Multiple delta genes and lateral inhibition in zebrafish primary neurogenesis

Both in vertebrates and in invertebrates, neurons originate as isolated cells within a neurogenic neuroepithelium; in the neurogenic regions, many cells have the potential to follow a neural pathway of differentiation, but only a scattered subset do so at any one time. The Delta-Notch lateral-inhibition mechanism plays a key part in this process, by preventing the immediate neighbours of each nascent neural cell from simultaneously embarking on neural differentiation. In vertebrates, this has been established mainly through study of the effects of misexpressing the Delta1 gene (Austin et al., 1995; Chitnis et al., 1995; Dorsky et al., 1997; Henrique et al., 1997a). These vertebrate observations pose an unsolved problem, however, as to how the neural cells become singled out from their neighbours in the first place.

This question has been examined in Drosophila, with regard to the neuroblasts of the central nervous system and the sensory mother cells of the peripheral nervous system. In both cases, the process appears to occur in two steps (Campos-Ortega, 1993; Ghysen et al., 1993). First, a cluster or patch of cells switches on expression of one or more proneural genes, conferring the potential for a neural fate. Then, within this proneural cluster, a second set of genes, the neurogenic genes, are brought into play to allow only a subset of the cells actually to become committed to neural differentiation.

The neurogenic genes code for components of a cell-cell signalling pathway, in which the transmembrane protein Notch serves as receptor and the transmembrane protein Delta as its ligand (Muskavitch, 1994; Artavanis-Tsakonas et al., 1995). Delta in one cell activates Notch in its neighbour, thereby delivering a lateral inhibitory signal – that is, inhibiting the neighbour from becoming committed to neural differentiation. The neurogenic genes, and in particular Delta, are under the control of the proneural genes and, in Drosophila, are expressed at the outset throughout the proneural cluster (Haenlin et al., 1990; Kooh et al., 1993; Hinz et al., 1994; Kunisch et al., 1994); thus it appears that all the cells in the cluster initially both deliver inhibition to one another and receive it. To explain how the system advances to a state where one cell – the prospective neural cell – escapes from inhibition, while its neighbours remain inhibited, it has been proposed that the activity of Delta itself is regulated by lateral inhibition, so that the more inhibition a cell receives from its neighbours, the less it is able to deliver back to them (Goriely et al., 1991; Heitzler and Simpson, 1991). This gives rise to a positive feedback loop that will tend to amplify any initial difference between neighbouring cells: it makes lateral inhibition a competitive process, allowing a single winner to emerge in each small region, delivering inhibition to its neighbours but receiving none back from them (Sternberg, 1993; Chitnis, 1995; Collier et al., 1996; Lewis, 1996).

Evidence to support this feedback model has come from analysis of the pattern of production of neural cells at the borders of mutant clones in the Drosophila epidermis (Heitzler and Simpson, 1991). Moreover, in the nematode worm, where a homologous system of genes serves to single out one cell for a special fate in the developing vulva, the process has been shown to involve a lateral inhibition feedback loop in which the Delta homologue (lag-2) is regulated at the RNA level (Wilkinson et al., 1994). In Drosophila, the mode of regulation of Delta seems to vary from tissue to tissue (Parks et al., 1995; Heitzler et al., 1996; de Celis and Bray, 1997; Seugnet et al., 1997). There is a defined molecular pathway operating in peripheral neurogenesis through which activation of Notch can inhibit transcription of Delta (Heitzler et al., 1996), but direct observations of Delta expression in the fly’s embryonic central nervous system, by in situ hybridisation or with antibodies, fail to show the expected pattern of regulation in proneural clusters (Kooh et al., 1993), suggesting that the regulation of Delta activity by lateral inhibition is post-translational or dependent on some ancillary factor, or that prospective neural cells in the embryonic CNS are marked out by some prior molecule that makes them resistant to lateral inhibition (Seugnet et al., 1997). Recent work has shown that birds, amphibians and mammals possess homologues of Delta and Notch, and that neurogenesis is regulated, as in the fly, by Delta-Notch signalling (Chitnis et al., 1995; Lewis, 1996; Henrique et al., 1997a). There is, however, a contrast with Drosophila in one important respect: in the vertebrate examples studied so far, Delta1 expression has been detected only in the nascent neurons, and not in their neighbours in the neurogenic region (Chitnis et al., 1995; Henrique et al., 1995; Myat et al., 1996). This raises the question whether there is a fundamental difference between Drosophila and vertebrates in the way in which the Delta genes are regulated and in the mechanism by which neural cells are singled out for their fate.

In this paper, we report the cloning of four Delta homologues from the zebrafish, and examine their function in primary neurogenesis – that is, in the formation of the earliest neurons in the CNS. We show that three of the genes, deltaA, deltaB and deltaD, are expressed in the neurogenic regions. But, in contrast with previous studies in vertebrates, their expression there is not restricted to the prospective primary neurons: deltaA and deltaD are expressed widely, in large groups of contiguous cells at sites of neurogenesis. Within these groups, some cells express the genes at higher levels, and these cells, which express deltaB also, can be identified as prospective neurons. By functional tests we show, furthermore, that the fish delta genes control primary neurogenesis through lateral inhibition, and that this inhibition operates in each case with feedback based on regulation of delta expression at the RNA level, providing a competitive Delta-Notch mechanism to single out the cells that are to differentiate as neurons.

Zebrafish eggs were obtained by natural spawnings from a colony of fish derived from stock from ‘The Goldfish Bowl’ pet shop, Oxford, England. Eggs were collected and maintained at 28.5°C in system water or E3 embryo medium (Haffter et al., 1996) with 10⁻⁵% methylene blue to inhibit fungal growth. Embryos were staged according to Kimmel et al. (1995); embryonic ages are given in hours postfertilization (hpf) at 28.5°C.

Zebrafish Delta/Serrate homologues were cloned by PCR, as described by Henrique et al. (1995), using similar primers, targeted to the DSL domain and an adjacent EGF repeat. The initial reaction used cDNA made from a 24 hour zebrafish random-primed cDNA library (gift of U. Strähle). Degenerate oligonucleotide primers TTCTGCCGICCICGIGA(C/T)GA(C/T) and TCIATGCAIGTIC- CICC(A/G)TT were used. 40 cycles of amplification yielded products varying in length from approximately 300 to 700 base pairs. Amplified fragments were cloned into pBluescript KS- (Stratagene) and sequenced. Two of the PCR fragments were used to screen a 20- 28 hour zebrafish λZAP cDNA library (made by Robert Riggleman and Kathryn Helde, a gift from David Grunwald), yielding clones that corresponded to four distinct delta genes. cDNAs were excised from the λZAP II vector using Stratagene’s Rapid Excision Kit, and subjected to double-stranded sequencing in both directions using ³⁵S, with the Pharmacia T7 polymerase kit and double-stranded nested- deletion kit. Sequences were analysed using the Wisconsin GCG computer programs.

Digoxigenin- or fluorescein-labelled RNA antisense probes were generated with a Stratagene RNA transcription kit. Enzymes forlinearisation and transcription for probe synthesis were as follows: islet-1 (Inoué et al., 1994) – XbaI and T3; paxb (Krauss, 1991) – BamHI and T7; krox20 (Oxtoby and Jowett, 1993) – XbaI and T3; and notch (Bierkamp and Campos-Ortega, 1993) – XbaI and T7. Clones for deltaA, deltaB and deltaD were cut with EcoRI and transcribed with T7, while the clone for deltaC was cut with XbaI and transcribed with T7.

Whole-mount in situ hybridisation followed Oxtoby and Jowett (1993) with minor modifications. Two-colour whole-mount in situ hybridisation was carried out essentially as described by Hauptmann and Gerster (1994).

To show Mauthner cells, embryos at 30 hpf were stained with the 3A10 monoclonal antibody (Furley et al., 1990) as described by Hatta (1992).

Specimens were stored in PTW (PBS + 0.1% Tween20) with 0.1% sodium azide at 4°C and were photographed in 100% glycerol, either intact or as flat mounts dissected off the yolk and flattened under a coverslip. Images were taken using a Leitz Diaplan microscope or an MRC-600 confocal microscope (BioRad) and assembled into figures using Adobe Photoshop.

Embryos were injected at the 1- to 4-cell stage, with 200 pl of a solution of 20-50 ng/µl RNA in water, unless otherwise stated (see Tables 2, 3 and 4), containing 0.2% phenol red. After injection, the embryos were allowed to develop until the 5- to 10-somite stage. Embryos injected with lacZ mRNA were fixed for 30 minutes at room temperature in 4% paraformaldehyde in PTW. After three washes in PTW, the embryos were stained for β-galactosidase activity by incubation for 30 to 90 minutes at 37°C in 400 µg/ml X-gal, 4 mM potassium ferricyanide, 4 mM potassium ferrocyanide, 4 mM MgCl₂, in PTW. The X-gal staining was not allowed to develop for more than 90 minutes in order to limit RNA degradation and embryo damage, and to avoid masking the in situ staining. After X-gal staining, the embryos were fixed overnight at 4°C and then dehydrated in methanol and processed for whole-mount in situ hybridisation as described above. Embryos not injected with lacZ mRNA were fixed overnight at 4°C in 4% paraformaldehyde in PBS and then dehydrated in methanol and processed for whole- mount in situ hybridisation.

To make sense RNA for injections, X-Delta1, X-Delta1^dn and X-NotchΔE constructs (Coffman et al., 1993; Chitnis et al., 1995) were linearized by NotI and transcribed with SP6 RNA polymerase. A plasmid corresponding to an XbaI-NotI lacZ open reading frame with a nuclear localisation signal (a gift of Jonathan Pearce), inserted into the CS2 vector (Turner and Weintraub, 1994), was linearized with Acc65I and transcribed with SP6 RNA polymerase. The deltaB transcription construct was made by replacing the NsiI-XhoI fragment containing part of the 3’ UTR of deltaB in Bluescript by a NsiI-SmaI fragment containing beta-globin 3’ sequences and poly(A) sequences from pSP64T (Krieg and Melton, 1984). The construct was linearized by AccI and transcribed with T3. In vitro preparation of capped mRNAs was performed with the MegascriptTM kit (Ambion).

By screening a zebrafish 24-hour cDNA library by PCR with primers based on the Drosophila Delta gene (see Materials and Methods), we identified fragments of four distinct zebrafish Delta homologues. We called these deltaA, deltaB, deltaC and deltaD. An exchange of sequence data revealed that our deltaA and deltaD were identical with genes independently discovered by B. Appel and J. Eisen in Oregon (Appel and Eisen, 1998) and also (in the case of deltaD) by P. Dornseifer, C. Takke, and J. Campos-Ortega in Köln (Dornseifer et al., 1997). While we have determined the complete cDNA sequence of deltaB (GenBank accession number AF006488), they have determined that of deltaA and deltaD, and have kindly allowed us to include their sequence data in our Fig. 1, for comparison with deltaB. The full sequence of deltaC will be presented elsewhere; as discussed below, it is not expressed in the neural plate during primary neurogenesis.

The deduced sequences of the proteins DeltaA (765 amino acids), DeltaB (616 amino acids) and DeltaD (718 amino acids) all have a size and domain structure similar to that of the Drosophila Delta protein: there is a large (>500 amino acids) N-terminal region corresponding to the extracellular portion of Delta, a short hydrophobic region corresponding to the transmembrane segment, and a C- terminal region corresponding to the intracellular portion of Delta. Sequence conservation is clear throughout the extracellular region (46-47% amino-acid identity with Delta for all three fish genes); this region includes a typical strongly conserved DSL (Delta-Serrate-Lag2 homology) domain (Tax et al., 1994) and 8 EGF repeats (as compared with 9 for Drosophila Delta). The intracellular domain, however, shows a striking lack of conservation (≤17% amino-acid identity with Drosophila Delta). The contrast between extracellular and intracellular domains in their degree of conservation is also evident when we compare the fish Delta proteins with one another and with the chick C-Delta1 protein (Henrique et al., 1995). Thus DeltaA and DeltaD, which are most similar to one another, show 80% identity in their extracellular regions but only 47% identity in their intracellular regions; they both show 77% identity to C-Delta1 extracellularly, but, respectively, 47% and 67% identity to it intracellularly.

DeltaB is more divergent: in all three regions – extracellular, transmembrane and intracellular – it differs from DeltaA and DeltaD by roughly the same amount as it does from C-Delta1, and more than DeltaA or DeltaD does from C-Delta1. This suggests that it may have arisen through a gene duplication/divergence event predating the divergence of fish from tetrapods. Its extracellular domain is 54-57% identical to that of DeltaA, DeltaD and C-Delta1, while its intracellular domain (75 amino acids) is much shorter than theirs and shows only 18-27% identity. As shown in Table 1, DeltaB bears a slightly closer resemblance to the Xenopus Delta2 protein (Jen et al., 1997) (59% identity overall) than to Delta1 (52%). All three fish delta genes are as much diverged from mouse Delta-like-3 (Dll3) (Dunwoodie et al., 1997) as they are from Drosophila Delta.

The high degree of extracellular sequence conservation among the members of the Delta family reflects the known importance of this region for binding to Notch and possibly other proteins (Lieber et al., 1992); the variability of the intracellular region, evident in virtually all pairwise comparisons of one Delta with another, suggests that this part of the molecule has a less critical function, or one that is less dependent on specific binding to other proteins.

Expression of all four genes – deltaA, deltaB, deltaC and deltaD – begins to be visible by in situ hybridisation during epiboly (Fig. 2A-D). deltaC and deltaD are both strongly expressed in the germ ring, throughout the region of involuting mesoderm, with the exception of the axial (midline) region – the region of the prospective notochord. Paired bilateral transverse stripes of expression, first of deltaC and then of deltaD, are seen a little more anteriorly, foreshadowing the formation of somites. As epiboly comes to an end, the mesodermal expression of deltaC and deltaD resolves into a strong tail-bud domain and, more anteriorly, bilateral pairs of stripes probably corresponding to somites that are about to form. Low-level expression persists subsequently in recently formed somites as well as in the presomitic mesoderm, at least up to 24 hpf (Figs 3-5): deltaC is expressed in the posterior parts of somites, deltaD in the anterior parts (Fig. 5). In their mesodermal expression in relation to somite development, deltaC and deltaD thus resemble the Delta1 gene of mouse, chick and frog (Bettenhausen et al., 1995; Chitnis et al., 1995; Henrique et al., 1995; Dornseifer et al., 1997; Hrabé de Angelis et al., 1997), the Delta2 gene of Xenopus (Jen et al., 1997), and the Dll3 gene of the mouse (Dunwoodie et al., 1997). The fish deltaA and deltaB genes, by contrast, do not appear to be expressed in the developing mesoderm at these early stages. The development of the mesoderm and its dependence on deltaD are discussed by Dornseifer et al. (1997) and will not be considered further here.

Our main concern here is with the developing nervous system. By 90% epiboly (9 hpf), transcripts of deltaA, deltaB and deltaD are seen in the epiblast (the future neurectoderm), in or near the axial midline and in two anterolateral pairs of patches (Fig. 2A,B,D). The patterns for the three genes appear similar but not identical (although the rapidity of change during epiboly makes it difficult to make precise comparisons between separately labelled specimens): for example, deltaB expression is seen in the midline while deltaA is expressed just lateral to this. Comparison with later stages suggests that the anterolateral patches correspond to the future sites of origin of the primary sensory neurons of the trigeminal ganglion and of the posterior CNS (the Rohon-Beard cells), and that the medial bands of deltaA expression flanking the midline correspond to the domains within which primary motor neurons will arise. Meanwhile, we detect no expression of deltaC in the prospective neural regions. Expression of the zebrafish notch gene (Bierkamp and Campos-Ortega, 1993), corresponding to tetrapod Notch1, is ubiquitous, but strongest in the dorsal midline region (Fig. 2E).

By the 1-somite stage (10.3 hpf) (Fig. 2F-L), the neural plate is well defined and paraxial stripes of expression of deltaA, deltaB and deltaD are seen, marking sites of primary neurogenesis. At the 5-somite stage (11.7 hpf), the patterns of gene expression are similar but more sharply delineated; neurogenesis has begun, as indicated by expression of the neuronal marker islet1 (Korzh et al., 1993; Inoué et al., 1994). We have chosen this stage (Fig. 3) for detailed analysis. We used the expression domains of paxb (at the midbrain- hindbrain junction) and of krox20 (in rhombomeres 3 and 5) as landmarks, shown in red by a double in situ hybridisation protocol. A small set of scattered cells expressing islet1 just lateral to the paxb domain could be identified as nascent placode-derived neurons of the trigeminal ganglion. The bands of islet1-expressing cells flanking the midline posterior to the krox20 domains could be identified as nascent primary motor neurons, while those lying laterally, at the margins of the posterior neural plate, could be identified as nascent Rohon- Beard cells. At all these sites, deltaA, deltaB and deltaD are also expressed, and in larger numbers of cells than express islet1. Taking these patterns in conjunction with those seen earlier and later, we conclude that expression of deltaA, deltaB and deltaD foreshadows the onset of neuronal differentiation as indicated by islet1. The three delta genes are also expressed in certain regions of the neural plate where islet1 expression is absent; these correspond to sites of differentiation of primary interneurons that do not express islet1, such as the Mauthner cells, other hindbrain reticulospinal cells and the primary interneurons of the spinal cord, lying in rows just medial to the Rohon-Beard cells (Figs 3, 5) (Mendelson, 1986). There is still no noticeable expression of deltaC in the neural plate: it will begin to be expressed strongly during secondary neurogenesis, especially in the developing retina (Fig. 4C). notch expression, on the other hand, is still widespread, although more intense in regions of delta gene expression (see also Bierkamp and Campos-Ortega, 1993).

We have used the 5-somite stage to examine whether the fish delta genes are expressed in nascent neurons only, like Delta1 in tetrapods, or more diffusely, in proneural clusters, like Delta in Drosophila. Figs 3 (right-hand panels) and 5 show detailed comparative views of the expression of deltaA, deltaB, deltaD and islet1 probes. While the domains of expression are broadly similar, there are marked differences at the level of the individual cells. islet1 and deltaB are expressed strongly and selectively in scattered cells that are mostly isolated from one another by non-expressing cells (Fig. 5E- H,J). deltaA and deltaD, meanwhile, are expressed more diffusely, in patches comprising many contiguous cells; in some of these cells the expression is strong, in others weak (Fig. 5I,K). The cells expressing deltaA most strongly appear scattered, while those expressing deltaD strongly occur more as clusters, but both are concentrated in the neurogenic regions where cells expressing deltaB are found.

Double labelling reveals that the cells within a deltaA patch that express deltaA strongly are precisely the cells that express deltaB (Fig. 5A-D); these double-labelled cells lie next to the midline (the site of origin of primary motor neurons), or at or just medial to the lateral edges of the deltaA domain (the sites of origin of Rohon-Beard cells and primary interneurons, respectively). Almost all (92%; 713 cells out of 777 counted, 6 embryos) of the deltaB-expressing cells express deltaA. Double labelling with deltaB and islet1 (Fig. 5E-H) shows, furthermore, that a large proportion (46%; 233 cells out of 503 counted, 7 embryos) of the deltaB-expressing cells also express islet1, while almost all (84%; 233 cells out of 278 counted, 7 embryos) of the islet1-expressing cells also express deltaB. The cells that express deltaB but not islet1 lie just medial to the presumptive Rohon-Beard cells and, from their location and time of appearance, are presumably primary interneurons. We conclude, in short, that cells expressing deltaB are nascent neurons, and that the nascent neurons are a subset of the cells expressing deltaA and deltaD.

To test whether signalling by Delta protein does indeed regulate commitment to a neural fate in the fish, we injected mRNA coding for different forms of the Delta protein into zebrafish embryos at the 1- to 4-cell stages, and observed the distribution of the islet1-positive cells in these embryos at the 5- to 10-somite stages. In some experiments, E. coli lacZ mRNA was coinjected as a marker for the presence of injected RNA, its protein product being detected by X-gal histochemistry. In most of the injected embryos, the X-gal staining was unevenly distributed and, in some cases, it was unilateral, which allowed comparison of the islet1 staining on the affected (sky blue) and control (white) sides.

To overactivate the Delta-Notch signalling pathway, we first injected mRNA coding for the full-length Xenopus Delta1 protein (Chitnis et al., 1995). This caused a reduction in the numbers of islet1-positive presumptive neurons of all classes (Fig. 6A,B). A similar result was obtained after injection of an mRNA coding for the full-length zebrafish DeltaB protein (Fig. 6E,F; Table 2). Similar experiments by Appel and Eisen (1998) and by Dornseifer, Takke and Campos-Ortega (Dornseifer et al., 1997) have likewise demonstrated inhibition of neurogenesis following injection of mRNA coding for DeltaA and DeltaD. The effects on neurogenesis seen with deltaB RNA were, however, much weaker than those with X-Delta1 RNA and were accompanied, at high doses, by distortions of the neural plate suggestive of disturbances occurring during gastrulation (see Tables 2 and 3).

For comparison with the effects of the Delta proteins, we injected RNA coding for NotchΔE, an extracellularly truncated form of X-Notch1 that is constitutively active and has been found in analogous experiments in Xenopus both to inhibit neurogenesis and to cause distortions of the neural plate (Coffman et al., 1993; Chitnis et al., 1995). The effects of injecting NotchΔE RNA in the fish were similar, resembling those of deltaB RNA (Table 3). The relatively weak inhibition of neurogenesis seen in both cases may reflect relatively short half-lives of these RNAs.

To test the effect of blocking the Delta-Notch signalling pathway, we injected RNA coding for the truncated Delta protein X-Delta1^dn (formerly known as X-Delta1^STU), which lacks most of its intracellular domain and is known in other species to have a dominant-negative action (Chitnis et al., 1995; Sun and Artavanis-Tsakonas, 1996): it acts by making the cells that express it insensitive to lateral inhibition (Henrique et al., 1997a). The result was a striking increase in the number of islet1-positive cells (Fig. 6C,D; Tables 2, 3). Whereas on the control side, these occurred in isolation or in groups of two or three, in corresponding regions on the affected side they occurred in densely packed patches. The powerful effect of X-Delta1^dn RNA injection was greatly reduced by coinjecting NotchΔE RNA (Table 3), as expected if X-Delta1^dn acts by preventing activation of endogenous Notch.

Note that injection of RNA coding for a molecule that blocks lateral inhibition (X-Delta1^dn) does not have, and would not in general be predicted to have, effects that are the precise inverse of those of RNA coding for an activating molecule (X-Delta1, DeltaB or X-NotchΔE). There are two reasons. First, the RNAs may have different lifetimes. Second, the effects of the blocking molecule are predicted to be irreversible and hence more striking, because cells that prematurely escape inhibition become irreversibly committed to neuronal differentiation; whereas the effects of the activating molecule are reversible and temporary – once the RNA and protein are degraded or diluted away, neuroepithelial cells that were inhibited can differentiate belatedly, so that the final outcome appears more nearly normal.

The above data, taken as a whole, match the results obtained from analogous experiments in Xenopus embryos (Chitnis et al., 1995), and likewise indicate that Delta-Notch signalling mediates lateral inhibition in primary neurogenesis.

In fish and amphibians, a unique giant neuron, the Mauthner cell, is normally generated on each side of the hindbrain, in rhombomere 4. The Mauthner cell is the first to be born of all the neurons in the body (at 7.5 hpf – Mendelson, 1986), and its genesis poses the problem of singling-out in the purest form: what mechanism guarantees that on each side of the brain precisely one cell, and no more, will be assigned this special fate? Analogies with Drosophila would suggest that positional signals might first define a small proneural cluster and that lateral inhibition, mediated by Delta-Notch signalling, might then operate within this group of potential Mauthner cells to pick out one of them. If so, forced overactivation of the Delta- Notch pathway should prevent Mauthner cell development, while artificial blockade of the pathway should enable a cluster of cells, instead of one solitary individual, to develop as Mauthner cells. We tested these predictions by injecting either X-Delta1 or X-Delta1^dn RNA into an early blastomere and staining with the 3A10 antibody to reveal Mauthner cells at 30 hpf (Fig. 7). Out of 22 embryos injected with X-Delta1 RNA, 6 had no Mauthner cells, 11 had only one and 5 still had two (one on each side). Conversely, out of 78 embryos injected with X-Delta1^dn RNA, at least 24 had more than two Mauthner cells, while 54 appeared to have just the usual two. Additional Mauthner cells when present often lie clustered very close together and are not easy to distinguish, so that these figures for the proportion of cases with additional Mauthner cells may be an underestimate.

The foregoing experiments indicate that lateral inhibition mediated by Delta-Notch signalling is required to prevent neighbours of nascent primary neurons from developing as primary neurons. But is this mechanism also responsible for singling out the primary neurons in the first place, through a competitive process of the type outlined in the Introduction? If so, we should expect a feedback regulation of Delta activity by Delta activity: activation of the Delta-Notch signalling pathway in a given cell should inhibit the production of Delta activity in that cell.

To test this, we analysed the expression of the deltaA, deltaB and deltaD genes in embryos injected with X-Delta1 or X- Delta1^dn RNA together with a lacZ marker. After injection of X-Delta1^dn mRNA, the deltaA (Fig. 8A), deltaB (Fig. 8B) and deltaD (Fig. 8C) stainings were both stronger and more extensive on the injected side of the embryo, as compared to the control side. Conversely, after injection of X-Delta1 mRNA, the deltaA (Fig. 8D), deltaB (Fig. 8E) and deltaD (Fig. 8F) stainings were reduced on the injected side. These results, summarised in Table 4, show that activation of the Notch signalling pathway by a Delta protein represses transcription of delta genes and that a blockade of Notch activation permits delta genes to be expressed at higher levels and in a larger number of cells than normal. We infer that deltaA, deltaB and deltaD do indeed participate in a feedback loop of the type hypothesised.

In embryos injected with X-Delta1^dn RNA (Fig. 8A-C), the overexpression of delta genes extends over a region significantly broader than the immediate neighbourhood of the normal set of primary neurons: it is not only the cells that would normally be in direct contact with the prospective primary neurons that show signs of having been released from an inhibition. The normal low-level expression of delta genes in cells other than the prospective primary neurons is therefore physiologically significant: where its effects are blocked, cells are enabled to display neuronal characteristics inappropriately.

One Delta gene has so far been described in the chick (Henrique et al., 1995), two in the mouse (Bettenhausen et al., 1995; Dunwoodie et al., 1997) and two in Xenopus (Chitnis et al., 1995; Jen et al., 1997). In the zebrafish, we have discovered four. Of these four fish genes, the closest counterpart of the tetrapod Delta1 gene is deltaD: it has the most similar sequence, and, like Delta1, is involved both in primary neurogenesis and in somite formation (Dornseifer et al., 1997; Hrabé de Angelis et al., 1997). deltaC, by contrast, is expressed in the developing somites but not during primary neurogenesis; in its expression pattern, at least, it resembles the Xenopus X- Delta2 gene (Jen et al., 1997). deltaA and deltaB are expressed in regions of primary neurogenesis, like deltaD, but not in the developing somites; they do not have precise counterparts among the tetrapod Delta genes described so far.

We have focused on the roles of deltaA, deltaB and deltaD in primary neurogenesis. We have shown that they have overlapping but significantly different expression patterns in the neural plate: deltaA and deltaD are expressed in patches of contiguous cells, while deltaB is apparently confined to the scattered cells within those patches that differentiate as neurons.

It is tempting to suggest that the deltaA/deltaD expression pattern comes first, and that the deltaB pattern emerges later, but we have not been able to resolve the temporal sequence: the first signs of expression of all three genes are first seen at a similar time, during epiboly; and at subsequent stages (see Fig. 2), as landmarks become clearer in the neural plate, we see the same nested pattern of expression described above for the 5-somite stage. If there is a delay from the onset of expression of deltaA and deltaD in a given cell to the onset of expression of deltaB, it is unlikely to be more than an hour or so.

deltaB expression persists longer in relation to the onset of cell differentiation than does Delta1 in tetrapods, since it is coexpressed with islet1 in developing primary neurons – something that is not seen in chick or Xenopus (Chitnis et al., 1995; Henrique et al., 1995 and unpublished observations). At later stages, however, a large proportion of islet1-expressing cells lack deltaB expression (data not shown), suggesting that deltaB does eventually switch off in these cells. As expected, expression of deltaB correlates with withdrawal from the cell cycle: in BrdU pulse-labelling experiments at the 5- to 6- somite stage (data not shown), 95% of the deltaB-expressing cells were unlabelled with BrdU, indicating that they had already finished their final S-phase (although not necessarily their final mitosis). Occasional deltaB-expressing cells (5%; 45/827 cells counted, 6 embryos) were, nevertheless, labelled with BrdU; similarly, a small proportion of islet1-expressing cells were labelled. Mitoses that are not completed until after neuronal commitment may explain why, although most primary neurons occur as isolated individuals, some are found in small clusters of two or three contiguous cells (see Fig. 5).

deltaA, deltaB and deltaD all show the characteristic signature of genes whose products deliver lateral inhibition: they are all normally expressed most strongly in the nascent neurons, yet the effect of artificially forcing expression at a high level in all the cells of a region of the neuroepithelium is to inhibit neurogenesis (Dornseifer et al., 1997; Appel and Eisen, 1998; and present data). Conversely, when the Delta-Notch pathway is blocked by an injection of X-Delta1^dn RNA, neurons are produced in excessive numbers and no longer as isolated individuals. The Mauthner cell provides a striking example: normally a single one is generated on each side of the brain, but when the Delta-Notch signalling pathway is blocked, several are produced in a cluster and, when the pathway is overactivated, none are produced (Fig. 7). The overproduction of Mauthner cells and of motor and sensory neurons seen after X-Delta1^dn RNA injection resembles that seen (in a moreextreme form) in the mindbomb (white tail) neurogenic mutant (Jiang et al., 1996; Schier et al., 1996), reinforcing the suggestion that the mindbomb gene codes for a component of the Delta-Notch signalling pathway.

Since the product of each of the fish delta genes deltaA, deltaB and deltaD has an effect similar to that of X-Delta1 or NotchΔE, and since the Notch-binding extracellular domain is highly conserved in all of them, it seems very likely that each of them is capable of activating Notch. deltaA, deltaB and deltaD not only have products with similar activities; they are also themselves all regulated similarly in their expression by the level of activation of the Delta-Notch signalling pathway – activation of the pathway drives their expression down, and inactivation of the pathway allows their expression to rise. We infer that in zebrafish embryos the product of each of these three delta genes can act on neighbouring cells to repress its own expression and the expression of the other two also. This finding implies the existence of a positive feedback loop that will tend to amplify differences between adjacent cells, as explained in the Introduction; it implies also that the levels of expression of the three delta genes should all be raised or lowered together, as observed.

The regulation of delta expression in one cell by Delta activity in its neighbour creates a competitive relationship between the adjacent cells. Provided the regulatory effect is steep enough, so that a change in the Delta activity in one cell causes an even larger change of Delta activity in its neighbour, the feedback loop in the control of delta expression is sufficient to render unstable a uniform state where all cells express the genes and all inhibit one another, and to drive the system towards a state where some cells express the delta genes strongly and deliver inhibition, while their neighbours receive inhibition and express the genes only weakly (Collier et al., 1996). In this way, the demonstrated properties of deltaA and deltaD, with their widespread early expression, are sufficient to explain how, through competition among the cells of a neurogenic patch, isolated cells become singled out to develop as primary neurons, even if the cells in the patch are initially all equivalent.

There is, however, a proviso: we have not proved that the feedback effects for deltaA and deltaD are steep enough by themselves to destabilise the uniform state of mutual inhibition. deltaB may have a special role in this respect. Its expression – strong in the future neurons, undetectable in their neighbours – suggests that it is much more steeply sensitive to the level of Notch activation than is the expression of deltaA or deltaD, and that it may thereby take the state of uniform mutual inhibition over the brink of instability and drive the emergence of isolated cells as winners of a competition based on lateral inhibition.

A Delta protein may exert effects not only by activating Notch, but also by interfering directly with the actions of other Delta proteins by binding to them (Fehon et al., 1990). Indeed, we have suggested elsewhere (Henrique et al., 1997a) that Delta1^dn may act in this way to make cells deaf to lateral inhibition: Delta1^dn may bind to and sequester endogenous Delta proteins on neighbouring cells, so that they are no longer available to activate Notch. It is possible that DeltaB in nascent neurons has a similar effect: by binding to the Delta proteins on neighbouring cells, it may help to protect the nascent neurons from lateral inhibition. Experiments to explore this possibility are in progress.

Although our findings imply that the selection of individual cells as primary neurons depends on a Delta-mediated competition, it remains possible that the competition is biased from the outset. Some cells may have an advantage over their neighours by virtue of higher starting levels of expression of proneural genes such as achaete/scute or atonal homologues (Cubas et al., 1991; Goriely et al., 1991), or larger endowments of proteins such as Numb (Zhong et al., 1996). In the early chick neural plate, for example, the initial level of expression of the achaete/scute homologue Cash4 varies markedly from cell to cell in an apparently random fashion (Henrique et al., 1997b). Where two adjacent cells by chance both start with a similar strong advantage, both may progress to become committed to neuronal differentiation before competition mediated by lateral inhibition has time to be fought through to a conclusion that favours one cell over the other. A mechanism of this sort could account for the occasional groups of two or more contiguous primary neurons that are seen in the zebrafish (if these are not all the result of cell division, as discussed earlier, or of cell migration bringing cells into contact after they have become committed). Indeed, one can envisage a continuum of possibilities, from the extreme where the primary neurons are determined entirely by prior factors, with lateral inhibition acting subsequently and serving only to prevent undetermined neighbours of these cells from also becoming primary neurons, to the opposite extreme where all cells are equivalent at the outset and lateral inhibition with feedback is the all-important mechanism in making them different.

We began this paper by pointing out a clash between the proposal, based on work in Drosophila and C. elegans, that cells within proneural clusters are singled out to become neurons by a competitive mechanism based on Delta-Notch signalling, and the observation in chick and Xenopus that Delta1 is expressed only in the prospective neurons. It may be that these cells in chick and Xenopus are indeed specified by some prior mechanism, independent of Delta genes, as just discussed, and that Delta-Notch signalling serves only to prevent the neighbours of these predetermined cells from developing as neurons at the same time. Flies, tetrapods and fish may all control neurogenesis differently, and there may be important differences between primary and secondary neurogenesis. But the present observations on the zebrafish suggest at least two other possibilities. First, tetrapods may have other homologues of Delta or Serrate, yet to be characterised, that are expressed like the zebrafish deltaA or deltaD in all the cells of a neurogenic patch and mediate competitive lateral inhibition, with Delta1 behaving like zebrafish deltaB to signal the outcome of the competition. Second, it is conceivable that in situ hybridisation may have given an incomplete picture of the expression of Delta1 in tetrapods: we have argued that, in the fish, the low-level expression of deltaA and/or deltaD in cells other than the prospective primary neurons is functionally important, and there is no guarantee that in situ hybridisation is sensitive enough in all cases to detect levels of delta expression that are sufficient to activate Notch. The Drosophila paradigm, in which neurogenesis begins with a proneural cluster all of whose members express Delta and Notch and thereby compete for a neural fate, may therefore apply also, with minor variations, not just to the zebrafish but to vertebrates in general.

We are indebted to Tom Schilling, Trevor Jowett, Patrick Blader, Uwe Strähle, David Grunwald, Jonathan Pearce, Chris Kintner, Bruce Appel and José Campos-Ortega for DNA templates and other reagents. We thank Tanya Whitfield, Judith Eisen, Julie Cooke and Steve Wilson for advice on techniques; Stephen Massey and Toby Simmonds for looking after the fish; Frédéric Rosa for welcoming S. S.-M. in his laboratory for some of the injection experiments; and David Ish-Horowicz, Alastair Morrison, Yun-Jin Jiang and other members of our laboratories for many discussions and for comments on the manuscript. We are especially grateful to Bruce Appel and Judith Eisen and to José Campos-Ortega and colleagues for generously sharing data and allowing us to see their manuscripts before publication. The first three authors of this paper contributed equally to the work. We thank the Imperial Cancer Research Fund, INSERM, the European Science Foundation, and the EU HCMP fund for support.