On the organisation of the regulatory region of the zebrafish deltaD gene

The Notch signalling pathway mediates lateral inhibition to bring about and stabilise cell-fate decisions during development. The organisation of the pathway was initially elucidated experimentally by studies on the development of progenitor cells of the central nervous system and of the sensory organs in the neuroectoderm of Drosophila (for a review, see Campos-Ortega, 1993). By expressing proneural genes, such as those of the achaete-scute complex (AS-C), clusters of neuroectodermal cells acquire the competence to adopt a neural developmental fate. Lateral inhibition serves to ensure that the proneural clusters actually provide both epidermal and neural progenitors. The main elements of the Notch pathway are a transmembrane ligand (Delta), a transmembrane receptor (Notch) and a transcriptional repressor (Suppressor of Hairless; [Su(H)]). For Drosophila, there is indirect evidence that, upon binding of the ligand, the intracellular domain of Notch (Nic) is cleaved off, and translocates into the nucleus (Lecourtois and Schweisguth, 1998; Struhl and Adachi, 1998), where, analogous to the situation in vertebrates (Hsieh et al., 1996), it is assumed to associate with Su(H). The Su(H)/Nic complex then activates 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, suppress expression of the proneural genes (Paroush et al., 1994; Dawson et al., 1995; Giebel and Campos-Ortega, 1997), thus giving neuroectodermal cells access to epidermal development.

A regulatory feedback loop between Notch and Delta, which modulates Delta activity by regulating its transcription (Haenlin et al., 1994; Hinz et al., 1994; Kunisch et al., 1994), is an essential element of lateral inhibition. Proneural proteins encoded by the AS-C activate transcription of Delta by binding to specific sites in the Delta promoter (Kunisch et al., 1994). Consequently, the amount of proneural protein contained in a given neuroectodermal cell determines the amount of Delta protein produced and, ultimately, the efficacy of that cell in activating the Notch receptor in neighbouring cells. Within an array of mutually interacting cells, differences in the levels of proneural proteins will tend to increase the strength of the signals emitted by one of the cells and decrease the probability that any of the surrounding cells adopt a neural fate. This in turn reduces the efficacy of the surrounding cells as sources of lateral inhibition, and thus confirms the neural progenitor cell in its developmental fate decision (Kunisch et al., 1994). However, rather than being a result of random fluctuations in proneural protein concentration in the cells of the proneural cluster, it appears that individual cells are biased towards one of the two developmental fates prior to the separation of the two cell types (Seugnet et al., 1997).

Four different Delta homologues, deltaA, deltaB, deltaC and deltaD have been identified in the zebrafish (Dornseifer et al., 1997; Appel and Eisen, 1998; Haddon et al., 1998). Their patterns of transcription have much in common, but they also exhibit some differences. Thus, three of the genes, deltaA, deltaB and deltaD, are expressed in various regions of the neural plate in similar patterns, whereas deltaC and deltaD are also expressed in various mesodermal derivatives.

One of the functions of the Notch pathway in vertebrates, as in Drosophila, is to select individual cells, from groups of initially equivalent ones, for specific fates (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, which is apparently organised similarly to that in Drosophila, in both Xenopus and zebrafish (Wettstein et al., 1997; Takke et al., 1999). However, it is not known whether, as is the case in Drosophila (Kunisch et al., 1994), this feedback operates on proneural proteins that can bind to the promoters of the Delta homologues.

Another major function of the Notch pathway in vertebrates is the segmentation of the mesoderm. Mouse embryos deficient for Notch1 or for the Delta homologue Delta-like 1 (Dll1) show severe somitic defects (Conlon et al., 1995; Hrabe de Angelis et al., 1997). Similar somite defects have been observed in mice that lack RBP-J_κ (Oka et al., 1995; de la Pompa et al., 1997), a vertebrate homologue of Suppressor of Hairless. While these observations suggest that the Notch pathway is required for segmentation of the paraxial mesoderm, it is not known how Notch signalling performs this function or at which step it affects during segmentation. However, there is evidence suggesting that Notch signalling is required to synchronise the activity of groups of presomitic mesodermal cells with respect to the activity of genes that are cyclically expressed (Jiang et al., 2000).

We address the question of how the expression of deltaD is regulated during embryogenesis. We attached various fragments of the deltaD locus to an enhanced version of the reporter gene gfp (Cormack et al., 1996), and transformed these into the germline in order to identify cis-regulatory regions. We found that regulatory sequences are dispersed over a stretch of approximately 12.5 kb of genomic DNA that includes the entire coding region. Germline transformation with a minigene comprising 10.5 kb of this genomic DNA coupled to the 3′ end of a full-length cDNA rescues the abnormal phenotype of embryos homozygous for the amorphic deltaD mutation after eight^AR33 (Holley et al., 2000). In our analysis, we concentrated on the early pattern of transcription of deltaD and identified genomic regions that drive expression in mesodermal and neuroectodermal domains. Two of the genomic regions relevant for expression in specific domains of the neural plate reveal a highly conserved sequence homology between the mouse Dll1 and the zebrafish deltaD promoter (Beckers et al., 2000). Both regions are regulated by the proneural proteins neurogenin 1 (Blader et al., 1997) and zash1a/b (Allende et al., 1994) acting as transcriptional activators. Activation of transcription of deltaD by proneural proteins thus represents a molecular target for the regulatory feedback loop controlled by the Notch pathway in lateral inhibition.

Two overlapping clones comprising a total of 21 kb were obtained from a genomic DNA library (Easy-to-handle eukaryotic genomic library (zebrafish), Mo Bi Tec, Göttingen, Germany). To synthesise the reporter gene constructs, the gfp-coding sequence (Cormack et al., 1996) was ligated to the SV40 polyA signal (a gift from N. Scheer) and the resulting fragment was cloned either upstream or downstream of various genomic fragments in pBluescript. Different endogenous restriction sites were used for construct assembly (upstream sites: –6 XbaI, –3.5 EcoRI, –1.8 Asp700, –1.3 NsiI; downstream sites: +2.8 NsiI and +6.5 Asp718, the latter site lying within pBluescript). For the rescue of the aei^AR33 mutant phenotype, the 3′ terminal end of a full-length deltaD cDNA (fusion of the cDNAs Dl-1 and Dl-2) (Dornseifer et al., 1997), comprising the sequences corresponding to the tenth and eleventh exons, was linked in frame to several copies of the coding sequence for the myc epitope. The resulting fragment was then ligated to genomic DNA using a NruI site located in the ninth exon. The resulting construct includes 10.5 kb of genomic DNA (6 kb of sequence upstream and 4.5 kb downstream of the transcription start site) and carries the entire coding sequence, as the fragment of the cDNA clone provides the last two exon sequences.

DNA fragments were isolated from pBluescript with either Asp718/NotI or one of the endogenous restriction enzymes indicated in the map of the deltaD locus in Fig. 2. The preparation of the DNA fragments followed the previous methods (Scheer and Campos-Ortega, 1999). Injected, putative founder fish (G0) were crossed inter se and their progeny (F₁) were screened with a fluorescence stereomicroscope [Leica-Stereomikroskop (MZ FLIII)] for GFP-mediated signals. Animals scored as positive were raised to adulthood and crossed either to wild-type fish or inter se. Animals that showed no fluorescence were screened by PCR using deltaD (5′CAACAGAGCATCAACCCGAGC3′)- and gfp (5′CGTGTCTTGTAGTTCCCGTCATC3′)-specific primers.

In situ hybridisation was performed as described by Bierkamp and Campos-Ortega (Bierkamp and Campos-Ortega, 1993). Antibody staining (anti-myc and anti-GFP) was carried out as described by Westerfield (Westerfield, 1994) with some modifications. Embryos and larvae older than 48 hours were incubated for 30 minutes with 0.01% collagenase/PBT before bleaching in 1% H₂O₂/PBT overnight. EliteABC (Vectastain) was used to enhance the sensitivity of the staining. All incubation steps were performed at 4°C overnight in 5% DMSO/10% goat serum/PBT.

To determine quantitatively effects of neurogenin 1- and zash1a/b-sensitive promoter elements, CAT assays were used. Several promoter constructs and the control plasmid pBS 6lacZ were co-injected with mRNA for either neurogenin 1 or zash1a into wild-type zygotes; the animals were allowed to develop to the five-somite stage and the levels of CAT and β-Gal were determined (Vize, 1996) using ELISA kits (Roche). Site-directed mutagenesis of E-boxes was carried out on the CAT constructs using the QuikChange™ Site-Directed Mutagenesis Kit from Stratagene. The following mutations were introduced into the HI and HII boxes (see Results).

Wild-type: GTGAAAAAACGCCATTTGGTTGGGAGCAGATGGTTGGCTTGGG

Mutant: GTGAAAAAACGCTATATCGTTGGGAGTATATCGTTGGCTTGGG

Wild-type: GTGAGGGGAGCAGGTGCTGTGTGAATTACCATACAGCTGAGAGCACAGAG

Mutant: GTGAGGGGAGTAGTTCCTGTGTGAATTACCATATAGTTCAGAGCACAGAG

For each reaction, 10 ng of plasmid DNA was used. Primers used for the mutagenesis experiments were:

HI 1s, GTGAAAAAACGCTATATCGTTGGGAGCAG;

HI 1a, CTGCTCCCAACGATATAGCGTTTTTTCAC;

HI 2s, CATTTGGTTGGGAGTATATCGTTGGCTTGGG;

HI 2a, CCCAAGCCAACGATATACTCCCAACCAAATG;

HI 1+2s, GTGAAAAAACGCTATATCGTTGGGAGTATATCGT;TGGCTTGGG;

HI 1+2a, CCCAAGCCAACGATATACTCCCAACGATATAGCGTTTTTTCAC;

HII 1s, GTGAGGGGAGTAGTTCCTGTGTGAATTACC;

HII 1a, GGTAATTCACACAGGAACTACTCCCCTCAC;

HII 2s, GTGAATTACCATATAGTTCAGAGCACAGAG;

HII 2a, CTCTGTGCTCTGAACTATATGGTAATTCAC;

HII 1+2s, GTGAGGGGAGTAGTTCCTGTGTGAATTACCATATAGTTCAGAGCACAGAG; and

HII 1+2a, CTCTGTGCTCTGAACTATATGGTAATTCACACAGGAACTACTCCCCTCAC.

Primer orientations are: s, sense; a, antisense. Mutated bases are bold.

During embryogenesis in the zebrafish, deltaD transcripts are distributed in a complex pattern, comprising both mesodermal and neuroectodermal expression domains. The following is a summary of the main features of the deltaD transcription pattern within mesodermal and neural derivatives in embryos (see also Dornseifer et al., 1997; Haddon et al., 1998).

Transcription of deltaD in the developing mesodermal primordia begins at 30% epiboly within the entire marginal region (Fig. 1A). At about 50% epiboly, transcription ceases in the embryonic shield, i.e. the prospective axial region of the embryo (Fig. 1B), and, at 60-70% epiboly, the expression domain extends from the marginal zone into the hypoblast. From 80-90% epiboly onwards, two transverse, band-like domains become visible – first within the hypoblast and later in the presomitic mesoderm (Fig. 1C-H). This pattern continues throughout somitogenesis. Posterior to the bands, a much lower density of transcripts can be discerned down to the tip of the growing tail, where a high concentration of transcripts is present (Fig. 1D-H). In addition to the presomitic bands, the anterior halves of the somites themselves contain deltaD transcripts (Fig. 1G,H).

Expression within the epiblast starts at 80-90% epiboly in the form of two bands that extend in the animal-vegetal axis to eventually become continuous stripes (Fig. 1C-F). Expression in the neural plate is characterised by a number of domains in the prosencephalic-mesencephalic primodium and in the primordium of the hindbrain. Three longitudinal expression domains, i.e. lateral, intermediate and medial, are present in the neural plate region corresponding to the hindbrain. Of these, the medial and lateral domains extend into the territory of the spinal cord, whereas the intermediate domain is restricted to the rhombencephalon (Fig. 1C-F). These longitudinal domains are also part of the patterns of expression of deltaA and deltaB (Haddon et al., 1998; Appel and Eisen, 1998). Experimental evidence (Dornseifer et al., 1997; Haddon et al., 1998; Appel and Eisen, 1998; Appel et al., 2001; Takke et al., 1999) indicates that primary sensory (so-called Rohon-Beard) neurones are selected from the cells that make up the lateral stripes, while primary motoneurones are generated from the medial ones, as a result of Notch-mediated signalling.

Within the primordia of forebrain and midbrain, at the tailbud and two-somite stages, four main domains of deltaD RNA expression can be distinguished (Fig. 1C-E), numbered 1-4 in the order of their appearance. The first one (domain 1 in Fig. 1C,D) is located at the level of the prospective mesencephalon; the second (2 in Fig. 1D-H) is prosencephalic, and is disposed in the shape of an arch, apparently delimiting the prosencephalic primordium anteriorly and laterally; the third (3 in Fig. 1E-H) is located medially in the prosencephalon; and the fourth (4 in Fig. 1E) lies in the metencephalon.

In order to determine the extents of coding and regulatory sequences in the deltaD gene, we chose to perform germline transformation of the after eight mutant aei^AR33, which corresponds to a loss of function of the deltaD gene (Holley et al., 2000). The extant after eight mutations (van Eeden et al., 1996) are recessive, homozygous viable mutations. The mutant phenotype is manifested in mesodermal and neuroectodermal derivatives. Fusion of the somites caudal to the seventh to ninth somite (see Fig. 3A) and mild hyperplasic defects of primary neurones are observed (Holley et al., 2000). Two overlapping genomic clones encompassing 12.5 kb of the deltaD locus were sequenced and the structure of the locus was defined from this sequence (Fig. 2). The deltaD-coding region consists of 11 exons and 10 introns (Fig. 2A). In order to add five copies of the myc epitope to the rescue construct, part of the genomic DNA was deleted and replaced by the 3′ end of a full-length cDNA (see Materials and Methods). The rescue clone consisted of 6 kb of genomic sequence upstream and 4.5 kb downstream of the putative transcription start site (as defined by the TATA box and the maximal extent of cDNAs), including the first nine exons of the protein coding region. The tenth and eleventh exons and the 3′UTR were provided by the 3′ terminal region of a full-length cDNA with the myc tags. Hence the entire coding sequence is present in this construct, which was injected into aei^AR33 homozygous zygotes. Upon reaching adulthood, the injected animals were backcrossed to aei^AR33 homozygotes and embryonic progeny of these crosses were screened for insertions of the construct by PCR using myc-specific primers. Progeny of a total of 102 injected animals were screened and six independent insertions were recovered, of which three [Tg(dlD:myc)^kca5–7] were further analysed. Mutants carrying each of the three deltaD:myc insertions expressed a deltaD-myc fusion protein of the expected size of 91 kDa, as shown on western blots probed with the anti-myc antibody (Fig. 3B). These animals showed none of the abnormal phenotypic traits characteristic of the aei mutants and were indistinguishable from the wild type.

The most obvious difference between insertion-bearing embryos and mutants was the appearance of normal somitic boundaries caudal to the ninth somite in the aei^AR33;Tg(dlD:myc)^kca5-7 transformants (Fig. 3A′), as this trait is completely penetrant in the aei mutants (van Eeden et al., 1996). Several genes, e.g. her1, myoD, mesp-a, papc and deltaC, exhibit a conspicuous pattern of transcript expression in stripes made up of presomitic and/or somitic, paraxial mesodermal cells in the wild-type. This pattern is perturbed in the aei mutants (Fig. 3C) (van Eeden et al., 1996; Holley et al., 2000; Durbin et al., 2000; Jiang et al., 2000). However, the expression patterns of these genes were normal in the transformants (Fig. 3C′) and indistinguishable from wild type (Fig. 3C′′). The number of primary neurones (detected with an islet1 probe), which is increased in the aei mutants (Holley et al., 2000), was lowered in the transformants to a wild-type level (Fig. 3D-D′′′). We therefore conclude that the artificial deltaD construct is functionally equivalent to the wild-type deltaD gene.

In order to identify genomic regions containing cis-acting elements that regulate transcription of deltaD, the coding region of the reporter gene gfp, beginning immediately after the 5′UTR, was fused to various fragments of the 12.5 kb genomic sequence that rescues the aei^AR33 phenotype. The resulting constructs were injected into wild-type zygotes and, after reaching adulthood, the injected animals (putative G0 transgenics) were crossed inter se and the progeny embryos were screened for gfp-mediated fluorescence and by PCR analysis of DNA prepared from pools of embryos, using a forward primer located in the deltaD promoter and a reverse primer in the gfp-coding sequences (see Materials and Methods). Transgenic animals were used to establish lines carrying the different transgenes. Although transcription of the endogenous deltaD gene begins at 30-40% epiboly (Dornseifer et al., 1997; Haddon et al., 1998) (Fig. 1), gfp-mediated fluorescence was first detected as late as the tailbud stage in embryos carrying some of the constructs used here (Fig. 4A). Thus, to allow comparison with the early endogenous expression pattern, we assayed for deltaD-gfp transcription by in situ hybridisation using a digoxigenin-labelled gfp probe. Fig. 2 shows the transgenes analysed in this study. All strongly expressed transgenes were associated with gfp-mediated fluorescence. The remaining transgenes were either weakly expressed, and could be analysed only by gfp in situ hybridisation, or were not expressed at all.

The pattern of gfp transcripts in transformants carrying construct 6gfp+3′, in which the gfp gene is sandwiched between the 5′ and 3′ halves of the entire 12.5 kb genomic DNA, displays all temporal and spatial elements of the deltaD expression pattern. Therefore, together with the rescue of the aei mutant, this observation strongly suggests that the entire deltaD locus is contained within a 12.5 kb segment of genomic DNA (Fig. 4A-F).

Three different regions of the deltaD promoter contain all regulatory elements required for transcription in the tailbud, in the presomitic paraxial mesoderm and in the somites (Fig. 5A). One element required for expression in the marginal zone and, at later stages, in the tailbud, is located in the 3′ region, as all constructs carrying the fragment 0 to +2.8 showed, albeit weakly, expression in the marginal zone and the tailbud (Fig. 5B). The strength of 3′-mediated expression was considerably increased when the +2.8 to +6.5 fragment was added. The 3′ region also contributes weakly to the control of diffusely distributed transcripts within the presomitic mesoderm and, from the eighth somite on, in the somites. Double in situ staining with myoD and gfp revealed that this somitic gfp expression is in fact an ectopic expression, as it is restricted to the posterior half of the somites (not shown). An additional regulatory element that controls transcription within the newly formed somites is located in the upstream DNA, between –6 and –3.5 (construct 6Δ3.5-0.5gfp) (Fig. 5C). The third mesodermal element is located proximal to the basal promoter, and is also responsible for control of transcription in the somites (construct 1.3gfp) (Fig. 5D). More importantly, this region drives expression in the stripe domains in the presomitic mesoderm. On its own, this element drives transcription weakly. However, transcription is quite intense when the element is combined with the –6 to –3.5 region. Besides the timing of the mesodermal expression, we also observed a change in the amount of transcripts depending on the upstream DNA fragment driving gfp. Embryos carrying the construct 6gfp gave rise to a normal level of transcripts in the presomitic mesoderm, i.e., comparable with that of the endogenous gene; embryos transgenic for 1.3gfp exhibited the same expression in the presomitic mesoderm. Strikingly, embryos transgenic for 6Δ1.8-1.3gfp showed a higher density of transcripts in the presomitic mesoderm (see Fig. 7A,F), whereas embryos transgenic for 6Δ3.5-1.8gfp did not show any expression in the presomitic mesoderm (see Fig. 7C,H). This suggests the existence of an element reducing transcription in the presomitic mesoderm located between –1.8 and –1.3.

Three different regions were found to drive transcription within the neural primordia (Fig. 6A,F,K). Particularly striking are two well-defined regions located between –3.5 and –1.8 (HI) and –1.8 and –1.3 (HII) (Fig. 2A), which show substantial sequence similarity to two regions in the promoter of the mouse Dll1 gene (Beckers et al., 2000). The 6 kb of genomic DNA upstream of the putative transcription start site (6gfp), which includes HI and HII, contains regulatory sequences necessary for the expression of all features of the transcription pattern in the neuroectoderm, with the exception of the early expression (tailbud- to five-somite-stage) of the horseshoe-like domain at the boundary of the prosencephalon (domain 2) and the anterior expression of domain 3 (Fig. 6B,G,L). This latter domain requires additional sequences in the downstream region, which could not be mapped to a defined segment (Fig. 6E,J,O). From the five-somite-stage on, gfp expression corresponds to the endogenous expression of deltaD. Absence of HI, as in the constructs 6Δ3.5-1.8gfp and 1.8gfp, is associated with the loss of reporter gene expression in the trigeminal ganglion and in the three stripes in the neural plate at the level of the prospective rhombencephalon and spinal cord (Fig. 7C,H). Absence of HII, as in construct 6Δ1.8-1.3gfp, leads to loss of various expression domains in forebrain, midbrain and hindbrain (Fig. 7A,F). Finally, absence of both HI and HII, as in constructs 6Δ3.5-1.3gfp and 1.3gfp, determines the loss of essentially all elements of the neural pattern (Fig. 6C,H,M).

Constructs containing the most proximal 2.8 kb of genomic DNA in the 3′ region, or even the complete 3′ fragment down to +6.5 which encompasses all the coding sequences of deltaD, are also associated with gfp expression in all neural expression domains (Fig. 6D,I,N). However, expression mediated by this 3′ DNA in these domains is significantly delayed and much weaker than when the 3′ region is accompanied by 5′ DNA. Therefore, we assume that control of neural expression by the 3′ region is subordinate to control by the 5′ region. It should be noted that there is a putative MyT1 binding site at +271. We will come back to this point later.

neurogenin 1 (Blader et al., 1997; Korzh et al., 1998) and zash1a/1b (Allende et al., 1994) encode putative proneural proteins. Injection of neurogenin 1 mRNA has been shown to ectopically activate transcription of the endogenous deltaD gene (Takke et al., 1999), and neurogenein 1 is therefore very likely to regulate deltaD directly. Similarly, injection of mRNA encoding neurogenin 1 into zygotes carrying the 3.5gfp+3′ transgene leads to ectopic activation of gfp transcription (data not shown) as in the case of the endogenous deltaD gene (Takke et al., 1999). In addition, it is worth noting that constructs containing HI and HII drive reporter gene expression in a pattern that reflects the distribution of neurogenin 1 and zash1a/1b transcripts. Thus, in the two-somite stage embryo, neurogenin 1 is transcribed in domain 1, in domain 4, and in the medial, intermediary and lateral stripes of deltaD expression in the prospective rhombencephalon and spinal cord (Fig. 7A,B,F,G) (Blader et al., 1997; Korzh et al., 1998; Takke et al., 1999). However, in the 12-somite stage embryo zash1a is transcribed in the epiphysis, ventrally in the diencephalon and dorsally in the hindbrain, whereas zash1b is transcribed dorsally in the telencephalon, in rhombomeres 2 and 4, and in cells of the spinal cord (Fig. 7C,D,E,H-J) (Allende et al., 1994). HI and HII may thus represent direct targets for the proneural proteins neurogenin 1 and zash1a/1b.

We used CAT assays to look for possible activation of deltaD by neurogenin 1 and/or zash1a/1b. For this, plasmids carrying specific regions of the deltaD promoter fused to the CAT-coding sequence were injected, together with mRNA for either neurogenin 1 or zash1a and zash1b, into wild-type zygotes (Fig. 8A). The amount of chloramphenicol acetyl transferase (CAT) was determined by ELISA in extracts of the injected animals when they had reached the five-somite stage. When co-injected with mRNA for neurogenin 1, constructs including HI (6CAT and 3.5CAT) expressed high levels of CAT; constructs lacking HI (1.3CAT, 1.8CAT and 6Δ3.5-1.8CAT) generated much lower amounts of the protein (Fig. 8B). Deletion constructs, particularly one comprising HI and the basal promoter (HI 0.5CAT), which still express significant amounts of CAT, show that the neurogenin 1-mediated CAT activation requires HI (Fig. 8D).

Pilot experiments with either zash1a or zash1b gave similar results for both proteins (results not shown). Therefore, as the DNA-binding regions of both proteins are identical (Allende et al., 1994), we assume that the conclusions drawn for zash1a can be extended to zash1b. After co-injection with mRNA for zash1a, constructs carrying HII (6CAT, 3.5CAT and 1.8CAT) express high levels of CAT, whereas those without HII (1.3CAT and 6Δ1.8-1.3CAT) gave rise to much lower amounts of the reporter (Fig. 8C).

Neurogenin 1 and zash1a/b are bHLH proteins, and bHLH proteins are known to bind to E-boxes. HI contains two E-boxes, which are conserved in position and orientation in mouse and zebrafish (Beckers et al., 2000). Three nucleotides were replaced in each E-box in HI (see Materials and Methods). Mutation of either of the E-boxes brings about a marked reduction in the amount of CAT expressed. Mutation of both E-boxes in HI causes a further reduction to the basal level (Fig. 8D). Two E-boxes are also located within HII, which, like those in HI, are conserved in position and orientation in the HII region of the mouse gene. Mutation of the distal E-box does not have any apparent effect on the amount of CAT, whereas mutation of the proximal E-box causes a strong reduction in the level expressed (Fig. 8E).

We have seen that the 0 to +2.8 fragment located in the 3′ region mediates reporter gene expression in all neural domains, although with considerable delay relative to the full-length gene, and that this control is temporally subordinate to that by the –3.5 to –1.3 region. It has also been mentioned that there is a putative MyT1-binding site at +271 (Park et al., 2000), as well as 13 E-boxes distributed throughout the 0 to +2.8 region. MyT1 has been described as a direct target of neurogenin 1 (Bellefroid et al., 1996; Koyano-Nakagawa et al., 1999). Therefore, to further characterise the regulatory region and the possible roles played by MyT1, the 13 E-boxes and neurogenin 1, two additional constructs were synthesised (Fig. 8A). One construct comprised the 0 to +2.8 region placed downstream of CAT; the control region in the other reporter construct carried a mutation in the putative MyT1-binding site. mRNA encoding either neurogenin 1 or MyT1 was injected into zygotes together with each one of the plasmids, and the level of CAT was determined as described above. However, none of these constructs could be activated by either MyT1, neurogenin 1 or both. Therefore, the putative MyT1 binding site and the 13 E-boxes contained in this region, appear to be irrelevant in the context studied here.

Two main conclusions bearing on the organisation of the deltaD promoter can be drawn from our observations.

(1) Cis-regulatory sequences are distributed over 12.5 kb of genomic DNA encompassing 6 kb upstream and 6.5 kb downstream of the transcription start site. This DNA can correct all abnormal phenotypic traits associated with the mutation aei^AR33 (van Eeden et al., 1996; Holley et al., 2000), and is thus sufficient to provide the normal function of the deltaD gene. No additional control sequences are expected to occur outside this region, as the 6gfp+3′ construct, which contains the entire 12.5 kb genomic region, displayed all the temporal and spatial features of deltaD transcription that we were able to assess. Genomic regions required for the regulation of expression in mesodermal and neuroectodermal derivatives can be distinguished within this DNA segment. In addition, two putative quantitative enhancers were identified, one between –6 kb and –3.5 kb in the 5′ region, and the other between +2.8 kb and +6.5 kb in the 3′ region. However, this proposal is based solely on the observation that their presence substantially increased the intensity of expression driven by other DNA fragments.

(2) Sequences required for the regulation of specific traits were found to be concentrated in discrete zones, whereas others are dispersed over a wider area. Thus, mesodermal elements were found to be dispersed in relatively large pieces of genomic DNA, whereas two discrete boxes, HI and HII, were found to be necessary and sufficient for almost all neural expression domains, as the comparison of reporter gene expression directed by several constructs clearly shows. Moreover, E-boxes within HI and HII were found to be targets for neurogenin 1 and zash1a, and these elements are therefore responsible for the expression directed by HI and HII in neural domains (see below; Fig. 9).

Besides HI and HII, genomic sequences in the interval 0 to +2.8 in the 3′ region appear to regulate transcription within the same regions of the neural plate, although with a striking delay in comparison with reporter expression driven by the upstream sequence, or relative to the endogenous deltaD gene. As the pattern elements controlled by this genomic interval suggest the participation of proneural proteins and/or their downstream targets, we used CAT assays to probe the role played by 13 E-boxes and a putative MyT1 binding site located there in responding either to neurogenin 1 or to MyT1. On the one hand, the pattern elements controlled by this interval include those which, as discussed above, are neurogenin 1-responsive. On the other hand, the MyT1 protein has been shown in Xenopus to be part of the neurogenic cascade, appearing following the activity of proneural bHLH proteins and rendering cells expressing these proteins insensitive to lateral inhibition (Bellefroid et al., 1996). However, surprisingly enough, the results of CAT assays suggest that these 13 E-boxes do not respond to neurogenin 1, nor does the putative MyT1 site react to MyT1. Therefore, the question as to how expression in neural primordia is driven by the 0 to +2.8 segment remains unanswered.

The similarity of the pattern of reporter gene expression directed by constructs including HI and HII to that of the proneural genes neurogenin 1 and zash1a/b (Blader et al., 1997; Allende et al., 1994) was the first hint that transcription of deltaD in these domains might be activated by the proneural proteins encoded by these genes. Indeed, the CAT assays have provided convincing evidence that both types of proneural proteins activate transcription of deltaD through the E-boxes present in them, either by directly binding, or by activating other bHLH proteins which themselves bind, to the E-boxes. Previous work had shown that injection of mRNA encoding neurogenin 1 activates deltaD transcription ectopically in embryos (Takke et al., 1999). Our present findings indicate that activation of transcription occurs by direct binding of either neurogenin 1, or some other bHLH protein induced by neurogenin 1, to specific E-boxes. It seems highly probable that neurogenin 1 binds directly to the deltaD promoter, as the timing of its expression and of that driven by the neurogenin 1-responsive reporter gene constructs are virtually coincident. In addition, Xenopus Delta1 can be activated by low doses of X-ngnr-1 mRNA, which do not elicit any response from several other downstream targets of X-ngnr-1 (Koyano-Nakagawa et al., 1999).

Transcription of Drosophila Delta is directly activated by the binding of proneural proteins to several E-boxes distributed through the promoter, and this provides the target for the regulatory feedback loop responsible for lateral inhibition (Haenlin et al., 1994; Kunisch et al., 1994). Therefore, in this respect, the promoters of Drosophila Delta and zebrafish deltaD are similarly organised, as transcription of both genes in the neuroectoderm can be activated by proneural proteins. In both cases, the amount of proneural protein determines the strength with which the Delta ligand signals to the Notch receptor and, therefore, the intensity of lateral inhibition (Kunisch et al., 1994; Takke et al., 1999). However, although in Drosophila a relatively large number of active binding sites for proneural proteins are distributed over a large segment of genomic DNA (Kunisch et al., 1994), far fewer binding sites are found in the zebrafish deltaD gene, and these are concentrated in two discrete regions. In fact, in Drosophila 12 E-boxes in a stretch of 4.3 kb of genomic DNA were found to serve as binding sites for homodimers of daughterless, achaete and lethal of scute, and their heterodimeric combinations, and all were required for the neuroectodermal pattern of transcription of Delta (Kunisch et al., 1994). In the zebrafish, however, only three E-boxes that respond to proneural proteins are sufficient to drive transcription in all neuroectodermal domains but one, namely domain 2, which is regulated by sequences contained within the up- and downstream regions. As there is only one Delta gene in Drosophila, whereas at least three delta genes in zebrafish exhibit largely overlapping, but not identical, expression patterns in neural primordia, it is conceivable that this is the reason why a larger number of E-boxes is required to modulate the transcriptional activity and thus the function of Delta in the fruitfly. This modulation may be achieved in the zebrafish by the subtle differences that can be observed between the transcriptional patterns of the three Delta genes expressed in neural primordia.

Our transgenic analysis of the deltaD locus has revealed six distinct cis-regulatory regions, five upstream and one downstream of the transcription start site, that direct gene expression in neuroectodermal and mesodermal subdomains of the embryo. The upstream region of the mouse Dll1 locus shows a similar organisation (Beckers et al., 2000). We propose that both promoters are organised in five modules, of which at least three are phylogenetically conserved. Two of these modules correspond to the regions HI and HII, identified on the basis of their high sequence similarity; in addition, both regions are located in the same relative positions and in the same orientation in both species. However, there is a difference in the pattern of expression driven by these elements in zebrafish and mouse. In stably transformed mouse embryos, HI coupled to the minimal Dll1 promoter is able to direct reporter gene expression primarily to the ventral tube and some derivatives of the neural crest, such as dorsal root and spinal ganglia. By contrast, transformants bearing HII fused to the minimal Dll1 promoter direct expression in the marginal zone of the dorsal region of the neural tube (Beckers et al., 2000). In zebrafish, the expression patterns of HI and HII do not exhibit a restriction to dorsal or ventral regions of the neural tube. It seems probable that this apparent difference is due to the different expression of neurogenin 1 and Mash1 (Ascl1 – Mouse Genome Informatics), and neurogenin 1 and zash1a/b, in mouse and zebrafish, respectively. Indeed, in the mouse embryo the expression patterns of neurogenin 1 and Mash1 show a similar restriction in the neural tube, and it appears that there is a complete overlap in the expression patterns of HI and HII with neurogenin 1 and Mash1 (Ma et al., 1996). Thus, the regulatory network appears to be conserved in zebrafish and mouse, although the expression domains of the corresponding proneural genes have changed during evolution.

With respect to the three mesodermal modules, two appear to be conserved whereas the other has diverged. In both species, one mesodermal element in the region immediately proximal to the minimal promoter [‘msd II’ according to Beckers et al. (Beckers et al., 2000)] is able to direct reporter gene expression in the presomitic mesoderm and nascent somites. However, sequence comparison of these two regions in zebrafish and mouse revealed only minor stretches of similarity, the significance of which still remains to be tested. A second module is represented by the putative silencer of transcription in the presomitic mesoderm at –1.8 to –1.3, flanked by enhancer elements. In mouse, Beckers et al. (Beckers et al., 2000) describe negative regulators flanked by positive regulators of expression in the presomitic mesoderm, i.e. a similar organisation to that in zebrafish. Unfortunately, again in this case the comparison of both DNA sequences has failed to show any similarity in this region. The third mesodermal module identified in the mouse Dll1 promoter, called ‘msd’, is located within the two elements HI and HII. This module is not present in the zebrafish deltaD and might be responsible for the difference in the expression in mature somites: zebrafish deltaD is expressed in the anterior halves of the mature somites, whereas mouse Dll1 is expressed in the posterior halves. Therefore, all these considerations reveal a great deal of phylogenetic conservation in the organisation of the regulatory regions of deltaD and Dll1 expression in zebrafish and mouse.

We thank P. Hardy for critical reading of the manuscript, Christel Schenkel and Iris Riedl for technical assistance, and our colleagues for discussions. 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. The Accession Number for the sequence of the genomic DNA of the deltaD locus is AF426384.