Regulation of neuronal specification in the zebrafish spinal cord by Delta function

During vertebrate neurogenesis, the neuroepithelium is patterned with the anterior region specified as brain and the posterior region as spinal cord. Subsequently, many distinct types of neurons differentiate at particular times and places within the developing central nervous system (CNS). The problem we address in this paper is how cells within the spinal cord are specified for different neuronal fates.

Specification, the process by which cellular identity is established during development (Kimmel et al., 1991), of ventral spinal cord cells results from inductive signaling. Axial mesoderm, which underlies the developing spinal cord and differentiates as notochord, induces neural tissue to form floorplate, a specialized cell type in the ventral midline (Placzek et al., 1990; Yamada et al., 1991). Both notochord and floorplate can induce formation of motoneurons (Yamada et al., 1991, 1993) and their absence, in mice and zebrafish, correlates with absence of motoneurons (Bovolenta and Dodd, 1991; Beattie et al., 1997). The signal for both the notochord and floorplate responsible for patterning ventral spinal cord appears to be Sonic Hedgehog (Shh), a secreted glycoprotein, which is initially expressed in axial mesoderm and later in floorplate and is necessary and sufficient to induce floorplate and motoneuron development (reviewed by Tanabe and Jessell, 1996).

Floorplate and motoneurons are just two of the diverse cell types in the ventral spinal cord and the question remains as to how this diversification is achieved. Shh might be a morphogen, since relatively high concentrations induce both floorplate and motoneurons, while lower concentrations induce only motoneurons (Roelink et al., 1995) and still lower concentrations induce interneurons dorsal to motoneurons (Ericson et al., 1997). Blockage of Shh signaling during a critical late phase of Shh requirement causes generation of interneurons rather than motoneurons, suggesting that different neuronal types have different requirements for Shh signaling (Ericson et al., 1996). Thus, graded distribution of Shh within ventral spinal cord may instruct a variety of different fates. Sequential signaling may contribute to spinal cord patterning as motoneurons are required for differentiation of specific interneurons (Pfaff et al., 1996).

We are investigating whether interactions among prospective neurons in ventral spinal cord are important for their specification by testing neuronal specification in zebrafish embryos, in which distinct neurons develop at characteristic positions and times (Kimmel and Westerfield, 1990). For example, motoneurons can be categorized as primary and secondary (Myers, 1985). Primary motoneurons are born before secondary motoneurons, are larger and fewer in number, and are spared in ned-1 mutant embryos in which secondary motoneurons degenerate (Grunwald et al., 1988). Individual primary motoneurons can be identified by differences in cell body position, axon projection and gene expression (Eisen et al., 1986; Inoue et al., 1994; Appel et al., 1995; Tokumoto et al., 1995). Additionally, various types of individually identifiable interneurons develop in the ventral spinal cord close to motoneurons (Bernhardt et al., 1990, 1992).

In Drosophila melanogaster, cellular interactions, mediated by proteins encoded by the so-called neurogenic genes Notch and Delta, operate to specify cells for different fates. For example, Notch and Delta are required to specify appropriate numbers of neuroblasts and epidermoblasts from ventral neuroectoderm (reviewed by Campos-Ortega, 1995). In the absence of Notch or Delta function, too many neuroblasts develop at the expense of epidermoblasts. Delta and Notch appear to function as ligand and receptor, respectively (Muskavitch, 1994). In a process called lateral inhibition (Simpson, 1990) or lateral specification (Greenwald and Rubin, 1992; Artavanis-Tsakonas et al., 1995; but see Muskavitch, 1994), high levels of Delta expression in specified neuroblasts are thought to activate Notch in neighboring cells, causing them to develop as epidermoblasts rather than neuroblasts (Campos-Ortega, 1995).

Notch and Delta homologues have been identified in Caenorhabditis elegans and a variety of vertebrate species (Artavanis-Tsakonas et al., 1995). Notch proteins that are constitutively active in the absence of ligand suppress neurogenic and myogenic developmental programs in cell culture and vertebrate embryos (Coffman et al., 1993; Kopan et al., 1994; Nye et al., 1994; Lardelli et al., 1996) as does overexpression of a full-length Delta protein (Chitnis et al., 1995). Conversely, expression of a dominant negative (dn) form of Delta in Xenopus laevis embryos causes too many cells to adopt neuronal fates (Chitnis et al., 1995). These observations are consistent with a role for vertebrate Delta and Notch proteins in mediating lateral inhibition, suggesting this is a conserved mechanism for specifying cells for different fates.

In D. melanogaster, neuroectodermal cells have either neural or epidermal fates. In contrast, all cells within the vertebrate neuroectoderm are destined for neural development, either as neurons or glia. Thus, lateral inhibition could regulate specification of neural cells as different types of neurons and glia. Although work in X. laevis showed that Delta-mediated lateral inhibition is important for controlling the number of neural plate cells that develop as neurons (Chitnis et al., 1995), the role of lateral inhibition in regulating neuronal identity in the developing spinal cord is unknown. We have extended previous work by testing how lateral inhibition contributes to specification of identified spinal cord neurons in zebrafish embryos. We report identification of a novel zebrafish Delta homologue, deltaA, expressed specifically in developing nervous system. deltaA is transiently expressed in cells specified for neuronal fates. Overexpression of full-length DeltaA suppresses neuronal specification and expression of a dn Delta causes too many cells to adopt neuronal fates. By examining the effect of dn Delta on neuronal identity, we find evidence that lateral inhibition is important for specification of different types of motoneurons and interneurons in the ventral spinal cord. Furthermore, our results suggest that in the ventral spinal cord, as in the D. melanogaster and vertebrate retinas (Cagan and Ready, 1989; Fortini et al., 1993; Austin et al., 1995; Dorsky et al., 1995, 1997), the time at which cells are competent to respond to inductive signals influences their specification for particular neuronal fates.

Approximately 1×10⁶ plaques of a lamba ZAP-II cDNA library constructed from 15-19 h (hours at 28.5°C) zebrafish embryos were screened at low stringency using a X. laevis X-Delta-1 cDNA (Chitnis et al., 1995; gift of C. Kintner) as probe to isolate zebrafish delta cDNAs. cDNA inserts were subcloned by helper phage cotransfection (Stratagene) and restriction enzyme mapping revealed two classes of cDNA, deltaA, described here, and deltaD (see Haddon et al., 1998). The longest representative of each class was sequenced, on both strands, with a Perkin-Elmer Applied Biosystems Automated DNA Sequencer Model 377 using FS dye terminator chemistry. Sequences were assembled and analyzed using the DNASTAR Lasergene package. The GenBank accession number for deltaA is AF030031.

Embryos from the University of Oregon laboratory colony were raised at 28.5°C. Staging was according to Kimmel et al. (1995).

In situ RNA hybridization was conducted as described previously (Thisse et al., 1993) except that the probes were not hydrolyzed. For RNA-antibody double-labeling experiments, embryos were fixed in 4% paraformaldehyde and first processed for antibody labeling using α-Isl mAb 39.4D5 (Tsuchida et al., 1994), which was obtained from the Developmental Studies Hybridoma Bank, followed by the peroxidase anti-peroxidase antibody system (Sternberger Monoclonals, Inc.). A brown precipitate was formed by incubating embryos in 0.5 mg/ml diaminobenzidine (DAB) and 0.001% H₂O₂. Embryos were then refixed and processed for in situ RNA hybridization. Embryos were mounted in glycerol and photographed using a Zeiss Axioplan and Kodak T160 film. For additional antibody labeling experiments, the following primary antibodies were used: rabbit anti-Hu (gift of Linda Hansen and Jim Weston), zn5 (Trevarrow et al., 1990) and rabbit anti-GABA (Sigma). Fluorescent secondary antibodies were obtained from Jackson ImmunoResearch Laboratories, Inc. Fluorescent images were collected using a Zeiss LSM 310 confocal microscope, assembled with VoxelView (Vital Images) and processed using Adobe Photoshop.

A BamHI-XhoI fragment of deltaA cDNA was subcloned into the CS2 vector (Turner and Weintraub, 1994; gift of D. Turner). This subclone eliminates the first methionine translation start codon of the long open reading frame and is predicted to encode an extracellular domain similar in size and sequence to other vertebrate Delta proteins (see Results and Haddon et al., 1998). CS2-X-Delta-1^STU (Chitnis et al., 1995) was a gift of C. Kintner and CS2-nucβ-gal was a gift of D. Turner. Plasmids were linearized with NotI and capped mRNA was synthesized using SP6 RNA polymerase and the mCAP kit (Stratagene). The integrity of in vitro synthesized mRNA was assessed using ethidium-agarose gel electorophoresis and the concentration determined by spectrophotometry.

deltaA and X-Delta-1^STU mRNA at 50-300 ng/µl and lacZ mRNA at 300 ng/µl were mixed with Phenol Red (Sigma), to allow visualization of injections, at a concentration of 0.01%. Injections of approx. 5 nl were made, using air pressure, into 1- to 8-cell-stage embryos. Injection was into the center of the yolk, from which cytoplasmic streaming distributes mRNA to all cells, or directly into a single cell. This restricts distribution of mRNA to descendants of the injected cell, representing a subset of all embryonic cells, as revealed by staining for β-galactosidase activity. Often, mRNA is restricted to the left or right side of the embryo. β-galactosidase activity was detected after fixing embryos in 4% paraformaldehyde for 1 hour at room temperature or overnight at 4°C by incubation in 4% 5-bromo-4-chloro-indoxyl-β-D-galactoside, 150 mM NaCl, 1mM MgCl₂, 1.5 mM K₄[Fe₃(CN)₆], 1.5 mM K₃[Fe₂(CN)₆], in 5 mM sodium phosphate buffer (pH 7.3) at 37°C. Alternatively, β-galactosidase was detected using an anti-β-galactosidase antibody (Promega).

Using a probe synthesized from X. laevis X-Delta-1 (Chitnis et al., 1995) at low stringency, we cloned two zebrafish delta genes, deltaA, described here, and deltaD. The structures of deltaA and deltaD are presented more fully in an accompanying paper (Haddon et al., 1998) and deltaD has been recently described (Dornseifer et al., 1997). Briefly, both genes are predicted to encode proteins having large extracellular domains as well as transmembrane and intracellular domains. The putative extracellular domain of each protein is highly similar to those of Delta-like proteins identified in fly, nematode and other vertebrates as each has the DSL domain and EGF-like repeats characteristic of this group (Artavanis-Tsakonas et al., 1995; Henrique et al., 1995). If translation is assumed to initiate at the first methionine codon of the long open reading frame, then deltaA is predicted to encode a protein of 803 amino acids. Alignment with other Delta proteins reveals that DeltaA would have an additional 38 N-terminal amino acids (see Haddon et al., 1998). However, the nucleotide sequence at this putative start site, ACCAATG, does not match a consensus translation initiation sequence, CANCATG, assembled for vertebrates (Cavener, 1987). In contrast, the sequence at the second in-frame methionine codon, CATCATG, matches the consensus translation start sequence. Translation initiation at this site would eliminate the 38-amino-acid N-terminal extension, resulting in an extracellular domain equivalent in size to those of other Delta proteins. The intracellular domains are quite diverged. The DeltaA intracellular domain is 58 amino acids longer than that of DeltaD, which makes it similar in size to the D. melanogaster Delta intracellular domain. However, these sequences have little similarity to one another (see Haddon et al., 1998).

We examined the expression patterns of deltaA and deltaD by in situ RNA hybridization. deltaA expression is initiated in the neuroectoderm before that of deltaD (data not shown). We focus here on deltaA expression in the developing trunk neural plate and neural tube. In zebrafish, cells that give rise to primary neurons of the trunk begin to exit the mitotic cycle as gastrulation is completed (Kimmel and Westerfield, 1990). The first of these known to differentiate are primary sensory Rohon Beard neurons (RBs) (Lamborghini, 1980; Bernhardt et al., 1990) and primary motoneurons (Myers, 1985), which arise from lateral and medial regions of the neural plate, respectively. deltaA expression is initiated in the epiblast prior to completion of gastrulation (not shown). At the 2- to 3-somite stage (10.5 h) low levels of deltaA RNA are distributed throughout the trunk CNS, with cells expressing higher levels found in the medial and lateral regions of the neural plate (Fig. 1A). These regions correspond to the positions at which primary motoneurons and RBs originate. Cells expressing high levels of deltaA RNA do not form contiguous domains. Rather, single cells or small clusters of several cells showing high expression are interspersed with cells having lower expression. deltaA expression is specific to the developing nervous system (Fig. 1B,C), and continues to be expressed broadly in the CNS throughout neurogenesis (Figs 1C, 2E). The expression of deltaA falls within the spatial and temporal expression profile described for zebrafish notch genes (Bierkamp and Campos Ortega, 1993; Westin and Lardelli, 1997). The correlation of deltaA and notch expression is consistent with the possibility that in zebrafish, as in other animal embryos, Delta and Notch proteins interact to regulate specification of cell fate.

Double labeling shows that deltaA is expressed in cells specified for neuronal fates. Soon after gastrulation is completed, and 6-8 hours before axogenesis, primary motoneurons and RBs are marked by an antibody raised against rat Islet-1 protein (α-Isl) (Korzh et al., 1993). We refer to these cells, marked by gene expression prior to axogenesis, as presumptive primary motoneurons (pPMNs) and presumptive Rohon-Beard neurons (pRBs). We first detect α-Isl reactivity in pPMNs and pRBs at about the 2- to 3-somite stage. Fig. 2A shows a dorsal view of the neural keel, at the position of the third and fourth somites of a 4-somite stage (11.3 h) embryo probed for Isl protein and deltaA message. α-Isl-reactive pPMNs express high levels of deltaA RNA. In slightly older embryos, most pPMNs at the same position no longer express deltaA RNA (Fig. 2B) while more posterior, later born pPMNs do (data not shown). Similarly, at early stages pRBs express deltaA RNA (Fig. 2C) but later, at the same axial position, most do not (Fig. 2D). These observations suggest that deltaA is rapidly downregulated following neuronal specification. Cells that express only deltaA RNA are also evident in medial and lateral neural plate. This may result from deltaA expression preceding Isl expression and from deltaA expression in cells specified for other neuronal fates. At 24 h, α-Isl labels cells in ventrolateral spinal cord that are probably fated to become secondary motoneurons (Appel et al., 1995). Many, but not all, of these cells express deltaA, consistent with the proposed transient nature of deltaA expression (Fig. 2E). At 24 h, and throughout later embryogenesis, deltaA is broadly expressed in the spinal cord, suggesting that it is expressed by many types of cells (Fig. 2E; data not shown). We conclude that deltaA is expressed as neuronal specification occurs and subsequently downregulated in cells that have acquired specific neuronal fates.

Recent studies suggest that vertebrate Delta and Notch genes regulate neurogenesis (Coffman et al., 1993; Austin et al., 1995; Chitnis et al., 1995; Dorsky et al., 1995, 1997; Dornseifer et al., 1997). To initiate our investigation into the function of delta genes in neuronal specification, we tested the effects of full-length DeltaA protein and a truncated form of X. laevis Delta-1 protein (dnDelta), which acts in dominant negative fashion (Chitnis et al., 1995), on neurogenesis in zebrafish using a probe for huC RNA as a marker of cells undergoing neuronal differentiation (Kim et al., 1996). Fig. 3A shows the expression pattern of huC in a 6-somite-stage control embryo injected, at early cleavage stage, with lacZ mRNA, which we use as a lineage tracer to reveal distribution of injected mRNAs. huC is expressed in discontinuous rows of cells in medial and lateral neural keel showing that some, but not all, cells in these regions are specified for neuronal development at this stage. Embryos coinjected with mRNAs encoding β-galactosidase and full-length DeltaA protein have fewer huC-expressing cells coincident with distribution of lineage tracer (Fig. 3B; Table 1), suggesting that high levels of DeltaA activity inhibit neurogenesis. At later stages of development, deltaA-injected embryos have fewer than normal secondary motoneurons, suggesting that deltaA overexpression generally inhibits neuronal specification (data not shown). In contrast, embryos injected with mRNA encoding dnDelta show an increased density of cells expressing huC spanning the normal spatial domains of huC expression (Fig. 3C; Table 1). To test if dnDelta interferes with DeltaA function, we injected both mRNAs together. In most embryos, the DeltaA inhibition of neurogenesis was suppressed, resulting in near normal or excess number of neurons (Table 1). Thus, dnDelta appears to disrupt DeltaA function. These data suggest that DeltaA, expressed in cells that have acquired neuronal fate, inhibits neuronal specification in neighboring cells. Disruption of Delta signaling activity, by dnDelta expression, results in failure of this lateral inhibition leading to specification of neuronal fate in excess cells. These observations are similar to those obtained by misexpression of full length and dominant negative X-Delta-1 proteins in X. laevis embryos (Chitnis et al., 1995).

To learn if Delta-mediated lateral inhibition regulates specification of neuronal identity in the spinal cord, we examined embryos expressing dnDelta with markers specific for ventral neural tube fates. First, we found that neuralplate-stage embryos have an increased number of medially located cells expressing islet1 RNA, a marker of motoneuronal fate (Inoue et al., 1994; Appel et al., 1995) (Fig. 4A,B). Next, we determined the extent to which ventral spinal cord cells are caused to adopt motoneuronal fates when Delta function is disrupted. Fig. 4C shows an uninjected 6-somite-stage embryo probed with α-Isl, to label presumptive motoneurons, and α-Hu, which labels all differentiating neurons (Marusich et al., 1994) to reveal the early pattern of neuronal specification. pPMNs form two single-cell-width columns flanking the floorplate. Within these columns, pPMNs are arranged in a regular, periodic fashion (Appel et al., 1995). α-Isl reactivity precedes that of α-Hu; thus, Isl-positive cells are occasionally unlabeled by α-Hu. Some cells are Hu-positive but Isl-negative; other cells express neither marker. This suggests that, during early stages of neurogenesis, the developing trunk ventral neural keel of zebrafish includes cells specified for primary motoneuronal fate (Isl⁺Hu⁺), cells specified as neurons but for non-motoneuronal fates (Isl⁻Hu⁺) and cells that are not yet specified as neurons (Isl⁻Hu⁻). In contrast, in embryos expressing dnDelta, all neuronal cells near the floorplate are Isl-positive and thus pPMNs (Fig. 4D). These cells are limited to rows one or two cells wide bordering the floorplate. α-Hu labeling shows that dnDelta also causes overproduction of neuronal cells more distant from floorplate; however, these cells are unlabeled by α-Isl and, thus, are not motoneurons. These observations indicate that Delta function is required to establish the appropriate number and distribution of primary motoneurons. Disruption of Delta function results in formation of supernumerary motoneurons, which appear to arise at the expense of other cell types that are normally interspersed among them.

To learn the identities of motoneurons generated when Delta function is perturbed, we examined islet2 expression, which normally marks only one or two specific primary motoneurons per hemisegment, CaP and VaP (Appel et al., 1995) (Fig. 4E). In contrast, we find clusters of 3-7 islet2positive cells in the ventral spinal cords of X-Delta-1^STUinjected embryos (Fig. 4F), suggesting that failure of lateral inhibition leads to overproduction of primary motoneurons. Furthermore, islet2-positive cells are limited to clusters midway between somite boundaries at the positions normally occupied by CaP and VaP motoneurons. This shows that, as revealed by islet2 expression, supernumerary primary motoneurons acquire identities appropriate for their location, supporting our previous results showing that spinal cord gene expression and motoneuronal identity are positionally specified (Eisen, 1991; Appel et al., 1995). Additionally, this result implies that the positional information that specifies identity of individual primary motoneurons is intact in the absence of normal Delta function. A similar conclusion has been drawn for neuroblast specification in flies, in which it was shown that the identity of neuroblasts in the peripheral nervous system is independent of the neurogenic loci (Goriely et al., 1991).

We examined the effect of dnDelta expression on development of secondary motoneurons using the zn5 monoclonal antibody (Trevarrow et al., 1990), which recognizes the DMGRASP protein (Fashena, 1996). In control embryos, zn5 labels secondary motoneuron cell bodies and axons (Fashena, 1996) (Fig. 5A). dnDelta expression greatly reduces the number of secondary motoneurons in embryos showing widespread distribution of lineage tracer in ventral spinal cord (28 of 40 embryos). Fig. 5B shows an embryo in which lineage tracer was distributed to clusters of cells in the spinal cord. zn5 labeling is evident outside these clusters but is largely absent from cells expressing lineage tracer. Additionally, the ventral roots appear to be reduced in size, suggesting that fewer secondary motoneuron axons contributed to the nerve. Thus, embryos expressing dnDelta have too many primary motoneurons and too few secondary motoneurons. One possible explanation for this finding is that cells normally specified as secondary motoneurons develop as primary motoneurons in the absence of Delta function.

Two types of interneurons, KA and VeLD, positioned close to motoneurons (Fig. 6A), can be identified with antibody specific to the neurotransmitter GABA (Bernhardt et al., 1992). KA neurons lie adjacent to the floorplate, ventromedial to motoneurons. VeLD neurons are more dorsal, intermixed with motoneurons. In X-Delta-1^STU-injected embryos with lineage tracer distributed broadly throughout the ventral spinal cord, KA and VeLD interneurons are reduced in number or absent (7 of 16 embryos; Fig. 6B). Thus, disruption of Delta-mediated lateral inhibition results in overproduction of primary motoneurons with corresponding decreases in secondary motoneurons, KA interneurons and VeLD interneurons. These observations indicate that Delta function is required for specification of a variety of neuronal fates in zebrafish ventral spinal cord.

To learn how different kinds of neurons are specified in the ventral spinal cord, we asked two questions. First, do zebrafish neurons influence the fates of neighboring cells through Delta/Notch-mediated lateral inhibition? Second, what is the consequence for ventral spinal cord cell fates when lateral inhibition is disrupted?

The lateral inhibition model stems from observations of bristle formation on the cuticle of Rhodnius prolixus made by Wigglesworth (1940), who concluded that bristles prevent formation of similar structures within a certain radius. When bristles were removed, they were regenerated by cells that would not normally give rise to bristles, suggesting that many cells have potential for bristle development but that, typically, only a subset of cells express this potential. Interactions which specify equivalent cells for different fates are now recognized as an important patterning mechanism (Greenwald and Rubin, 1992). In D. melanogaster, lateral inhibition operates to specify cells for neuroblast or epidermoblast fates in the embryonic CNS and the adult PNS; in the absence of lateral inhibition, mediated by neurogenic genes, too many neuroblasts develop at the expense of epidermoblasts (reviewed by Simpson, 1990; CamposOrtega, 1995). A simple prediction of the lateral inhibition model is that the neurogenic gene Delta, which functions cell-nonautonomously, is expressed at higher levels in neural cells than in neighboring non-neural cells. Indeed, genetic mosaic analysis indicated that cells tend to be specified for neural fate when they have a greater dosage of Delta relative to adjacent cells (Heitzler and Simpson, 1991). However, Delta protein is expressed uniformly throughout cell populations undergoing fate specification and is not evident in delaminated neuroblasts (Kooh et al., 1993). This was taken as evidence to support a model of mutual inhibition (Goriely et al., 1991) in which every cell of an equivalence group is prevented from adopting a neural fate and only those cells that escape the inhibitory field undergo neural development (Muskavitch, 1994).

Zebrafish deltaA is expressed broadly throughout the neural plate and, later, spinal cord. However, this expression is nonuniform as some cells express deltaA at higher levels than others. Double labeling shows that early induced cells specified for primary motoneuron and RB fates express high levels of deltaA, and that deltaA is later expressed by cells specified for secondary motoneuron fate as well as by other types of cells. deltaA expression appears to be quickly downregulated in these cells soon after their birth. Thus, in many ways, expression of deltaA is consistent with predictions of the lateral inhibition model. In actuality, the patterns of deltaA expression and neuronal differentiation do not conform precisely to the model. In many instances, adjacent cells express high levels of deltaA. Likewise, adjacent cells may develop as primary motoneurons or RBs (see Figs 1A, 3A, 4C). Perhaps, as a result of cell movements during neurogenesis, cells that are specified for neuronal fates are brought into contact with one another. Additionally, in zebrafish, some cells may be specified for neuronal fates before completion of mitosis (B. Appel and K. Stoesser, unpublished results; Haddon et al., 1998). Thus, small clusters of adjacent neurons might arise, circumventing lateral inhibition of similar fate (see Haddon et al., 1998).

To date, four zebrafish Notch homologues have been described (Bierkamp and Campos-Ortega, 1993; Westin and Lardelli, 1997). Although we have not directly compared them, the expression patterns of zebrafish notch genes and deltaA appear to coincide within the developing nervous system. Thus, zebrafish notch and delta genes are expressed at the appropriate times and places to mediate lateral inhibition in the CNS.

If DeltaA acts as an inhibitory signal, increasing its level of expression should prevent neuronal specification. When we inject mRNA encoding full-length DeltaA protein into embryos, we find that the number of cells expressing a panneuronal marker is greatly reduced. This result is identical to that obtained when X-Delta-1 is overexpressed in frog embryos (Chitnis et al., 1995). Indeed, X-Delta-1 overexpression in zebrafish embryos also inhibits neuronal specification (B. Appel, unpublished observations) as does overexpression of zebrafish DeltaD (Dornseifer et al., 1997; B. Appel, unpublished observations). At least three zebrafish delta genes are expressed in the developing nervous system (this work; Dornseifer et al., 1997; Haddon et al., 1998). Thus, lateral inhibition may be mediated by multiple Delta family members. Comparison of delta gene expression patterns and identification of mutations that affect individual delta genes will reveal if these genes function within different subpopulations of cells in the CNS and whether they have similar or overlapping functions in the same cells.

Recently, two zebrafish mutations have been described, whitetail (Jiang et al., 1996) and mind bomb (Schier et al., 1996), which result in overproduction of early specified neurons in zebrafish. These phenotypes are similar to some of those we describe here arising from expression of dnDelta. Thus, the whitetail and mind bomb mutations, which are allelic, may identify a component of the Delta/Notch signaling pathway. It will be interesting to further compare the mutant phenotype with the dnDelta phenotype as well as to learn the identity of the affected gene.

Neuronal diversity is often attained through a stereotypical sequence of cell differentiation. For example, in each ommatidium of the D. melanogaster retina, specific cells develop in a precise order. Inductive signaling is important for specification of these cell types and early differentiating cells appear to recruit later-differentiating cells to the ommatidium (Tomlinson and Ready, 1987). Inactivation of Notch function, by means of a temperature-sensitive Notch mutation, at progressively later times during retinal development affects the fates of progressively later-developing cells (Cagan and Ready, 1989), suggesting that Notch function is required as each fate is specified. Expression of a ligand-independent, activated Notch causes retinal cells to adopt inappropriate fates, perhaps as a consequence of delaying their differentiation (Fortini et al., 1993). Thus, Notch activity may regulate the ability of cells to respond to inductive signaling.

Likewise, in the vertebrate retina, different cell types develop in a particular sequence. Lineage analysis of retinal cells suggests that progenitors are multipotent and that environmental signals contribute to specification (reviewed by Cepko et al., 1996). Reduction of Notch activity increases the number of cells that adopt the earliest specified fate, that of retinal ganglion cells, while overexpression of Delta or of activated Notch reduces ganglion cell number (Austin et al., 1995). Consistent with the lateral inhibition model, when cells overexpressing X-Delta-1 have normal neighbors, they take earlier fates, while dominant negative X-Delta-1 causes nearly all expressing cells also to take the earliest specified fates of ganglion cells and cone photoreceptors (Dorsky et al., 1997). Together, these observations suggest that disruption of Delta/Notch signaling allows excess cells to be specified for early fates and that Delta and Notch functions normally regulate the number of retinal cells specified for any particular fate.

In X. laevis embryos, expression of activated Notch results in neural hypertrophy (Coffman et al., 1993). Notch activation appears to delay cell differentiation and an increased number of progenitor cells commit to neural development, leading to the proposal that Notch signaling regulates the number of cells that can respond to any given inductive signal.

In the zebrafish spinal cord, regulation of primary and secondary motoneuron specification fits well with this model. Primary and secondary motoneurons are born beginning about 9-10 h and 14-15 h, respectively (Myers et al., 1986), at least one cell cycle apart (Kimmel et al., 1994). Zebrafish shh expression is initiated in presumptive dorsal mesoderm at about 7 h (Krauss et al., 1993). In cyclops; floating head double mutant embryos, which lack differentiated notochord and floorplate, shh expression is initiated normally but not maintained after gastrulation and primary, but no secondary motoneurons develop (Beattie et al., 1997). Thus, early shh signaling may induce development of primary, but not secondary, motoneurons. In chick, motoneuronal induction requires exposure to Shh for 1-2 cell cycles (Ericson et al., 1996). The zebrafish cell cycle length during early neurogenesis is about 4 hours (Kimmel et al., 1994). Thus, secondary motoneurons are born 1-2 cell cycles after initiation of shh expression. Together, these observations suggest that secondary motoneurons are equivalent to chick motoneurons in their requirement for Shh signaling, but that primary motoneurons require only brief exposure. We have shown here that, in the absence of lateral inhibition, too many primary motoneurons develop concomitant with the loss of secondary motoneurons. Thus, we propose that Delta-Notch signaling specifies primary and secondary motoneuronal fates by regulating how neural precursor cells respond to Shh signaling. In this model, some cells of the neural plate respond to Shh by immediately developing as primary motoneurons. DeltaA is expressed at high levels in these cells and inhibits neighboring cells from responding to Shh in the same way. Later, as DeltaA is downregulated in primary motoneurons, neighboring cells are released from lateral inhibition and respond to Shh by adopting secondary motoneuronal fate. Alternatively, secondary motoneurons may be specified by a late-arising signal that acts with, or subsequent to, Shh. In either case, by controlling the number of cells that respond to inductive signals, Deltamediated lateral inhibition facilitates specification of distinct types of motoneurons in the zebrafish spinal cord.

We do not know how specification of KA and VeLD interneurons fits into these models, as we know less about when and where they are born with respect to motoneurons. These neurons are anti-GABA reactive beginning at about 18 h (Bernhardt et al., 1992; B. Appel, unpublished observations). Thus, they may be born at the same time as primary motoneurons or the earliest secondary motoneurons. Perhaps Shh is able to specify a variety of fates by virtue of a cell’s position within a Shh gradient as well by length of exposure to Shh (Ericson et al., 1996). Additionally, early differentiating neurons may express signals that instruct the fates of laterdifferentiating ones, a possibility demonstrated by the requirement of motoneurons for development of a specific interneuron in mice (Pfaff et al., 1996). Either way, Delta/Notch signaling, by mediating lateral inhibition, controls the number of cells competent to respond to any particular instructive signal, thus diversifying neuronal fates in the vertebrate ventral spinal cord.

We are extremely grateful for gifts of plasmids and antibody from Chris Kintner, David Turner, Linda Hansen and Jim Weston. We also thank Abbie Jensen and Charles Kimmel for their thoughtful comments on the manuscript. Thanks to Michele McDowell for sectioning, to the University of Oregon Zebrafish Facility staff for fish rearing, and to Tracy Norris for sequencing, which was supported by an NSF Shared Instrumentation Grant BIR 9512344 and a grant from the MJ Murdock Charitable Trust. The 39.4D5 antibody, developed by Thomas Jessell, was obtained from the Developmental Studies Hybridoma Bank maintained by the Department of Pharmacology and Molecular Sciences, Johns Hopkins University School of Medicine, Baltimore, MD, and the Department of Biological Sciences, University of Iowa, Iowa City, IA, under contract NO1-HD-6-2915 from the NICHD. This work was supported by NIH grants HD07658 and NS23915.