcript combined with a comparison of its amount in Math1^+/+ and Math1^{β-gal/β-gal} is indicative of its importance for CGC development.

Among all Notch receptors tested, the level of Notch2 was the highest, being 145-fold higher than Notch1 and more than 20-fold higher than Notch3-4 (Fig. 5A). A striking difference was detected also for Notch ligands,where the level of Jag1 and to a lower extent Dll1 was the highest among the five ligands tested (Fig. 5A).

When expression of the receptors was tested in Math1^+/+and Math1^{β-gal/β-gal} littermates the largest reduction of 2.8-fold was detected for Notch4(Fig. 5B). Among the Notch ligands, the level of Dll1 was reduced by 2.5 fold, while Dll3,Dll4 and Jag2 transcript levels were also significantly decreased by 1.7-fold (Fig. 5C). Overall, all Notch receptors and ligands tested were expressed in the developing cerebellum. However, the differences in the level of downregulation in Math1^{β-gal/β-gal} implied that only some of the Notch receptors and ligand were related to Math1 function.

Seeing that the Notch signaling pathway was related to CGC development, we next examined two Notch effectors Hes1 and Hes5 in E14.5 rhombic lips and primary cultures from Math1^+/+ and Math1^{β-gal/β-gal} littermates by RT-PCR(Fig. 6). Both Hes1and Hes5 were found to be expressed in wild-type rhombic lip, with a higher level of the latter. Although the expression of Hes1 andβ-actin was similar in the two genotypes, Hes5 expression was reduced in Math1^{β-gal/β-gal} rhombic lip, when compared with Math1^+/+(Fig. 6). Moreover, the decrease in Hes5 expression level in Math1^β-gal/+ was even more pronounced in rhombic lip cells cultured for 3 and 6 days (Fig. 6). The reduction of Hes5 in CGC progenitors from Math1 null suggested a positive control of Math1 over Hes5, but not Hes1, expression, which was not identified previously.

To establish a more causal relationship between Math1 and Hes5, we have tested the ability of Math1 gene product to recognize and bind Hes5 (Fig. 7). Recombinant and purified MATH1 and E47 were allowed to heterodimerize and then subjected to an electrophoretic mobility shift assay. As targets we have used a proven MATH1 target (Fig. 7A)(Akazawa et al., 1995) and an E-box located downstream to Hes5, similar to the position of the E-box-containing enhancer involved in the positive autoregulation of Math1 (Helms et al.,2000). Although MATH1/E47 could bind directly Hes5, MATH1 by itself did not, in contrast to the known capability of the homodimeric E47(Akazawa et al., 1995; Helms et al., 2000). A cold oligonucleotide containing the E-box target blocked the binding, which indicated the specificity of the protein/DNA interaction. The ability of Hes5 and Math1 gene products to affect the transcription of each other by a luciferase reporter assay is currently not feasible as Math1 promoter has not been identified yet.

In the fly atonal has a proneural role in the PNS, but not in the CNS, where it controls arborization. The question therefore arose as to the function of Math1 in the developing cerebellum: does it have a proneural role in the specification of rhombic lip stem cells or progenitors,or does it work later in development, during differentiation? Moreover, as Math1 was found to participate in the Notch signaling in inner ear hair cells and in the intestine, we were interested in learning whether this is a general theme that takes place also during CGC development. By studying the effect of Math1 knockout on rhombic lip development in vivo and in vitro, we found that Math1 was not essential for the specification of rhombic lip cells, but for their proper differentiation. The lack of Math1 interrupted the normal downregulation of Math1promoter activity, and inhibited the ability of rhombic lip cells to develop processes in culture. Moreover, Math1-null mice displayed a selective downregulation of Notch receptors and ligands in the rhombic lip, and revealed a novel negative autoregulatory loop controlling Math1 expression through the Hes5 effector.

Whole-mount X-Gal staining demonstrated clearly that rhombic lip CGC precursors were born in Math1^{β-gal/β-gal} mice,but failed to migrate out to form the EGL, consistently with previous studies(Ben-Arie et al., 1997; Ben-Arie et al., 2000). The thinner rhombic lip identified by whole-mount staining of Math1-null mice was in full agreement with previous analyses and the fact that decreased proliferation was detected by a BrdU incorporation assay(Ben-Arie et al., 1997). As we examined the entire cerebellar region, the absence of ectopic staining in Math1^{β-gal/β-gal} mice excluded the likelihood of abnormal migration of Math1/lacZ-expressing cells. The possibility that ectopic cells were not stained, as they did not maintain Math1promoter activity, is less probable, as the rhombic lip kept staining until E18.5 in vivo, and the precursors maintained Math1/lacZ expression for an extended period in vitro.

The fact that rhombic lip CGC precursors activated Math1/lacZ expression did not provide a definite answer to the question of the specification status of the progenitors. To address whether Math1 is needed for proper fate determination we examined the expression of two more transcription factors, Zipro1 and Zic1 known to be expressed in CGC and their progeny(Alder et al., 1999; Aruga et al., 1998; Aruga et al., 1994; Nagai et al., 1997; Yang et al., 1999). The continuous expression of both genes in the rhombic lip and in cultured progenitors was shown to be Math1 independent, which lead us to the conclusion that Math1 was not required for the initial specification of granule cell progenitors and for the maintenance of granule identity, both in vivo and in vitro.

The relationship between Math1, Zic1 and Zipro1 is noteworthy. We show that in Math1-null mice both Zic1 and Zipro1 were normally expressed in CGC in vivo and in vitro, which may indicate that they act upstream to Math1. However, this notion is contradicted by other data. First, Zic1 expression in the developing neural tube is broad and becomes confined to the rhombic lip only by E12(Aruga et al., 1994), whereas Math1 expression at the neural tube begins at E9(Akazawa et al., 1995; Ben-Arie et al., 1997; Ben-Arie et al., 2000). Similarly, Zipro1 is expressed also in granule cells of the olfactory bulb and dentate gyrus, where no Math1 expression was reported(Yang et al., 1996). Second, Zic1 and Zipro1 knockout and overexpression in mice demonstrated that these genes regulate cerebellar patterning and EGL proliferation at stages later then those affected by Math1 deletion(Aruga et al., 1998; Yang et al., 1999). Third, Zic1 was recently shown to bind an enhancer of Math1 and to downregulate Math1 expression. However, Zic1 acts through repression of the positive autoregulation of Math1 itself(Ebert et al., 2003), which is not the major regulatory element of Math1 expression, as the autoregulation depends on initial activation of Math1 by independent upstream genes. Taken together, Math1, Zic1 and Zipro1 seem to affect cerebellar development through parallel, yet crosstalking, signaling pathways.

Rhombic lip cells from both Math1^+/+ and Math1^β-gal/+ E14 embryos reaggregated in culture, as expected (Alder et al., 1996). However, only after a longer incubation period in vitro (between 3 and 6 days)did a complex network of processes form, without the addition of supplements like BMPs of NGF. Immunoreactivity with β-tubulin, phosphorylated neurofilaments, NF160, NCAM and the distinct process morphology, confirmed a progress of the rhombic lip cells towards a neural phenotype. By contrast, Math1^{β-gal/β-gal} cultures developed few processes and growth cones, and lacked well developed neural extensions. During normal development in vivo, CGC do not grow extensions until they are situated in the inner EGL and become competent to start the inward radial migration to form the IGL (Hatten and Heintz, 1995). Therefore, culturing and analysis of the process outgrowth were not supposed to mimic the in vivo situation, but rather allow examination of the developmental potential of the progenitors, separating it from the need to migrate to the EGL, the place at which this morphological change normally takes place.

Normally, at the rhombic lip stage, CGC undergo proliferation and consequently migrate out of the rhombic lip: two abilities that are affected in Math1-null mice. As both functions mark the progress in the developmental program, which require Math1 for the regulation of its target genes, they can be regarded as Math1-dependent differentiation events. We propose that improper differentiation is the cause for developmental arrest in the rhombic lip. A simplistic view of the lack of EGL may suggest that Math1 was essential for activation of genes, which convey a migratory ability, or that their products are part of the migratory machinery per se. However, as the transcription of those genes is under the control of Math1, directly or indirectly, the lack of migration from the rhombic lip may be regarded as the outcome of improper differentiation of the progenitors. Hence, we suggest that only after Math1 is activated do rhombic lip cells acquire the ability to further differentiate.

Helms et al. (Helms et al.,2000) reported a positive autoregulation of Math1 over its own expression, through an E-box-containing downstream enhancer, which was shown to bind Math1. Transgenic mice expressing a Math1/lacZreporter, under various control elements flanking Math1 ORF,recapitulated most of the endogenous Math1 expression. However, the same transgene was not expressed when the mice were crossed with Math1-null mice, as no MATH1 was available to activate its enhancer(Helms et al., 2000). The fact that a Math1/lacZ reporter is expressed in Math1^{β-gal/β-gal} mice, which are a completely null for Math1, established the existence of additional Math1-independent control elements that activates Math1expression. Moreover, as we found a continuous expression of Math1/lacZ in rhombic lip cultured cells, it seemed that the major control over Math1 expression is MATH1 independent, and that the positive autoregulation contributes mainly to the refinement of Math1levels.

During normal cerebellar development Math1 is expressed in granule cell precursors and in the rhombic lip and outer EGL, and is turned off in postmitotic cells in the inner EGL (Akazawa et al., 1995; Ben-Arie et al.,2000; Helms and Johnson,1998). However, upstream genes and control mechanisms regulating the expression of Math1 are not yet fully identified. In the spinal cord of Gdf7 mutant mice, Math1 expression does not continue after E10.5, but the addition of GDF7 or BMP7 markedly increased Math1 expression (Lee et al.,1998). Similarly, the dorsal midline cells adjacent to the rhombic lip express GDF7, BMP6 and BMP7, which were demonstrated to induce Math1,En1/2, Zic1 and Wnt3a in the ventral mesencephalon/metencephalon neural tube. The induction of those genes normally confined to dorsal cells that develop into CGC precursors indicates the ability of BMP factors to determine the neural subtype fate, and suggests that BMPs regulate Math1 expression (Alder et al.,1999).

The evolutionarily conserved Notch signaling pathway mediates cell-to-cell communication to regulate cell fate decisions and patterning in both invertebrates and vertebrates. In the developing nervous system Notch signaling was classically regarded as a mechanism that keeps cells in an undifferentiated state. However, recently Notch signaling was found to be important for differentiation of glial cells and the organization of neuronal processes (Frisen and Lendahl,2001; Justice and Jan,2002). To shed light on the role of Notch signaling in CGC development, and based on the observations that various components of the pathway are expressed in various stages of cerebellar development, we analyzed their expression in the rhombic lip. We have found that the Notch2receptor and Jag1 ligand are the most abundant species, although all known receptors and ligands tested were expressed. Our findings are in agreement with previous studies that were mostly concerned with later stages of cerebellar development (Irvin et al.,2001; Kusumi et al.,2001; Solecki et al.,2001; Tanaka et al.,1999). Moreover, we have identified a selective downregulation of Notch4, Dll1, Dll3, Dll4 and Jag2 in the rhombic lip of Math1-null mice.

Because in Math1-null mutants the level of Notch receptors and ligands was affected, we examined whether Math1 had a transcriptional control over the Notch effectors Hes1 or Hes5 in the rhombic lip. RT-PCR analysis of Hes1 and Hes5 expression in rhombic lip tissue and in cultured cells after 3 and 6 days demonstrated a continuous downregulation of Hes5, but not of Hes1, in Math1-null mice. EMSA analysis has indicated that MATH1 can bind an E-box-containing sequence flanking Hes5, which suggests a novel control mechanism of Math1 over the transcription of Hes5,which is known to act as Math1 suppressor. Taking the new and established data together, we suggest a possible model linking some of the genes and interactions involved in CGC development(Fig. 8). According to our hypothesis, Hes5 normally downregulates Math1, which in turn further activates Hes5 transcription (directly or indirectly), and thus an increasing suppression of Math1 develops. However, in Math1^{β-gal/β-gal} cells, this feedback loop is interrupted, as there is no Math1 gene product to further activate Hes5. Therefore, the level of Hes5 gene product cannot increase, and Math1 promoter remains active, which is in full agreement with our observations. The model also provides an explanation for the delay in the downregulation of Math1 promoter activity, seen in cultured rhombic lip cells from Math1^β-gal/+ mice after 6, but not 3, days in vitro. Accordingly, at E14.5 there is a balance between MATH1 and HES5 levels, in which both the positive and negative regulatory loops take place. However, with time, the level of MATH1 increases due to the positive autoregulation, which finally leads to an increase in the level of HES5 until it reaches the threshold needed to attenuate Math1 transcription. Further experiments are needed in order to establish and verify the interplay between all the genes and proteins presented in the suggested model.

An inhibitory effect of Hes1 over Math1 activity was previously demonstrated, as transfection of a mouse pluripotent cell line with Math1 induced transcriptional activation of a luciferase reporter,which was inhibited by cotransfection with Hes1(Akazawa et al., 1995). Hes1 and Hes5 knockout mice have supernumerary hair cells,which express Math1 (Zine et al.,2001). Based on the characterization of both lines, it was suggested that in inner ear hair cells Math1 controls Jag2expression, which is repressed by Hes1 and Hes5 through the inhibition of Math1 activity(Zine et al., 2001). Moreover, Hes1 was also demonstrated to highly repress Math1-induced hair cell generation in cochlear explants(Zheng et al., 2000). However,the expression level and cellular localization of Hes1 were unaffected in the intestine of Math1-null mice(Yang et al., 2001),suggesting that the interrelations between the genes are also context dependent and vary in different tissues.

Mutual effects between bHLH factors and Hes genes are not limited to Math1 (reviewed by Guillemot,1999; Kageyama et al.,1997). Cau et al. (Cau et al.,2000) have demonstrated a complex interplay between Hes genes and Mash1 in the olfactory epithelium. Mash1 was expressed ectopically in Hes1 mutants, but normally in Hes5 mutants. By contrast, in Mash1 knockout the expression of Hes1 was unaffected, while Hes5 level was severely reduced(Cau et al., 2000). However,retroviral overexpression of Hes5 repressed Mash1 expression in oligodendrocytes precursors (Kondo and Raff, 2000). It was therefore proposed that Hes1represses Mash1, while Mash1 activates Hes5, which in turn represses Mash1. This mode of action is very similar to the model we propose for Math1 action.

Interestingly, during recent years Notch signaling has been linked not only to neural and glial cell fate determination, but also to the control of process outgrowth (Frisen and Lendahl,2001). Upregulation of Notch was shown to inhibit process extension or even cause their retraction, while repression of Notch signaling enhanced process outgrowth (Berezovska et al., 1999; Franklin et al.,1999; Sestan et al.,1999). However, in cultured CGC we have noticed that downregulation of Notch receptors and ligands in Math1-null mice was accompanied by a reduction in process outgrowth. However, the exact molecular mechanism underlying the relationship between this downregulation and process extension should be further examined.

The correlation between the expression of Notch effectors, such as Hes1 and Hes5, in controlling process outgrowth has been demonstrated in various experimental systems. Expression of Hes1 in PC12-E2 cells inhibits NCAM-dependent process outgrowth(Jessen et al., 2003), and its expression inhibits both the intrinsic and NGF-induced process outgrowth of embryonic day-17 rat hippocampal neurons in culture(Castella et al., 1999). Similarly, constitutive expression of the intracellular domain of Notch1, which activates Hes1 promoter in SH-SY5Y neuroblastoma cells, inhibits their spontaneous and induced process outgrowth(Grynfeld et al., 2000). Interestingly, axonal injury of corticospinal and dorsal root ganglion neurons suppresses Hes gene expression, possibly as part of the initiation of a regenerative response (Kabos et al.,2002). Hence, our data support the hypothesis that Math1influences process outgrowth, via the Notch pathway, by regulation of Hes5 expression.

We thank Esther Golenser and Theodora Bar-El for their valuable assistance. We are grateful to Huda Y. Zoghbi (Houston, TX) for providing us with Math1 knockout mice, to Benjamin Aroeti for generously sharing his tissue culture facility, and to Hermann Rohrer and Uri Gat for carefully reading the manuscript. We are indebted to Hermona Soreq for her fruitful advice and support. This work was supported by grants from the US-Israel Binational Science Foundation (1998-066), the Israel Science Foundation(587/02), the European Community (QLG3-CT-2000-00072) and the Roland Center for Neurodegenerative Diseases.