Zic1 represses Math1 expression via interactions with theMath1 enhancer and modulation of Math1 autoregulation

The differentiation of a proliferating neural progenitor into a postmitotic neuronal cell is a critical transition during mammalian neural development. During this time period, extensive cell lineage commitments are formed based on signals from dorsoventral and anteroposterior signaling centers(Altmann and Brivanlou, 2001;Briscoe et al., 2000;Chiang et al., 1996;Lee et al., 2000). These signals are integrated with cues controlling the timing of differentiation,such as those modulated by the effects of lateral inhibition(Cai et al., 2000;Gaiano et al., 2000;Morrison et al., 2000). The neural bHLH family of transcription factors, which is expressed in specific populations of neural precursors, figure prominently in this process. The regulation of these transcription factors, including Mash1 (Asc11 —Mouse Genome Informatics), Math1 (Atoh1 — Mouse Genome Informatics) and Ngn1 (Neurog1 — Mouse Genome Informatics) may serve to translate dorsoventral positional information into specific neural cell-type identity(Fode et al., 2000;Gowan et al., 2001;Lee et al., 2000;Lee et al., 1998). Accordingly, the expression of bHLH genes in a restricted pattern appears to be required for the generation of a diverse population of adult neurons. Loss-of-function mutations in these factors result in the loss of specific subsets of neurons and neonatal lethality(Ben-Arie et al., 1997;Bermingham et al., 1999;Bermingham et al., 2001;Fode et al., 1998;Guillemot et al., 1993;Ma et al., 1998). Overexpression of these neural bHLH factors appears to induce neuronal differentiation (Cai et al.,2000; Farah et al.,2000; Kim et al.,1997; Lee et al.,1995; Ma et al.,1996; Sun et al.,2001). In addition, substantial defects in neural development arise when the expression of Math1 is deregulated spatially,temporally or quantitatively (Helms et al., 2001; Isaka et al.,1999). Thus, the proper initiation, maintenance and termination of neural bHLH transcription factor expression are essential for normal neural development.

The bHLH factor Math1 is expressed transiently in proliferating neural precursors in multiple domains of the developing murine nervous system,including the dorsal hindbrain and neural tube, hair cells of the vestibular and auditory systems, mechanoreceptor (Merkel) cells of the hairy skin, and the cells of the external granule layer of the developing cerebellum(Akazawa et al., 1995;Ben-Arie et al., 2000;Ben-Arie et al., 1996;Helms and Johnson, 1998). Analysis of the Math1 gene has identified two discrete ∼500 bp sequences 3′ of the Math1-coding sequence that exhibit significant homology to sequences flanking the human ortholog HATH1(ATOH1 — Human Gene Nomenclature Database)(Helms et al., 2000). These sequence domains are sufficient to direct heterologous lacZ reporter activity to domains of native Math1 expression in transgenic mice. The activity of the Math1 enhancer in transgenic mice requires both a conserved E-box (binding site for bHLH factors) and Math1 expression(Helms et al., 2000). Similar autoregulation has been reported for the Drosophila homologatonal (Sun et al.,1998) and a related chick gene CATH5(Matter-Sadzinski et al.,2001). Thus, sustained expression of Math1, as with its homologs, appears dependent on previous Math1 expression. Little is known, however, about other factors controlling Math1 expression. Initiation of Math1 expression appears to be dependent on BMP signaling from the roof plate. Ablation of the roof plate in mouse embryos resulted in the absence of multiple dorsal markers, including Math1(Lee et al., 2000). In naive neural tube explants, application of BMP factors induced Math1expression (Alder et al.,1999). In chick neural tube, expression of constitutively active BMP receptor results in increased Math1 expression(Timmer et al., 2002). The direct effectors of regulation of Math1 expression by BMP signaling are unknown.

In a screen to isolate factors involved in the regulation of Math1expression, we have identified Zic1, a zinc-finger transcription factor that is also induced by BMPs in the neural tube(Alder et al., 1999;Aruga et al., 2002b;Aruga et al., 1994). It belongs to a family of related proteins including Zic2, Zic3, Zic4 and Zic5, all of which share homology to the Drosophila gene odd paired(Aruga et al., 1996a;Aruga et al., 1996b;Nakata et al., 2000), in addition to weaker homology to the Gli family of vertebrate genes and theirDrosophila homolog cubitus interruptus(Aruga et al., 1994). Loss-of-function and gain-of-function studies with Zic factors in the neural tissue of Xenopus, mouse, and chicken have suggested that Zic1 inhibits neuronal differentiation and maintains cells as progenitors(Aruga et al., 2002b;Brewster et al., 1998;Kuo et al., 1998;Mizuseki et al., 1998). Although Zic1 and Math1 are both expressed in the dorsal neural tube, and both are induced in response to BMP signaling, we have found that high expression of Zic1 and Math1 occurs only in distinct cells. We demonstrate that in the chick neural tube Zic1 represses Math1 expression. This repression blocks Math1 auto-activation and inhibits bHLH induced neuronal differentiation. Thus, BMP signaling induces both Zic1(Aruga et al., 2002b) and Math1(Alder et al., 1999;Timmer et al., 2002), setting up antagonistic pathways that must be in balance to form a spinal cord with the correct cell number and cell type.

The yeast one-hybrid screen was carried out using the Matchmaker one-hybrid kit (Clontech, Palo Alto, CA). The yeast strain YM4271 was used to generate a dual His/lacZ reporter strain as follows. Math1 enhancer region 898-1271 (AF218258) (Helms et al.,2000) was cloned in either one or four copies into the pHISi and pLACi vectors (details available upon request). The vectors were linearized and sequentially integrated into YM4271. Ectopic activity of either vector following integration was minimal (pHISi-Math required 15 mM 3-aminotriazole to suppress nonspecific growth). An E10.5 mouse embryo library (a gift from E. Olson) was screened. Putative Math1 enhancer-binding factors were isolated from His+, lacZ+ clones, transformed into DH5α E. coli and sequenced.

Math1/lacZ transgenes were constructed by PCR from theMath1 enhancer (Helms et al.,2000) (Accession Number, AF218258) and cloned into the BGZA reporter construct (Yee and Rigby,1993), which places the enhancer region 5′ to theβ-globin basal promoter, the lacZ-coding region and SV40 polyadenylation sequences. Specific enhancer fragments used and mutations within these fragments are shown in Fig. 5 and Fig. 1B. Tg22 also contains a mutation in a possible Zic1-binding site within the non-conserved regions between enhancers A and B. Templates for in vitro transcription coupled translation (TNT) were generated as follows. The truncated Zic1 construct was derived by PCR using theZic1-containing yeast one-hybrid clone and primers 5′ACCATGGGCCCACACGGCCATAC 3′ (sense) and 5′ GGGTCGGCATGTTTTGTTTC 3′ (antisense), and then subcloned into pSP64-TEN. Full-lengthZic1 was generated by hybrid PCR from a genomically derivedZic1 fragment corresponding to the 5′ region of the coding sequence, as well as the yeast one-hybrid clone used to generate the truncatedZic1 construct (details available upon request). All chick expression plasmids are in pMIWIII (Muramatsu et al.,1997). pMIWIII-^mycZic1 was constructed by cloning the full-length coding sequences in frame into pMIWIII-^mycNgn1, after the removal of the Ngn1-coding sequence(Gowan et al., 2001). The co-electroporation vector EGFP-N1 is from Clontech. pMIWIII-^myccontrol contains a Myc tag fused to a missense Ngn1(Gowan et al., 2001). The Math1 expression construct is from Gowan et al.(Gowan et al., 2001). All constructs were sequenced to confirm the lack of PCR incorporation errors.

Electrophoretic mobility shift assays (EMSA) were performed using protein generated via in vitro transcription coupled translation starting with the templates described above using the TNT SP6 kit (Promega, Madison, WI). EMSA was performed in 1×Zic1 buffer (Aruga et al., 1994), which contained 20 mM HEPES (pH 7.9), 10 mM MgCl₂, 50 mM KCl, 0.5 mM DTT, 0.1 mM EDTA, 10 μM ZnCl₂, 10% glycerol, 50 μg/ml poly dI/dC and 10⁵ cpm probe. Oligonucleotide probes were generated by annealing complementary oligonucleotides (see Fig. 3for probe sequence), followed by radiolabeling in 1×PNK buffer containing 50 ng annealed oligonucleotides, 50 mM Tris-HCl, 10 mM MgCl₂, 0.1 mM EDTA, 5 mM DTT, 0.1 mM spermidine, 50 μCiγ ³²P ATP (Amersham, Piscataway, NJ), 10 U polynucleotide kinase (Roche, Indianapolis, IN). Labeled oligonucleotides were purified over a G-50 sephadex column and quantified by scintillation analysis. The enhancer B (374 bp) probes were generated by adding 50 μCi α³²P dCTP to a standard PCR reaction with primers 5′ AAGCCGAGTGTACGTTTAGT 3′ and 5′ TAAATGTGGCCGCTCAGCTC 3′. The reaction was purified using High Pure PCR Product Purification kit (Roche) to reduce primer-dimer contamination, and quantified by scintillation analysis. Pre-binding reactions of unlabeled competitor oligonucleotides at ∼100-fold molar excess occurred for 15 minutes at room temperature. Binding reactions were performed for 30 minutes at room temperature, followed by electrophoresis on a prerun 5%-0.08% acrylamide-bisacrylamide non-denaturing 0.5×TBE gel for 3-5 hours at 20 mA constant current, followed by drying and exposure to X-ray film for several hours.

White leghorn eggs were obtained from SPAFAS (Norwich, CT) or Texas A&M(College Station, TX). DNA solutions (concentrations ranged from 1-3μg/γ) were injected into the lumen of the neural tube using a General Valve Corporation Picospritzer II (Fairfield, NJ) and pulled glass pipettes. Electroporations were performed using a BTX ECM 830 power supply and BTX Genetrodes electrodes (San Diego, CA) with 3-50 ms pulses at 25 V. Embryos used for lacZ reporter construct analysis were electroporated at stages 14-17, and harvested for β-gal activity assays 24-36 hours later as previously described (Timmer et al.,2001). Embryos used for analysis of ectopic expression of Zic1 were electroporated at stages 13-14 and harvested at 24 hours for eitherβ-gal activity or immunocytochemistry. Embryos used to assess contribution to neural crest were electroporated at stages 11-13 and harvested 48 hours later. All electroporations included the EGFP-N1 plasmid (Clontech)to mark cells that took up the DNA.

Mouse and chick embryos were dissected in 1×PBS on ice, rinsed, fixed in 2-4% formaldehyde for 2-4 hours at 4°C, rinsed with 1×PBS and sunk in 30% sucrose at 4°C overnight. Embryos were embedded in OCT compound (Tissue Tek) and cryosectioned at 30 μm. Slides were incubated in the appropriate dilution of primary antibody in PBS/1% goat serum/0.1% triton X-100, followed by either goat anti-rabbit or goat anti-mouse IgG, conjugated to Alexa Fluor 488 and 594 (Molecular Probes). Primary antibodies used for this study include: rabbit anti-Math1(Helms and Johnson, 1998),mouse anti-Math1 (Gowan et al.,2001), rabbit anti-Zic (gift of R. Segal), rabbit anti-β-galactosidase (5prime-3prime), mouse anti-c-myc (9E10 Santa Cruz Biotechnology), mouse anti-Mash1 (Lo et al., 1991) and mouse anti-islet1 (Developmental Studies Hybridoma Bank). Fluoresence imaging was carried out on a BioRad MRC 1024 confocal microscope. EGFP signal was imaged using a standard FITC filter. For detection of β-gal activity, chick embryos were dissected in room temperature 1×PBS, fixed for 30 minutes at room temperature in 2% formaldehyde,rinsed and stained in X-gal solution as described(Timmer et al., 2001).

In situ hybridization was performed as described(Gowan et al., 2001) using in situ probes for Cath1 (PCR generated), Cash1 (obtained from T. Reh), and chick Ngn1 and Ngn2 (obtained from D. Anderson).

In previous studies, a Math1 enhancer that resides ∼3.4 kb 3′ of the Math1-coding sequence was sufficient to direct restricted expression of a lacZ reporter to most domains ofMath1 expression in transgenic mice(Helms et al., 2000). This enhancer contains two discrete blocks of sequence (denoted enhancer A and B)that are highly conserved with sequence near the human Math1 ortholog, HATH1. To delineate important sequence elements further, we analyzed the Math1 gene of a more evolutionarily distant organism. We cloned the chick homolog ofMath1, Cath1, and identified genomic sequence 3′ of the coding region conserved with the Math1 enhancer (AF467292). Like humanHATH1, Cath1 contains two extensive blocks of homology with theMath1 enhancer, and this homology is located ∼1.7 kb 3′ of the Cath1 coding sequence. The Cath1 enhancer sequence is less conserved with Math1 (75% and 62% for the two regions) relative to the conservation between Math1 and HATH1 (91% and 87% for the two regions) (Fig. 1A). Notably, an E-box site, previously shown to be essential for Math1enhancer function in transgenic mice(Helms et al., 2000), and the Zic1 binding site described below, are conserved between all three species(Fig. 1B).

We then tested for functional conservation by assaying the activity of the chick Cath1 enhancer in mouse transgenic assays, and conversely,testing the activity of the mouse Math1 enhancer in the chick neural tube by in ovo electroporation. The sequences with highest homology between the chick and mouse enhancers were tested for activity. A 1.2 kbMath1 enhancer sequence (which corresponds to the region conserved with the chick enhancer) was tested using the BGZA reporter(Yee and Rigby, 1993) both in transgenic mice, and in the chick neural tube after electroporation. Expression of a 1.2 kb Math1/lacZ transgene in mouse E10.5 embryos resulted in restricted expression of lacZ in the dorsal neural tube(Fig. 1C,E) as had been seen previously with a longer sequence (Helms et al., 2000). Electroporation of this same construct in the chick neural tube (stage HH14-17) resulted in lacZ expression on the electroporated side of the neural tube in a similar dorsal pattern when assayed 24-36 hours after electroporation(Fig. 1F). The activity of this 1.2 kb Math1 enhancer in chick is identical to previously published results using a 1.7 kb enhancer sequence(Timmer et al., 2001). Conversely, we tested the activity of a 950 bp Cath1 sequence with the BGZA reporter in transgenic mice. This Cath1/lacZ construct resulted in lacZ expression in a dorsally restricted pattern in the neural tube of E10.5 embryos from the midbrain/hindbrain boundary caudally to the tail (Fig. 1D), identical to that seen with the comparable Math1/lacZ transgene(Fig. 1C). Thus, conserved sequences between Math1 and Cath1 reflect a functional conservation that suggests conserved mechanisms control Cath1/Math1transcription in chick and mouse.

To identify factors that interact with the Math1 enhancer, we initiated a yeast one-hybrid screen using a 374 bp region of the enhancer(nucleotides 898-1271) to screen an E10.5 mouse embryo library (a gift from Dr Eric Olson). Upon screening ∼1×10⁶ independent clones for binding to the 374 bp sequence, the zinc-finger transcription factor Zic1 was identified six times. This study addresses the possible role of this factor in the regulation of Math1 transcription.

Zic1 is a zinc-finger transcription factor originally identified based on enriched expression in the cerebellum(Aruga et al., 1994). Published reports on the expression of Zic1 are consistent with a role for this molecule in the regulation of Math1 expression. In mouse neural tissue, Zic1 mRNA is detected in the neural plate at E7.5(Nagai et al., 1997), in advance of the initiation of Math1 expression at E9.0(Akazawa et al., 1995;Ben-Arie et al., 1996). By E10.5, Zic1 expression is restricted to the dorsal third of the neural tube, including spinal and hindbrain regions that overlap the domain ofMath1 expression (Aruga et al.,1994; Nagai et al.,1997). Like Math1, Zic1 is also expressed in cerebellar granule cell progenitors from rhombic lip stages through postnatal stages(Akazawa et al., 1995;Aruga et al., 1994;Ben-Arie et al., 2000;Helms and Johnson, 1998;Nagai et al., 1997). The multiple overlapping domains of Zic1 and Math1 expression are consistent with Zic1 playing a role in the regulation ofMath1.

To compare the distribution of Math1 to the Zic factors at the single cell level, we used double immunofluorescence with a rabbit polyclonal antisera that recognizes multiple members of the Zic factor family (gift of R. Segal)and mouse monoclonal antibodies specific to Math1(Gowan et al., 2001). At E10.5 in mouse, Math1 is restricted to the dorsal neural tube in cells adjacent to the roof plate (Fig. 2B)(Akazawa et al., 1995;Ben-Arie et al., 1996;Helms and Johnson, 1998). Expression of Zic factors is broader, extending from the roof plate through the dorsal third of the neural tube with the highest expression within the roof plate region (Fig. 2A)(Aruga et al., 1994;Nagai et al., 1997). Unexpectedly, Zic expression appears to be diminished specifically in the dorsal domain where Math1 is expressed(Fig. 2C). Analysis at higher magnification reveals cells co-expressing low levels of both Zic and Math1,but cells co-expressing Zic and Math1 at high levels are not detected(Fig. 2D-F). This inverse expression pattern suggests a negative role for Zic1 in Math1transcription. Conversely, as seen in the Math1 mutant, the Zic expression pattern is uniform in the dorsal neural tube(Fig. 2G), suggesting Math1 may inhibit Zic expression as well.

To confirm and characterize the specific Zic1-binding site within theMath1 enhancer, we used electrophoretic mobility shift assays (EMSA). The Zic1 clones identified in the yeast one-hybrid screen did not contain a sequence encoding the full-length Zic1; they lacked the domain N-terminal to the Zn-finger domain. In all previous published DNA-binding studies with Zic factors, similarly truncated forms of the proteins were used (Aruga et al.,1996a; Mizugishi et al.,2001; Salero et al.,2001). We in vitro transcribed and translated this partialZic1 clone and tested it for protein/DNA interactions in EMSA. The enhancer sequence used to identify Zic1 in the yeast-one-hybrid screen does not contain any site resembling published Zic1-binding sites(Aruga et al., 1994;Mizugishi et al., 2001). We identified an oligonucleotide (Z-site) within the Math1 enhancer that could be bound by Zic1 at an apparent higher affinity than binding to the previously defined consensus (Aruga et al.,1994) (Fig. 3A,compare lanes 2 and 3). Addition of unlabeled wild-type oligonucleotides competed for this binding, while addition of unlabeled mutated oligonucleotides failed to compete (Fig. 3A, lanes 4, 5, and 6). This novel Zic1 binding site is located∼25 bp from the previously identified E-box site required forMath1 enhancer activity (see Fig. 1B). No additional Zic1 binding sites within this enhancer fragment were identified as shown using EMSA with the 374 bp Math1enhancer probe and Zic1 extract. A single complex, in addition to that seen with control extract, was detected (Fig. 3B, lane 2). This complex was lost when the identified Zic1 site was mutated (Fig. 3B, lanes 2,4, and 6). Thus, this enhancer sequence contains only one Zic1-binding site.

To characterize the novel Zic1-binding site further in the Math1enhancer, we performed EMSA with mutant oligonucleotide probes spanning the Z-site oligonucleotide. This analysis defined a 15 bp region required for Zic1 binding (Fig. 3C, consensus). EMSA using probes from the homologous human (HATH1) and chick (Cath1) regions revealed that the ability to interact with Zic1 is conserved among all three species (Fig. 3C). This consensus has some similarity (8 out of 12 nucleotides) to a recently published Zic1 binding site in the apolipoprotein E enhancer(Salero et al., 2001) and no significant similarity to the published Gli site. In conclusion, a novel Zic1 binding site conserved between multiple species is present in theMath1 enhancer.

To test whether Zic1 functions in vivo in regulating Math1/Cath1expression, we examined Cath1 expression when Zic1 was overexpressed in the chick neural tube. A construct containing the full-lengthZic1-coding sequence fused to five N-terminal Myc tags in the pMiWIII vector was electroporated into stage HH13-14 chick neural tubes and assayed for Cath1 expression by immunofluorescence at 24 hours post-electroporation (HH20-21). Neural tubes ectopically expressing^mycZic1 had a dramatic decrease in the number of cells expressing Cath1 (Fig. 4B). This loss of Cath1-positive cells was not seen with neural tubes electroporated with a control expression vector (Fig. 4A). These data support a negative role for Zic1 in the regulation of Cath1 expression. Electroporation of ^mycZic1 after normal initiation of Cath1 expression (HH18) had no effect on Cath1 expression (data not shown). Thus, Zic1 appears to repress expression of Cath1 but this repression is stage dependent and occurs during early stages of Cath1 expression.

As suggested by the binding activity described above, Zic1 also represses alacZ reporter transgene driven by the Math1 enhancer(nucleotides 1-1365). The Math1 enhancer directs lacZexpression to the dorsal neural tube when co-electroporated into the chick neural tube at HH14-17 with a control vector, pMiWIII expressing five Myc tags and assayed at 24 hours (Fig. 4C, n=6). By contrast, co-electroporation of theMath1/lacZ transgene with the ^mycZic1 expression construct dramatically reduces the activity of the enhancer(Fig. 4D, n=6). These data support the conclusion that Zic1 represses Math1/Cath1expression and this activity may be through the 1365 bp Math1enhancer.

Zic1 represses Cath1 and Math1/lacZ transgene expression in the chick neural tube. This suggests that mutation of the Zic1 binding site in theMath1 enhancer might result in increased enhancer activity due to loss of Zic1 binding. To test this, we mutated or deleted specific nucleotides to disrupt Zic1 binding, and tested enhancer activity in the chick electroporation assay. Electroporation of multiple Math1 enhancer/lacZ constructs that include the Zic1 site into chick embryos resulted in consistent restricted expression of lacZ in the dorsal neural tube (Fig. 1F,Fig. 4C andFig. 5; Tg18 and 20)(Timmer et al., 2001). Contrary to our prediction, mutation (Fig. 5, Tg19 and 22) or deletion(Fig. 5, Tg21) of the Zic1 site within the context of the Math1 enhancer completely eliminated expression of the construct. Thus, the Zic1 site is necessary for enhancer activity in the chick neural tube, suggesting a positive role for this site in enhancer function. Given our findings that Zic1 is acting to repressCath1 expression, the loss of enhancer activity when Zic1-binding site is lost suggests that Zic1 may either compete for this site with an activator, or that Zic1 also has a role as an activator, but only in the presence of a co-factor.

We have previously demonstrated that the Math1 enhancer requires Math1 for activity and contains an essential E-box element (the binding site for Math1) supporting the involvement of autoregulation in Math1expression (Helms et al.,2000). We have found a similar requirement for the E-box for activity of the enhancer in the chick neural tube(Fig. 5, Tg23). Further data supporting positive autoregulation is seen by the increase in expression of endogenous Cath1 upon electroporation of Math1 in the chick neural tube (Fig. 6A, the endogenous increase in Cath1 was distinguished from the exogenousMath1 using a Cath1-specific probe for mRNA in situ analysis). The increase in Cath1 was mostly seen in the dorsal neural tube, consistent with previous results showing context dependent activity of Math1 in this assay (Gowan et al.,2001). An even more dramatic increase was seen in the expression of the Math1/lacZ transgene upon co-electroporation withMath1 (Fig. 6B). Note that the expression is throughout the DV axis, in contrast to the normal dorsal-restricted expression of the Math1/lacZ transgene(Fig. 1F,Fig. 4C,Fig. 5B). These data further support our previous result that demonstrated the involvement of positive autoregulation in controlling Math1 expression.

To address the interplay of autoregulation with the repression caused by Zic1, we electroporated Math1 and Zic1 together, and assayed for expression of both Cath1 and the Math1/lacZ transgene. The autoactivation of Cath1 by Math1 is clearly blocked by^mycZic1 (Fig. 6C). Furthermore, the dramatic increase in Math1/lacZ transgene expression seen with electroporated Math1 was completely blocked by co-electroporating^mycZic1 (Fig. 6D). Similar results were seen with embryos electroporated at HH14(Fig. 6) or at the later HH18 stage (data not shown). Thus, one mechanism for Zic1 repression ofMath1/Cath1 expression is by blocking positive autoregulation.

Roles for Math1 and Zic1 in the developing dorsal neural tube have been suggested previously from loss-of-function studies in mouse(Aruga et al., 2002b;Bermingham et al., 2001;Gowan et al., 2001) and overexpression studies in Xenopus(Brewster et al., 1998;Kim et al., 1997;Kuo et al., 1998;Mizuseki et al., 1998). Studies of the Math1 mutant mouse demonstrated that Math1 is required for the formation of a specific population of interneurons(Bermingham et al., 2001;Gowan et al., 2001), and inXenopus, Xath1, drives ectopic neuronal differentiation(Kim et al., 1997). By contrast, overexpression of Zic factors in Xenopus have suggested that Zic factors mediate neural competence, but also induce dorsal cell fates such as neural crest. The consequences for dorsal neural tube neuronal differentiation are not clear, as the previous referenced reports suggest that dorsal cell types may be induced, repressed or not affected(Brewster et al., 1998;Kuo et al., 1998;Mizuseki et al., 1998). Studies of the Zic1 mutant mouse suggest that Zic1 normally represses neuronal differentiation and maintains cells as neural progenitors(Aruga et al., 2002b). Electroporation of Math1 and Zic1 into the chick neural tube provides another paradigm to address the function of these factors in dorsal neural tube development.

Electroporation of a GFP expression construct alone at HH11-12 and analysis 48 hours later illustrates the distribution of cells that have taken up the DNA. Many GFP-expressing cells are found in the dorsal root ganglia (DRG), a neural crest derivative, on the electroporated side(Fig. 7A,B). A few cells are found in the DRG opposite to the electroporated side, suggesting that neural crest cells can cross the midline. Within the neural tube, the GFP is distributed throughout the ventricular zone, presumably in neural progenitor cells, as well as cells that are found laterally where the differentiated neurons reside (Fig. 7A,B;asterisk marks the midline). By contrast, when a Math1 expression construct is co-electroporated with the GFP expression construct, cells expressing GFP are rarely found in either DRG (Fig. 7C,D). Furthermore, within the neural tube, the GFP is no longer evenly distributed within ventricular zone but rather is mostly found lateral to this progenitor domain (Fig. 7C,D, asterisk indicates the midline). This effect is seen within 24 hours of the electroporation (data not shown). The distribution of the GFP-expressing cells in the presence of Math1 suggests that Math1 promotes neuronal differentiation within the neural tube in a manner that prevents these cells from adopting the neural crest fate.

Based on previous overexpression studies with Zic factors inXenopus (Brewster et al.,1998; Kuo et al.,1998; Mizuseki et al.,1998), electroporation of Zic1 in the chick neural tube might be expected to increase the number of cells joining the neural crest lineage and inhibit neuronal differentiation in the neural tube. We found that indeed,neuronal differentiation is blocked as illustrated by the lack of GFP expressing cells lateral to the ventricular zone in the neural tube(Fig. 7E,F). This is consistent with Zic1 mediating the repression of the neuronal differentiation factor Math1, and similar to recent reports expressing chick Zic1 in the chick neural tube (Aruga et al., 2002b). However, the cells did not preferentially choose the neural crest lineage, but seem to actively avoid it, as seen by the absence of GFP expressing cells in the DRG. Rather, all GFP expressing cells appear to be restricted to the ventricular zone along the midline. This suggests that Zic1 may function to maintain the cells in a progenitor state. Furthermore, the midline location and change in morphology of the electroporated cells suggests these progenitors have additional perturbations and may be blocked in a particular phase of the cell cycle.

The data presented here, and in previous studies, suggest that Zic1 plays a role in maintaining cells as neural progenitors, and thus may play a role in the timing of differentiation. We have shown that Zic1 represses expression of the neuronal differentiation factor Math1. However, Zic1 is expressed not only in the region encompassing the Math1 domain, but also extends ventrally to the dorsal regions of expression of three other bHLH transcription factors: Ngn1,Ngn2 and Mash1 (Gowan et al.,2001). To examine whether the Zic1 repression is specific to Cath1, or whether repression of bHLH transcription factors is a general mechanism for inhibiting differentiation, we analyzed expression of chickNgn1, chick Ngn2 and Cash1 in the^mycZic1 electroporated chick neural tubes. As seen previously with Cath1 immunostaining (Fig. 4),Cath1 mRNA is repressed relative to the non-electroporated side when^mycZic1 is electroporated (Fig. 8A). Similarly, Cash1 expression is repressed with^mycZic1 (Fig. 8B). By contrast, Ngn1 and Ngn2 appear to be unaffected (or even slightly induced) by ^mycZic1(Fig. 8C,D). Thus, Zic1 differentially regulates these bHLH factors at this stage in the chick neural tube.

We used double label immunofluorescence to examine if Zic factors and Mash1 are co-expressed, or whether they are mutually exclusive like Zic factors and Math1. The Mash1 and Zic domains of expression overlap in the dorsal half of the Mash1 expression domain (Fig. 8E). Within the overlapping domain, individual cells co-express Mash1 and Zic factors (Fig. 8F-H). Thus, in contrast to the direct repression of Math1/Cath1 by Zic1, the repression of Cash1/Mash1 is likely to be through a distinct indirect mechanism.

During development of the mammalian nervous system, regulation of the spatial and temporal transcription pattern of neural bHLH transcription factor genes is required for the correct differentiation of neural progenitors. We have begun to probe the mechanism of this transcriptional regulation by identifying enhancers in the Math1 gene required for tissue-and temporal-specific expression, and using these enhancer sequences to identify transcription factors that play a role in this process. Using this strategy,we identified Zic1 as a repressor of Math1 expression. Zic1, like Math1, is induced by BMPs. Thus, BMP signaling activates pathways that both promote and inhibit neuronal differentiation. Math1 promotes differentiation,and Zic1 suppresses it, in part, through inhibition of bHLH genes such asMath1. The balance between these two BMP-induced pathways is likely to be central in obtaining the correct composition of dorsal cell-types in the spinal cord.

The inverse relationship in expression of Zic factors and Math1 in the dorsal neural tube is consistent with an inhibitory role for Zic1 inMath1 expression. In ovo experiments analyzing Zic1 function demonstrated that overexpression of Zic1 can repress both the endogenousCath1 and Cash1 genes, as well as a Math1enhancer/lacZ reporter construct. Expression of related genes encoding the bHLH factors Ngn1 and Ngn2 were not repressed. These data suggest that Zic1 does not mediate global repression of neural differentiation, but instead is directed at distinct neuronal subtypes, in particular those descendant from Math1/Cath1- and Mash1/Cash1-expressing cells.

The data presented here strongly support a model in which Zic1 directly represses Math1 expression. This repression could function to block initiation of Math1 expression, to modulate the autoregulation phase of expression, or both. It is intriguing that the Zic1 binding site and E-box site required for autoregulation are separated by only 25 bp in the Math1 enhancer. It is possible that the two sites cannot be occupied at the same time, thus providing an explanation for why Zic1 was unable to repress endogenousCath1 expression when Zic1 was added after Cath1 was already present. Localized repressive effects on the chromatin catalyzed by Zic1 could undermine the ability of factors binding the E-box to activate transcription. Consistent with this possibility, factors related to Zic proteins, the Gli factors, recruit components of the mouse Sin3/histone deacetylase complexes by interaction with the Fused kinase (Cheng and Bishop, 2002). Gli and Zic factors interact in specific cell contexts, and this interaction can be direct, as shown in immunoprecipitation and yeast two-hybrid experiments (Koyabu et al., 2001). These data suggest that a combination of Zic/Gli factors (or possibly Zic alone) could act to recruit complexes to chromatin that repress transcription activity, suppressing activation of the autoregulatory E-box in the Math1 enhancer, and thus block or limit the establishment of a Math1 expressing domain in the dorsal neural tube.

Although it is clear that Cath1 is directly repressed by Zic1, the repression of Cash1 by overexpression of Zic1 may be indirect, based on the relatively abundant co-expression of these two molecules in the dorsal neural tube. An increase in expression of the neurogenic repressor Notch1 and its effector Hes1 was seen in Zic1 overexpression paradigms(Aruga et al., 2002b),providing a possible link to the indirect repression of Cash1expression. The observation that Cash1 but not Ngn1/2 was repressed by Zic1 could reflect a difference in how these genes are regulated by Notch signaling. Such differences in sensitivity of bHLH genes to inhibition by Notch signaling have been seen previously(Cau et al., 2002;Chitnis and Kintner, 1996;Lo et al., 2002).

The conclusion that Zic1 acts to repress Math1 expression was confounded by our findings that the Math1 enhancer lost activity in the chick neural tube when the Zic1 site was mutated. Loss of activity of the mutated enhancer suggests the Zic1-binding site is also required for activating Math1 expression. We propose alternate mechanisms by which Zic1 may temporally modulate Math1 transcription through binding to the Math1 enhancer. First, Zic1 may compete with a positive regulator for binding to the Math1 enhancer. Mutation of nucleotides critical for Zic1 binding to the Math1 enhancer compromises the activity of the enhancer in assays in vivo, suggesting that nucleotides bound by the putative activator of Math1 transcription and Zic1 are similar. Alternatively, a positive factor interacting with Zic1, or post-translational modifications of Zic1, may convert it to an activator, explaining the requirement for the Zic1-binding site for enhancer activity. Zic genes share homology to the Gli family of transcription factors, which can alternately act as repressors or activators of transcription depending on whether the repressor domain of the protein has been proteolyticly removed(Ruiz i Altaba, 1999;Sasaki et al., 1999). Zic factors may have a similar function, as they have been shown to either repress or activate expression of reporter constructs containing Zic/Gli-binding sites in a cell-type-specific manner (Brewster et al., 1998; Mizugishi et al., 2001; Salero et al.,2001; Yang et al.,2000; Kuo et al.,1998). In the Zic1 mutant, the number of Math1-expressing cells in the dorsal neural tube was decreased at E11.5(Aruga et a., 2002b). This is consistent with a role for Zic1 in activating Math1 expression. However, this interpretation is complicated by the fact that in the absence of Zic1 premature neuronal differentiation occurs. Thus, the progenitor pool will be diminished and result in fewer cells that are competent to expressMath1, particularly when looking at the later stages of development of this progenitor population. Further research into regulation of Zic1 protein function will be required to reveal if Zic1 can act as a bifunctional protein.

What is the importance of repression of Math1 expression by Zic1?One possibility is that a delay in Math1 expression would allow expansion of the progenitor population needed to generate the correct number of cells prior to cells exiting the cell cycle and differentiating. Because Math1 is a regulator of neuronal differentiation of dorsal neuronal cell types(Bermingham et al., 2001;Gowan et al., 2001), a Zic1-mediated delay in the onset of Math1 expression would allow dorsal progenitors to proliferate to the needed population size. When the repression of Math1 by Zic1 is subsequently overcome, possibly by the expression of a positive factor that shares the binding site with Zic1, or a modification of Zic1 itself, and Math1 expression is initiated, the differentiation of dorsal neuronal cell types results. Such a model is supported by multiple data generated in Xenopus, where ectopic expression of Xzic factors result in the inhibition of neuronal differentiation (Brewster et al.,1998; Kuo et al.,1998; Mizuseki et al.,1998). A role for Zic1 and Zic2 in inhibition of neuronal differentiation has also been suggested in mice, as in Zic1 andZic2 mutants, premature neuronal differentiation was detected(Aruga et al., 2002a;Aruga et al., 1998;Aruga et al., 2002b). The role for Zic1 in controlling the timing of differentiation is similar to functions attributed to Notch. The increase in expression of Notch and its effector Hes1 when Zic1 was overexpressed suggests this function of Zic1 in inhibiting differentiation may be mediated by the Notch pathway. Furthermore, the negative regulation of Zic factors by Math1, suggested by the increased expression of Zic factors in the Math1 mutant, may contribute to the progression of neural progenitors towards the differentiated state.

An additional consequence for Zic1-based repression of Math1expression may be to allow early-born cells of the dorsal neural tube to differentiate into a neural crest cell fate. Cells of the neural crest begin to migrate from the dorsal regions of the neural tube shortly after fusion of the neural tube, continuing up to ∼E9.0-E9.5 in the mouse. This latter time period coincides with the induction of Math1 expression in the dorsal neural tube. Electroporation of Math1 into the chick neural tube (prior to native Cath1 induction) prevents transfected cells from accumulating in the DRG, consistent with Math1 acting to bias cells toward neuronal cell fates and blocking the adoption of neural crest fates. Zic repression of Math1expression may allow the generation of neural crest cells early in embryonic development. Overcoming this repression allows the induction of Math1expression and subsequent generation of dorsal interneurons. In conclusion,the role of Zic1 may be to control the timing of the initiation ofMath1 expression, to modulate the number of cells that take on a neural crest fate versus those remaining in the neural tube that will adopt a dorsal interneuron fate.

The regulation of Math1 expression is complex in that it both integrates signals that pattern the neural tube and responds to signals that initiate neuronal differentiation. Thus, the spatial and temporal components of the Math1 expression pattern must be tightly controlled. The size of the conserved enhancer suggests that the integration of spatial and temporal signals may occur at the level of the enhancer. Identification of autoregulation (Helms et al.,2000), cross-inhibition between neural bHLH classes(Gowan et al., 2001) and the role of Zic1 in regulating different aspects of the Math1 expression pattern provide inroads into our understanding of these processes. Further experiments should be able to address how BMP signaling functions inMath1 regulation, and how it relates to Zic1 function. In addition,the studies to date have been focused on Math1 expression in the dorsal neural tube at E10.5. Will the same regulatory relationships function in the other domains of Math1 expression including the EGL of the cerebellum, and sensory cell progenitors in the inner ear and skin? Many transcription factors and signaling pathways functioning in the dorsal neural tube are present in these tissues as well, suggesting that common themes in regulation are likely.

We gratefully acknowledge the critical comments and discussions of this work with Drs R. MacDonald and R. M. Henke. We thank Dr R. Segal for anti-Zic antisera. We are particularly grateful for the assistance of Dr L. Niswander for initiation of the chick experiments in her laboratory, and the technical assistance of Andrew Abney for transgenic mouse production. This work was funded by grants to JEJ from the Muscular Dystrophy Association, March of Dimes and National Institutes of Health HD37932 and NS32817.