her3, a zebrafish member of the hairy-E(spl) family, is repressed by Notch signalling

Signalling mediated by Notch is a central component of a variety of developmental processes in invertebrates and vertebrates. During lateral inhibition in Drosophila, Delta binds to Notch on adjacent cells and determines the cleavage of its intracellular domain, which translocates into the nucleus (Hsieh et al.,1996; Lecourtois and Schweisguth, 1998; Struhl and Adachi, 1998). Here, it associates with Suppressor of Hairless[Su(H)], a member of the CSL (RBP-Jκ) family(Schweisguth and Posakony,1992), bound to its DNA recognition sequence, and forms a complex that activates the transcription of downstream genes(Jennings et al., 1994; Bailey and Posakony, 1995; Lecourtois and Schweissguth,1995), most notably those of the E(spl)-C(Knust et al., 1987). These latter genes encode transcriptional repressors of the bHLH/WRPW family(Klämbt et al., 1989; Delidakis and Artavanis-Tsakonas,1992; Knust et al.,1992), which in conjunction with Groucho(Paroush et al., 1994; Dawson et al., 1995; Fisher et al., 1996; Giebel and Campos-Ortega,1997) suppress the expression of transcriptional activators,including the proneural genes (Oellers et al., 1994; Van Doren et al.,1994; Singson et al.,1994; Tata and Hartley,1995; Heitzler et al.,1996; Nakao and Campos-Ortega,1996). A regulatory feedback loop modulates Deltaactivity by regulating its transcription(Haenlin et al., 1994; Hinz et al., 1994; Kunisch et al., 1994; Heitzler et al., 1996). Proneural proteins activate the transcription of Delta by binding to specific sites in its promoter (Kunisch et al., 1994). Consequently, the amount of proneural protein contained in a given cell determines the amount of Delta protein produced and,ultimately, the efficacy with which that cell activates Notch in neighbouring cells. The net result of these interactions is that only single cells within cell clusters take on the neural fate, while adjacent cells become epidermoblasts (for a review, see Simpson,1997).

The Notch signalling pathway is similarly organised in vertebrates (for a review, see Lewis, 1996). As in Drosophila, one of the functions of the pathway is to select individual cells for specific fates from groups of initially equivalent cells(Chitnis et al., 1995; Henrique et al., 1995; Chitnis and Kintner, 1996; Dornseifer et al., 1997; Wettstein et al., 1997; Appel and Eisen, 1998; Appel et al., 2001; Haddon et al., 1998; Takke et al., 1999). Injections of mRNA encoding variants of components of the Notch pathway have provided evidence for a regulatory feedback loop, organised similar way to that described for Drosophila, in both Xenopus and zebrafish(Wettstein et al., 1997; Takke et al., 1999). Activation of Notch receptors leads to activation of E(spl)homologues (Jarriault et al.,1995; Tamura et al.,1995; Hsieh et al.,1996; Kopan et al.,1996; Wettstein et al.,1997; Schroeter et al.,1998; Takke and Campos-Ortega,1999; Takke et al.,1999) mediated by CSL proteins(Wettstein et al., 1997). Thus, in mice, a loss-of-function mutation in the RBP-Jκ gene(Rbpsuh – Mouse Genome Informatics), the Su(H)homologue, leads to repression of Hes5 and, as a consequence, to upregulation of the proneural gene Math4a (Neurog2 – Mouse Genome Informatics) (de la Pompa et al.,1997). In zebrafish, this feedback loop operates on the proneural gene neurogenin 1 (neurog1; previously known as ngn1) (Blader et al.,1997), the product of which binds to E-boxes in the promoter of the Delta homologue deltaD, and activates its transcription(Hans and Campos-Ortega,2002).

We describe here a new zebrafish E(spl) homologue, her3,the expression of which is restricted to neural territories, where it represses transcription of the proneural gene neurog1. Gel retardation assays show that Her3 binds specifically to N-boxes in the promoter regions of neurog1 and of her3, thus contributing to its own regulation. The function of her3, like that of other members of the E(spl) family, depends on Notch signalling. However, her3 differs from the other family members in that its transcription is repressed rather than activated by Notch1a signalling.

Zebrafish embryos were obtained from spontaneous spawnings. Adult fish were kept at 28.5°C on a 14 hour light/10 hour dark cycle. The embryos were staged according to Kimmel et al. (Kimmel et al., 1995).

PCR using degenerate primers was performed on reverse-transcribed total RNA from zebrafish embryos at the 90%-epiboly to two-somite stage. PCR fragments encoding peptides with similarity to Drosophila bHLH proteins were used to screen a zebrafish cDNA library prepared in λZAP (Stratagene)from RNA isolated from 3-15 hour zebrafish embryos (gift from C. Fromental-Ramain and P. Chambon, Strasbourg). The GenBank Accession Number for the her3 cDNA sequence is X97331. A genomic her3 clone (15 kb) was obtained from a genomic DNA library (`Easy-to-handle eukaryotic genomic library' (zebrafish) Mo Bi Tec, Göttingen). The Accession Number for the her3 genomic sequence is AY277702. The upstream sequence was subcloned into pBsGAL4 (Scheer and Campos-Ortega, 1999). The plasmid pBs2xMARher4 was generated by excision of the Notch1a:intra coding sequence from pBs2xMAR notch1a-intra with SmaI(Scheer and Campos-Ortega,1999). The gap was filled with the EcoRI digested and blunt-ended coding sequence of the her4 cDNA(Takke et al., 1999). To generate stable transgenic lines, plasmid DNA preparations and injections were carried out following the procedures described by Scheer and Campos-Ortega(Scheer and Campos-Ortega,1999).

pCS2+her3:gfp was made by PCR using the plasmid pCS2+her3diluted 1:10 as template (see her3 for and her3 rev primers in Tables S1-S3). The amplified 680 bp fragment was cut with EcoRI and BamHI,cloned in the pCS2+EGFP vector previously cut with the same enzymes and sequenced.

Eight constructs encoding Her3 variants and two Her3 fusion proteins were generated by PCR using different 5′ and 3′ primers, and cloned into the pCS2+ (Turner and Weintraub, 1994). The coding sequence was amplified from the her3 cDNA using her3 specific primers. The primers used for constructs with terminal deletions (see Tables S1-S3) provided an artificial ATG for the construct her3ΔN, an artificial WRPW motif and stop codon for construct her3ΔC, and an artificial stop codon for construct her3ΔWRPW. PCR products were digested with the appropriate restriction enzymes and cloned into the pCS2+ vector. In order to remove the endogenous stop codon to permit the generation of the her3 fusion proteins, the coding sequence of her3 was amplified with the primers SP6 and her3 fusion (including a BamHI site), digested with EcoRI and BamHI, and cloned into the EcoRI/BamHI sites of the vectors pCS2+VP16 and pCS2+eng, which supply the sequences encoding the VP16 transactivation and the engrailed repressor domains, respectively, in the correct reading frame. The primer combinations used for the generation of the constructs and fusion proteins are listed in Tables S1-S3. Capped RNA was synthesised in vitro by transcription with SP6 polymerase from the constructs described above, or from a pCS2-nuclear β-Galactosidase(nβgal) template DNA, using a Message Kit from Ambion.

The RNA was injected in a volume of 5 nl into one of the first two blastomeres. In most cases, lacZ mRNA was co-injected at concentrations previously shown to be innocuous for the zebrafish embryo(Takke et al., 1999).β-Galactosidase was detected by antibody staining.

Morpholino-modified antisense oligonucleotides directed against her3 (Gene Tools; MO her35′-TGCAGCCATTGTCCTTAAATGCTCA, 2 blocker 5′-TTAAAAAATCCAGATGAATAAGGAC-3′) and the mismatch morpholino 5′-TGGAGGCATTGTGCTTAAATCCTGA-3′ were injected at the one- to four-cell stage at a concentration of 50-200 μM in 1×Danieau(Nasevicius and Ekker, 2000). Morpholinos were injected together with 0.2% Texas Red in a total volume of 5-10 nl. As an additional control, mRNA encoding a her3-gfp fusion was co-injected with the morpholinos.

Total RNA was extracted from 50 wild-type embryos, or 50 embryos injected with MO her3, using the RNA-Clean™ System (Angewandte Gentechnologie Systeme GmbH). Before precipitation, the RNA was treated with 2U of DNase (Boehringer Mannheim) for 30 minutes. Two μg of RNA was reverse transcribed using Superscript-RT (Gibco-BRL) and 100 ng of random hexamers(Boehringer Mannheim) in a 20 μl reaction, and 0.5 μl of this reaction was subjected to PCR. As an internal control, we used primers that amplified a 400 bp fragment of the gene for elongation factor e-IF4a. After 3 minutes at 95°C, amplification was carried out for 1 minute at 95°C, 1 minute at 58°C and 1 minute at 72°C (27 cycles for her3 and 28 for neurog1), with a final extension step for 10 minutes at 72°C. The primers are given in Tables S1-S3.

Protein preparation followed the protocol described previously(Chang et al., 1997). For each experimental determination, three lanes were loaded with increasing concentrations of protein (refer to the legends of Figs 3, 4 and 7). The specificity of binding was tested in competition assays with unlabelled oligonucleotide. Binding reactions were performed in the presence of oligonucleotides containing N boxes, one of the DNA sequences recognised by E(spl) proteins (consensus CACNAG) (Sasai et al., 1992; Tietze et al., 1992; Oellers et al., 1994), from the promoter region of her3 and neurog1(Blader et al., 2003). Band-shift assays were carried out according to Fried and Crothers (Fried and Crothers, 1984a; Fried and Crothers,1984) and Hendrickson and Schleif(Hendrickson and Schleif,1984). The sequences of the oligonucleotides used in binding assays were as follows (N-boxes are in bold, mutations are underlined).

• NP wt 5′ AAT TCC AAG CTC ACA AGC TCA CAC GAG CTG ATT G 3′

• Mut 1+2 5′ AAT TCC AAG CTC CAT GGC TCA CCA TGG CTG ATT G3′

• Mut 1 5′ AAT TCC AAG CTC CAT GGC TCA CAC GAG CTG ATT G 3′

• Mut 2 5′ AAT TCC AAG CTC ACA AGC TCA CCA TGG CTG ATT G 3′

• N1 5′ AAT TCT GAT TGG ATG TCC AGC AGA AAG TAT GGA TG 3′

• N2 5′ AAT TTG CAT TTT CAC CCC ACA CGA CCG AGG TTT CA 3′

• N1 mut 5′ AAT TCT GAT TGG ATG TCA CGC ACA AAG TAT GGA TG 3

• N2 mut 5′ AAT TTG CAT TTT CAC CCA CCA CCA CCG AGG TTT CA 3′

Hybridisation of digoxigenin-labelled RNA probes to embryo whole mounts was performed as described previously (Bierkamp and Campos-Ortega, 1993). Embryos injected with RNA were prepared for in situ hybridisation and for antibody staining, as described by Dornseifer et al. (Dornseifer et al.,1997). For sectioning, embryos were embedded in Araldite(Serva).

her3 was identified among PCR fragments obtained using degenerate primers directed against the regions of the E(spl)-C genes encoding the bHLH domain and the C terminal WRPW motif. The fragments were subcloned, sequenced and used as probes to screen a zebrafish cDNA library. The her3 cDNA 13.2.1.4 comprises a 687 bp open reading frame encoding a protein of 229 amino acids with all the features characteristic of the Drosophila hairy/E(spl) family. The her3 product shows the highest degree of similarity overall to hes3 (54%) and to the Drosophila mδ protein (54%). Within the bHLH domain this similarity increases to 86% for hes3 and 69% for mδ (Bestfit,GCG programme).

Whole-mount in situ hybridisation revealed that her3 RNA is expressed only in neural territories during embryogenesis(Fig. 1). Expression was first detected at about 30% epiboly in a coherent patch of cells within the dorsal region of the epiblast, in a region corresponding to the prospective anlage of the neural plate (Fig. 1A) (Woo and Fraser, 1995). At 80% epiboly, transcripts disappear from the medial region, splitting the primary domain into two(Fig. 1B,C). By the tail-bud stage, each of these has evolved into two extended longitudinal expression domains, which progressively separate from each other during the formation of the first somites. One of the domains is located rostrally and will eventually split into several groups of her3-positive cells distributed throughout the mesencephalon and rhombencephalon(Fig. 1E-H). Double in situ hybridisation using probes for egr2b and her3, pax2a and her3, or her5 and her3(Fig. 1I-L) show that the anterior margin of the her3 domain corresponds to the anterior margin of the mesencephalic primordium (Lun and Brand, 1998) extending to rhombomere 5. At the tail-bud stage, the her3 transcription domain ends cranially at the anterior border of the her5 domain, which itself initially extends throughout the midbrain/hindbrain anlage (Müller et al., 1996; Bally-Cuif et al.,2000), but is later restricted to the so-called intervening zone at the midbrain/hindbrain boundary (Geling et al., 2003). Therefore, later in development, the cranial margin of the her3 transcription domain continues to define the anterior margin of the mesencephalon, whereas the her5 domain is located further caudal. However, the her3 transcription pattern within the mesencephalic/rhombencephalic region is very dynamic and no attempt was made to define the different groups of her3-transcribing cells in this region.

The caudal expression domain occupies the intermediary regions of the neural plate, starting at the level of the presumptive cervical spinal cord,and consists of a contiguous patch of cells(Fig. 1D-H). Comparison of in situ hybridisation with neurog1 and her3 probes shows that,within this region, transcription of neurog1 is restricted to those regions of the neural plate in which her3 is not expressed(Fig. 1D′-F′). Although double in situ hybridisation with neurog1 and her3failed to produce reliable results, the comparison of in situ hybridisation with each single probe strongly suggests that the two genes are expressed in mutually exclusive domains both in the developing spinal cord and in the mesencephalic/rhombencephalic territories. During later stages of neurulation, her3 transcripts vanish from most of the cells in the anterior part of the spinal cord, becoming restricted to individual cells, while a contiguous expression domain persists in caudal regions, where neurulation is still in progress (Fig. 1E-H).

To characterise the regulatory region of the her3 gene, a 4.7 kb segment of the 5′ upstream DNA was fused to the GAL4 coding sequence,and transformed into the germline of wild-type embryos. Three independent chromosomal insertions were recovered, of which only one was able to transactivate UAS:notch1a-intra. In situ hybridisation with a gal4 cDNA probe showed that the gal4 transcription pattern in the transgenic embryos reflects, with minor deviations, the expression of the endogenous her3 gene (Fig. 2A-C). Thus, the expression of the gal4 transgene in mesencephalic and rhombencephalic regions is slightly delayed relative to that of her3, and the density of gal4 transcripts is higher than in the case of endogenous her3 RNA. The transgene is also expressed in the otic placodes, unlike the endogenous her3 gene. It is not clear whether these differences are due to position effects. For our present purposes, though, it is important to note that the 4.7-kb DNA that drives transcription in the intermediary region of the neural plate at the level of the spinal cord contains two high-affinity N boxes and a single high-affinity Su(H) binding site (TGTGAGAA) (Bailey and Posakony, 1995; Lecourtois and Schweissguth, 1995; Rebeiz et al., 2002). There are no low-affinity binding sites for E(spl)-related proteins in this DNA fragment.

The higher density of the gal4 mRNA in the transgenic embryos might be explained by higher stability relative to the endogenous her3 mRNA. Another possible explanation, however, is that regulatory elements that suppress transcription within medial and lateral regions of the neural plate in response to Notch-dependent signalling(Haddon et al., 1998; Takke et al., 1999) are missing in the 4.7 kb genomic segment used for the transgene. To test this possibility, the her3::gal4 transcription pattern was analysed in embryos that had been injected with mRNA encoding the constitutively active Notch receptor, Notch1a-intra. In these embryos, gal4 transcription was suppressed (Fig. 2D,E). This result indicates that the 4.7 kb genomic fragment does indeed respond to Notch signalling, thus supporting the hypothesis that the gal4 mRNA is more stable than her3 transcripts.

Embryos injected at the two-cell stage with her3 mRNA encoding the full-length protein were collected at the one- to three-somite stage and tested by in situ hybridisation for the expression of various other genes(neurog1, deltaA, islet1 and elavl3). After injection of her3 mRNA, a large proportion of embryos showed a pronounced asymmetry in the neural plate, which was considerably broader within theβ-galactosidase-expressing territory(Fig. 3, Table 1). The same effect has been reported after misexpression of a Xenopus Notch variant(Coffman et al., 1993), and following misexpression of deltaD(Dornseifer et al., 1997) and her4 (Takke et al.,1999) in the zebrafish. The significance of this effect on the size of the neural primordium is unclear. In most embryos that show an enlargement of the neural plate, there is a concomitant reduction in the numbers of cells expressing markers of primary neurones (islet1 and elavl3, Fig. 3A-C, not shown). As her3 is not normally expressed in the regions of the neural plate from which primary neurones originate, we asked whether this effect is due directly to the product of the injected her3 mRNA, or is caused by the products of other genes, the activity of which may depend on Her3, and which are expressed in the territories in which primary neurones arise. The proneural gene neurog1 elicits ectopic development of islet1-positive cells (Blader et al., 1997; Takke et al.,1999) and is repressed by Her4, another member of the hairy-E(spl)family (Takke et al., 1999). Thus, it is conceivable that the reduction in the number of primary neurones seen in the injected embryos is due to repression of neurog1. Indeed,probing with neurog1 of embryos injected with her3 mRNA revealed that similar proportions of embryos have an enlarged neural plate and a reduced density of neurog1 transcripts(Fig. 3D,E). This observation suggests that the effect of misexpression of her3 on primary neurones is, at least in part, due to repression of neurog1. Injection of her3 mRNA also weakly suppresses transcription of deltaA(not shown), a target of neurog1(Takke et al., 1999). Gel retardation assays suggest that repression of neurog1 transcription may be mediated by direct binding of Her3 protein to N-boxes present in the neurog1 promoter (CTC ACA AGC TCA CAC GAG CTG) at position –129 relative to the ATG(Fig. 3F)(Blader et al., 2003). Thus,when increasing amounts of Her3 protein are incubated in the presence of oligonucleotides containing N-boxes from the neurog1 promoter region(see Materials and methods), clear shifts in the electrophoretic mobility of the labelled probe are detectable. The shift can be selectively inhibited either by adding non-radioactive wild-type oligonucleotides, or using oligonucleotides containing mutations in both N-box sequences. In the latter case, the affinity of the nucleotide for Her3 is considerably reduced, whereas mutation of either box has a weaker effect. Furthermore, non-radioactive mutant oligonucleotides do not effectively compete with the labelled wild-type probe for binding of Her3.

Two different antisense oligonucleotides (morpholinos: MO her3 and 2 blocker) were designed to inhibit her3 function by preventing the synthesis of Her3 protein (Nasevicius and Ekker, 2000). In addition, a mismatch morpholino was injected as a control. Whereas the mismatch morpholino did not cause any detectable deviation from the wild-type pattern, both experimental morpholinos, MO her3 and 2 blocker, had identical effects with a penetrance of over 90%. neurog1 transcription was activated ectopically within the primordium of the rhombencephalon (Fig. 4A,B). RT-PCR analyses confirmed the presence of additional neurog1 transcripts in embryos after the injection of the morpholinos(Fig. 4F). In addition to the ectopic expression of neurog1 in the rhombencephalon, other neurog1-positive cells were occasionally seen in other regions of the neural plate. Although it is difficult to assess the functional significance of these latter changes in the neurog1 pattern of the morpholino injected animals, they may explain the relative abundance of neurog1RNA found in the RT-PCR analysis (Fig. 4F).

Double in situ hybridisation with egr2b showed that ectopic induction of neurog1 is restricted to rhombomeres 2 and 4(Fig. 4B), encompassing the region that connects the clusters of motoneurones and sensory neurones(Geling et al., 2003). Ectopic expression of neurog1 is associated with ectopic activation of a number of genes involved in lateral inhibition (deltaA, deltaD, coe2and her4) in the same regions of rhombomeres 2 and 4(Fig. 4C-E, not shown), and ectopic neurones can be subsequently detected within these same regions(Fig. 4D). The available evidence indicates that all these genes may be activated by Neurog1(Bally-Cuif et al., 1998; Dubois et al., 1998; Takke et al., 1999; Hans and Campos-Ortega, 2002)and therefore, their ectopic activation in rhombomeres 2 and 4 might well be caused by the ectopic induction of neurog1, leading to the formation of ectopic neurones.

As reproducible phenotypic effects of blocking her3 mRNA translation with morpholinos were rather weak, i.e. restricted to a fairly small region of the neural anlage, whereas the her3 transcription domain is much broader, we suspected that the efficacy of the injected morpholinos might be compromised. To test whether the morpholinos can completely knock-down her3 activity, a construct encoding a Her3:gfp fusion was synthesised, and 400 ng/μl mRNA transcribed from this plasmid was injected together with each of the two morpholinos. Another batch of embryos was injected with 400ng/μl her3:gfp mRNA without the morpholinos. After injection, embryos showing Texas Red staining were selected, and the number of embryos with GFP-mediated fluorescence was determined between 30% and at 85% epiboly. 100% of the embryos in the control series showed a fluorescent signal (Fig. 5). The simultaneous injection of either morpholino completely eliminated the fluorescent signal. This result suggests that both morpholinos completely inhibit the translation of her3 RNA.

Injection of her3 morpholinos results in an increase in the density of her3 transcripts in all the her3 expression domains (Fig. 4G). This suggests that Her3 represses transcription of its own gene within the limits of its expression domain. To test whether her3 is indeed subject to autoregulation, gel retardation assays were carried out. Two different N-boxes are present in the promoter region of the her3 gene (N1: TCC AGC AGA AAG, and N2: CCC ACA CGA CCG). Gel retardation assays show that labelled oligonucleotides containing either of these N boxes display a clear shift in electrophoretic mobility (Fig. 4H) following incubation with Her3. The shifts can be inhibited with increasing concentrations of the non-radioactive oligonucleotide. The presence of two or three shifted bands suggests that Her3 may bind to the target DNA in mono-, di- and trimeric forms. A similar inhibition of the DNA binding is observed if oligonucleotides with mutated N-boxes are used (see Materials and methods).

To gain insight into the structural requirements for Her3 function, we constructed seven deletion variants of the her3 cDNA (see Fig. 6), and injected the corresponding mRNAs into embryos at the two-cell stage. Injected embryos were processed for in situ hybridisation with a neurog1 probe. The effects of mRNA injections were comparable for five of the seven variants, her3^ΔN, her3^Δb, her3^ΔbHLH, her3^ΔHLH and her3^ΔWRPW, which encode derivatives that lack the LCC (low compositional complexity) domain, the basic domain, the bHLH domain, the HLH domain and the C-terminal tetrapeptide WRPW, respectively. All these variants have partially lost the ability to suppress neurog1transcription (Table 1). By contrast, the last two variants tested, her3^Δorange and her3^ΔC, which encode products that lack the so-called orange domain (Dawson et al.,1995) and a C-terminal segment, respectively, behave like gain-of-function mutants. Injection of mRNA for either construct leads to a large increase in the number of embryos that show reduced neurog1transcription.

The data described in the previous section indicate that Her3 suppresses the transcription of neurog1. However, they do not show whether Her3-dependent transcription repression is mediated directly or indirectly,e.g. by activating one or more other genes whose products then suppress transcription of downstream genes. To distinguish between these two alternatives, two constructs encoding C-terminal fusions of Her3 to either the transactivation domain of VP16 (her3VP16) or the repressor domain of engrailed (her3eng) were synthesised(Fig. 6). Embryos injected with the corresponding mRNAs were probed by in situ hybridisation with neurog1 and elavl3 cDNAs after gastrulation. Embryos expressing the Her3eng fusion had fewer primary neurones(Fig. 7B; Table 2) and showed a suppression of neurog1 transcription (not shown). That is, injection of her3eng mRNA had the same effects as the injection of her3 mRNA. By contrast, embryos expressing the Her3VP16 fusion showed strong ectopic activation of neurog1 transcription(Fig. 7D). These results are consistent with a direct role for Her3 as a transcriptional repressor.

Genes with homology to E(spl) have repeatedly been shown to be activated by Notch signalling, both in Drosophila(Lieber et al., 1993; Jennings et al., 1994; Bailey and Posakony, 1995; Lecourtois and Schweissguth,1995) and vertebrates(Furukawa et al., 1995; Jarriault et al., 1995; Hsieh et al., 1996; Kopan et al., 1996; Wettstein et al., 1997; Takke and Campos-Ortega, 1999; Takke et al., 1999). Therefore, we tested whether her3 is also under the control of Notch signalling. We used Gal4-mediated gene misexpression to activate, in the first instance, a constitutively active variant of the zebrafish notch1a receptor(UAS:myc-notch1a-intra) (Scheer and Campos-Ortega, 1999; Scheer et al., 2001; Lawson et al., 2001). Crosses were made between heterozygous hsp70::Gal4 and UAS:myc-notch1a-intra individuals, and embryos derived from these crosses were heat-shocked, at the 50% epiboly stage, for 30 minutes at 40°C and allowed to develop until the 1- to 2- to the 5- to 6-somite stage. Fixed embryos were either stained with antibodies against Myc and probed by in situ hybridisation with her3 cDNA, or processed directly for her3 in situ hybridisation without anti-Myc staining. her3 transcription was found to be downregulated, particularly in the mesencephalic/rhombencephalic domain in embryos expressing Myc(Fig. 8A). To test whether the observed repression of her3 is indirect, i.e. caused by Notch1a-mediated activation of transcriptional repressors, we misexpressed her4, another member of the E(spl) family that is activated by Notch1a:intra (Takke et al.,1999), by the Gal4-UAS technique and by mRNA injection. First, we crossed hsp70::Gal4 heterozygotes with UAS:her4 homozygotes,heat-shocked the progeny embryos at 50% epiboly and analysed them by her3 in situ hybridisation. In all cases, individual embryos were genotyped by PCR after in situ hybridisation(Fig. 8G). Second, wild-type zygotes were injected with her4 mRNA and the embryos were then analysed in the same manner. The results were the same in both cases: her3 transcription was strongly repressed(Fig. 8B,C). It should be noted that stronger repression of her3 transcription was obtained upon misexpression of her4 than after misexpression of notch1a-intra, suggesting that activation of her4 is a prerequisite for transcriptional repression of her3. her3transcription was also repressed after the injection of neurog1 mRNA(not shown), most probably owing to transcriptional activation of genes that encode some of the links in the feedback loop that regulates lateral inhibition, e.g. deltaA, deltaD and her4(Takke et al., 1999).

To probe further the functional significance of these observations, we analysed her3 transcription in embryos in which Notch signalling had been perturbed by injecting deltaDPst mRNA, which encodes a dominant-negative deltaD variant (Takke et al., 1999) [see Haddon et al.(Haddon et al., 1998), for the effects of the corresponding variant of Xenopus Delta, deltastu, in zebrafish]. Embryos injected with deltaPst mRNA showed ectopic her3 transcription in various regions, both within and outside the neural plate (Fig. 8D).

Gel retardation assays suggest that Her4-mediated repression of her3 occurs by direct binding of Her4 to N-boxes(Tietze et al., 1992; Sasai et al., 1992; Oellers et al., 1994) present in the her3 promoter (see Materials and methods). Thus, incubation of increasing concentrations of Her4 protein in the presence of oligonucleotides containing N-boxes from the her3 promoter region results in clear shifts in the electrophoretic mobility of the labelled probes. The shifts suggest binding of oligomeric forms of Her4 to the target DNA(Fig. 8H).

We tested to what extent transcription of other zebrafish genes of the hairy/E(spl) family, such as her5 and her6, is also dependent on Notch signalling. To do this, we probed embryos from both of the crosses mentioned above (hsp70::Gal4 to UAS:myc-notch1a-intra and hsp70::Gal4 to UAS:her4)with her5 and her6 cDNAs. Whereas transcription of her5 is strongly repressed in embryos from both crosses(Fig. 8E,F), transcription of her6 seems not to be affected at all (not shown). Therefore, the fact that is a member of the hairy/E(spl) family does not necessarily imply that it is activated as a result of Notch signalling.

Two main conclusions can be drawn from the work presented above. First, her3 encodes a zebrafish homologue of the Drosophila E(spl)proteins, which represses the transcription of neurog1 and is a target of Notch1a signalling. Second, in contrast to other members of the E(spl) family, the her3 gene itself is repressed rather than activated by Notch signalling.

The first conclusion is based on structural and functional considerations. Structurally Her3 shows considerable sequence identity to proteins of the E(spl) family in its bHLH domain, the region that binds DNA and is involved in target recognition (Akazawa et al.,1992; Tietze et al.,1992; Oellers et al.,1994). Furthermore, Her3 also exhibits the other characteristics of members of this family, such as the C-terminal tetrapeptide WRPW and the orange domain (Dawson et al.,1995), which corresponds to helix III/IV defined by Knust et al.(Knust et al., 1992). Davis and Turner (Davis and Turner,2001) classify the hairy-E(spl) proteins, on structural grounds,into four different groups. In their phylogenetic tree, Her3 belongs to the group of E(spl) proteins. With respect to functional criteria, the effects of fusions to the transactivation domain of VP16 and the repression domain of Engrailed indicate that Her3 is a transcriptional repressor. Gel retardation assays and deletion analyses support the contention that Her3 represses transcription by binding directly to so-called N-boxes, a major DNA target for the E(spl) proteins (Sasai et al.,1992; Tietze et al.,1992; Oellers et al.,1994) (for a review, see Davis and Turner, 2001).

The variant her3^ΔWRPW, which encodes a Her3 derivative that lacks the C-terminal tetrapeptide WRPW, reveals an additional element of functional similarity to E(spl). In Drosophila,the WRPW motif is essential for the association of hairy-E(spl) proteins with the co-repressor groucho (Wainwright and Ish-Horowicz, 1992; Paroush et al., 1994; Dawson et al.,1995; Fisher et al.,1996; Giebel and Campos-Ortega, 1997), and its removal results in a non-functional polypeptide. Similarly, injections of her3^ΔWRPW mRNA show that this variant has partially lost the ability to suppress target gene expression. Similar results have been reported for her5, another member of the same protein family (Geling et al., 2003). However, the WRPW tetrapeptide appears to be functionally dispensable in the case of other members of the family. Thus, a Her4 variant lacking the WRPW domain was found to behave like the wild-type protein(Takke et al., 1999).

Finally, we find that her3^Δorange and her3^ΔC, which encode products that are devoid of the orange domain and of the region between the orange domain and the WRPW motif, respectively, behave like gain-of-function mutants. This conclusion is based on the fact that their expression leads to a more pronounced reduction in neurog1 transcription than does the wild-type Her3. The deletion derivative encoded by her3^ΔC is similar to the product of the E(spl)^D allele of Drosophila(Knust et al., 1987; Klämbt et al., 1989),with the exception of the WRPW-coding region, which is still present in the former and absent in the latter. The Her3^ΔC deletion behaves like the product of E(spl)^D when the expression of this gene is driven by Gal4 in a Gal4-UAS experiment(Giebel and Campos-Ortega,1997). When expressed under the control of Gal4,E(spl)^D behaves like a dominant-negative variant. Dominant-negative effects were interpreted as being due to inhibition of the function of the endogenous E(spl) proteins by competitive or neutralising interactions with the truncated proteins (Giebel and Campos-Ortega, 1997). As the gel shift analyses suggest that Her3 may bind to DNA as dimers or trimers (Fig. 4F), association of the endogenous proteins with those supplied exogenously might also explain the gain-of-function and dominant-negative effects seen with the Her3 variants. In Drosophila, deletion of either the orange domain or the WRPW leads to strong impairment of the E(spl)function (Wainwright and Ish-Horowitz,1992; Schrons et al.,1992; Dawson et al.,1995; Fisher et al.,1996). It is assumed that the region between the orange domain and the WRPW motif may be required as a spacer to accommodate Groucho, so that its removal prevents the association of the WRPW with Groucho(Dawson et al., 1995; Giebel and Campos-Ortega,1997).

Taken together, the results described above suggest that Her3 binds directly to DNA and acts as a transcriptional repressor. However, mechanisms of transcriptional repression other than direct DNA binding can not be excluded. Thus, in addition to the bHLH domain required for DNA binding, both the orange domain and the WRPW tetrapeptide appear to play a prominent functional role. In fact, despite the abundance of data available (see Davis and Turner, 2001), it remains difficult to make generalisations with regard to how E(spl) proteins function.

Our present results point to the proneural gene neurog1 as one of the targets of Her3 function. Indeed, neurog1 transcription, as well as that of several target genes of Neurog1, is repressed following injection of her3 mRNA. Gel retardation assays show that neurog1repression might be due to direct binding of Her3 to N-boxes in the neurog1 promoter. This function is clearly compatible with the known function of members of the E(spl) family as strong suppressors of proneural gene function. Our data do not allow us to decide whether Her3 acts on deltaA, islet1 and elavl3 directly or via neurog1.Injection of morpholinos, either MO her3 or 2 blocker, leads to ectopic expression of neurog1 and subsequent induction of a number of targets of Neurog1, as for example deltaA, deltaD, coe2 and her4, and the ectopic induction of primary neurone development.

However, ectopic induction of neurog1 following morpholino injections is restricted to rhombomeres 2 and 4, whereas the remaining domains of neurog1 expression remain unaffected. As injection of her3 mRNA affects neurog1 transcription in all its expression domains, the relatively mild effect of morpholino injection is a striking result. Although we do not yet have a satisfactory explanation for this observation, two possible hypotheses can be considered. First, it is conceivable that under normal conditions regulatory interactions between Her3 and neurog1 are restricted to the regions of rhombomeres 2 and 4 that connects r2MN and r2SN, and r4MN and r4SN, respectively. In this case, the remaining expression domains of her3 would not manifest regulatory interactions with neurog1. However, in view of the complementary nature of the transcription patterns of her3 and neurog1,this seems rather improbable. The second hypothesis is based on functional redundancy of the genes of the E(spl) family. There are several examples of members of this family being expressed in overlapping domains,both in Drosophila and in vertebrates. In Drosophila, six out of the seven E(spl)-C genes show identical expression pattern in the neuroectoderm (Knust et al.,1987; Knust et al.,1992; Klämbt et al.,1989); in the zebrafish, her4 and her2 exhibit virtually identical expression patterns in the neural plate(Takke et al., 1999; Takke and Campos-Ortega, unpublished), her1 and her7 within the presomitic mesoderm (Oates and Ho,2002; Gajewski et al.,2003), and at least one other gene of the her family display the same expression pattern as her5 (L. Bally-Cuif, personal communication) (Müller et al.,1996; Bally-Cuif et al.,2000). The overlap of their expression domains and their common function explains why the genes of the Drosophila E(spl)-C show marked functional redundancy in early neurogenesis (reviewed by Campos-Ortega, 1993) (see also Delidakis et al., 1991; Schrons et al., 1992). A similar redundancy can be invoked to explain why neurog1transcription outside the region delimited by rhombomeres 2 and 4 is not affected, at least to levels detectable by in situ hybridisation, by misexpression of her3. However, as RT-PCR shows a considerable increase in the amount of neurog1 RNA(Fig. 4G), transcription of neurog1 might in fact be affected in all its domains, albeit either below levels detectable by conventional in situ techniques, or with low penetrance. We have mentioned that, in addition to the rhombencephalon, neurog1-positive cells are occasionally seen in other regions of the neural plate. This would require the expression of other her genes in the her3 domains under discussion, for which there is as yet no evidence. Therefore, for the time being our results do not allow us to decide between the two possibilities.

Despite all the similarities between Her3 and the DrosophilaE(spl) proteins, there is one important difference, which concerns the response to signalling through Notch1a. We find that her3transcription is repressed by Notch signalling, in clear contrast to the behaviour of several other members of the E(spl) family, both in Drosophila and in vertebrates, which are activated(Jarriault et al., 1995; Lecourtois and Schweisguth,1995; Tamura et al.,1995; Hsieh et al.,1996; Kopan et al.,1996; Wettstein et al.,1997; Takke et al.,1999; Takke and Campos-Ortega,1999) (for a review, see Lewis, 1996). This conclusion is based on the results of three types of experiment: (1) the downregulation of her3 transcription observed following misexpression, both by mRNA injection and by the Gal4:UAS technique, of notch1a-intra and her4; (2) the upregulation of her3 transcription in embryos expressing a dominant-negative variant of DeltaD; and (3) the downregulation of gal4 transcription in her3::gal4 embryos, in which gal4 is driven by the regulatory region of her3. Moreover,our results suggest that transcriptional repression of her3 is mediated by direct binding of Her4 to the N-boxes present in the promoter of her3. The functional significance of the Notch1a-mediated repression of her3 remains unclear. As Her3 represses neurog1expression, being thus required to block neurogenesis, one possibility is that Notch signalling promotes, rather than blocks, neurogenesis. However, this possibility requires experimental support, which, for the time being, is still missing.

The results obtained using the Gal4-UAS technique for targeted gene misexpression, i.e. using both UAS:notch1a-intra and UAS:her4, suggest that Notch1a signalling suppresses transcription of the her5 gene as well, whereas transcription of her6, yet another zebrafish hairy-E(spl)-related gene, remains unchanged under the same experimental conditions. Therefore, Notch1a signalling allows us to classify the Her genes into three different groups, depending on whether transcription is activated (her1 and her4)(Takke and Campos-Ortega,1999; Takke et al.,1999), repressed (her3 and her5; present results) or remains unaffected (her6; present results). Whereas members of the first and the second group repress neurogenesis in zebrafish as well as rodents, members of the third group promote neurogenesis, at least in mice (Hes6) (Bae et al., 2000; Koyano-Nakagawa et al.,2000).

her5 might well be a special case among the Her genes. Thus,Geling et al. (Geling et al.,2003) refer to unpublished data indicating that her5within the neural plate is independent of Notch signalling in vivo. However,we find that her5, like her3, is downregulated by Notch1a-intra, thus suggesting that her5 can be a target of Notch signalling under conditions of ectopic Notch expression. Altogether, these results suggest that Notch1a-intra can repress both her3 and her5 expression, but only her3 expression is activated by inhibition of Notch1a-intra, supporting the contention that her3 is a target of Notch signalling in the embryo in vivo, and hence a new player in the regulation of neurogenesis in zebrafish.

A case of Notch-mediated repression of Hes genes has previously been described in the mouse. Using an in vitro assay, Beatus et al.(Beatus et al., 1999) found that misexpression of the intracellular domain (IC) of Notch 3 represses transcriptional activation of Hes genes mediated by the intracellular domain of Notch1. If cells from any of a number of different lines are transfected with Notch1 IC together with a much larger amount of Notch3 IC, Notch1 IC-mediated transcriptional activation of Hes1 and Hes5 is repressed. Moreover, Hes5 transcription is repressed in vivo by Notch3 IC when expression is driven by the nestin promoter.

Formally, the effect of Notch3 IC signalling on Hes5 is the same as that of Notch1a on her3. However, the mechanism of transcriptional repression may not be the same in each case. The promoters of both Hes1 and Hes5 contain RBP-Jκ [Su(H)] binding sites and can be activated by Notch1 IC. Beatus et al. (Beatus et al., 1999)suggest that Notch3 IC may either compete with Notch1 IC for access to RBP-Jκ; or, alternatively, as Notch3 IC cannot activate Hes genes, it may compete with Notch 1 IC for a co-factor present in limited amounts. In the case of her3, sequence comparisons uncover a single high-affinity Su(H)-binding site in the 4.7 kb genomic fragment that was used here to drive spatially regulated transcription of her3, and has been shown to contain Notch responsive elements. Because our results suggest that Notch1a cannot activate her3, it is not clear what role this binding site might play. Given that we have not tested whether other Notch receptors can activate her3, we cannot exclude the possibility that a mechanism like the one proposed by Beatus et al.(Beatus et al., 1999) is involved in the repression of her3 transcription. However, the available data suggest that her3 repression is caused by the activation of her4 by Notch1a signals, and by binding of the Her4 protein to the N-boxes present in the her3 promoter. This hypothesis is supported by the observation, mentioned above, that her4transcription is activated by Notch1a signalling probably via Su(H) binding to several sites in the promoter. Consequently, the mechanism would be different from that in mouse.

We thank Christel Schenkel for expert technical assistance, and Paul Hardy and Thomas Klein for critical reading of the manuscript. We also thank the reviewers of this paper, particularly Reviewer 1, for very constructive criticisms. This work was supported by grants from the Deutsche Forschungsgemeinschaft (DFG, SFB 243, SFB 572) and the Fonds der Chemischen Industrie to J.A.C.-O.