Mouse Dll3: a novel divergent Delta gene which may complement the function of other Delta homologues during early pattern formation in the mouse embryo

The supply and patterning of tissue and cellular differentiation within that tissue are fundamental to the development of an organism. For example, during the initial stages of gastrulation in the mouse, the epiblast, a pluripotent and rapidly dividing epithelium, supplies tissue in the form of the newly induced mesoderm. The mesoderm is then patterned leading to a large diversity of differentiated tissue types. Throughout gastrulation and organogenesis these processes are repeated as the complexity of the embryo increases. These events are not restricted to the embryo but are also required in the adult during tissue replenishment. A signalling pathway first described in Drosophila has been implicated in tissue supply, patterning and cellular differentiation. The signal is supplied by Delta and received in the neighbouring cells by the receptor, Notch. The Delta gene encodes a transmembrane protein (Vassin et al., 1987; Heitzler and Simpson, 1991) which affects the development of adjacent cells displaying the transmembrane receptor, Notch (Wharton et al., 1985).

In Drosophila, a well-documented case where Delta/Notch signalling is required is during the production of neural precursors (see Artavanis-Tsakonas, 1995). A single neuroblast is produced from a precursor population (equivalence group) even though each cell in that population is capable of becoming a neuroblast. This production of individual neuroblasts occurs repeatedly from the equivalence groups until a full complement of neuroblasts has been derived. In each case, the cell that becomes the neuroblast signals via Delta to its neighbours and the activation of Notch in these cells prevents them from adopting a neural fate. This cell-cell dependent inhibitory signalling is referred to as lateral inhibition. Loss-of-function mutations in either of the corresponding genes, results in the production of as many neuroblasts as there are cells in the precursor population. In the peripheral nervous system (PNS), Delta/Notch signalling is required again to divert cell fate during differentiation (Hartenstein and Campos-Ortega, 1986; Hartenstein and Posakony, 1990; Spana and Doe, 1996; Guo et al., 1996).

In Drosophila, Delta and Notch are pivotal in numerous cell fate choices during both embryonic and adult life. In the embryo, homozygous null mutations for Delta or Notch result in the expansion of one cell type at the expense of another. In ectoderm discussed above, the embryonic nervous system is expanded while the embryonic epidermis is reduced (Lehman et al., 1983). In mesoderm, an expanded myoblast population is apparent (Corbin et al., 1991). Similarly a phenotype exists in the endoderm lineage (Tepass and Hartenstein, 1995). In the adult, additional roles for Delta and Notch include specification of bristle precursors throughout the adult epidermis, cell type specification in the adult eye and differentiation of follicle cells during oogenesis (see Muskavitch, 1994). Genes homologous to Delta and Notch are present in C. elegans where glp-1 and lag-2 are required to allocate precursors of specific vulval fates (Greenwald et al., 1983). These pleiotropic functions appear to reflect the recurrent use of these genes in many scenarios where cell fate specification is occurring.

Vertebrate homologues of Notch have been identified in Xenopus (X-Notch-1), chick (C-Notch-1) and mouse (Notch-1, Notch-2, Notch-3, Notch-4) (Coffman et al., 1990; Henrique et al., 1995; Franco del Amo et al., 1992; Lardelli and Lendahl, 1993; Lardelli et al., 1994; Uyttendaele et al., 1996). Like Drosophila and C. elegans, it is apparent that the Notch signalling pathway influences cell fate specification in vertebrates. Constitutively activated forms of Notch-1, which lack the extracellular portion of the protein, inhibit neurogenesis in Xenopus (Chitnis et al., 1995; Dorsky et al., 1995; Coffman et al., 1993). In addition, activated Notch inhibits muscle formation in Xenopus as well as in mouse cells in vitro and drives T lymphocyte lineage selection in the mouse (Kopan et al., 1994; Robey et al., 1996).

More recently, vertebrate homologues of Delta have been identified in Xenopus (X-Delta-1, X-Delta-2), chick (C-Delta-1) and mouse (Dll1) (Chitnis et al., 1995; Henrique et al., 1995; Bettenhausen et al., 1995; Jen et al., 1997). The Delta-1 and Dll1 genes appear to be orthologues due to conservation of amino acid sequence and gene expression. During embryo-genesis, the predominant regions of Delta-1/Dll1 gene expression are the neuroectoderm and the presomitic mesoderm. In the neuroectoderm, the punctate patterning of C-Delta-1 expression foreshadows the spatiotemporal pattern of neuronal differentiation. Cells that have exited from the cell cycle transiently express C-Delta-1 prior to differentiating into neurons (Henrique et al., 1995; Myat et al., 1996). Ectopic expression of Delta-1 in Xenopus and chick inhibits the production of neurons in the neural plate and retina (Chitnis et al., 1995; Dorsky et al., 1997; Henrique, unpublished), presumably due to the ubiquitous activation of Notch-1. In addition, genetic analysis in the mouse demonstrates that Delta/Notch signalling is required during somitogenesis, since null mutant mice for Notch-1 (Swiatek et al., 1994; Conlon et al., 1995) and Dll1 (Hrabe de Angelis et al., 1997) exhibit defects in somite formation.

Here we report the identification of a novel vertebrate homologue of Drosophila Delta. This gene Dll3 has been isolated in the mouse and is the third vertebrate Delta homologue to be identified. The Dll3 gene, like Dll1/Delta-1, is predominantly expressed in the neuroectoderm and paraxial mesoderm during embryogenesis. We show that Dll3-expressing cells in the neuroepithelium, like those that express Dll1 are postmitotic and that Dll3, like X-Delta-1, can inhibit primary neurogenesis in Xenopus. We suggest that, in the mouse, Dll3 and Dll1 cooperate in the formation of somite boundaries during segmentation of the paraxial mesoderm. We therefore argue that Dll3 and Dll1 function together in several contexts during mouse embryogenesis.

The PS-(Ect+End) subtracted cDNA library was generated according to the procedure described in Harrison et al. (1995). Here, single-stranded DNA was generated from the Primitive Streak library and subtracted with biotinylated RNA generated from the Ectoderm and Endoderm libraries. The subtracted PS-(Ect+End) library was hybridised to 10,000 gridded clones from the Primitive Streak library according to Harrison et al. (1995). Hybridising clones were sequenced and those representing novel cDNAs were used to generate antisense riboprobes for RNA in situ hybridisation.

Total RNA was isolated from embryonic tissue and ES cells according to Chomczynski and Sacchi (1987). Poly(A)+ RNA was isolated using the poly(A)Tract system (Promega). CGR8 ES cells (Mountford et al., 1994) were cultured according to Wilson et al. (1993). A DNA fragment representing 470 bp of Dll3 3′UTR was generated by PCR (1760-2225 bp), gel purified using QIAEX gel extraction (QIAGEN) and ³²PdCTP incorporated using a Random-Primed DNA Labelling Kit (Boehringer Mannheim). RNA was denatured and size fractionated (1% agarose, 2.2 M formaldehyde), transferred to a nylon membrane (Hybond-N+), cross linked to the membrane by exposure to UV light and hybridised according to Sambrook et al. (1989) using 1×10⁶ cts/minute/ml of hybridisation buffer.

Embryos were collected from timed C57BL6 ×DBA matings. Noon on the day of appearance of the vaginal plug is designated 0.5 dpc. Embryos were dissected from the uterus and Reichert’s membrane removed as described by Beddington (1987) in M2 medium (Hogan et al., 1994) containing 10% fetal calf serum (FCS, Advanced Protein Products) instead of bovine serum albumin. Embryos for whole-mount RNA in situ hybridisation were rinsed in phosphate-buffered saline (PBS) and fixed with 4% paraformaldehyde (PFA) in PBS. Embryos for RNA isolation were immediately dissolved in denaturing solution (Chomczynski and Sacchi, 1987).

Antisense riboprobes were transcribed according to Harrison et al. (1995). For Dll3, riboprobes were derived from PS93 (1075-2243 bp) or Delta-S (1-2102 bp). Whole-mount RNA in situ hybridisation was carried out according to Wilkinson (1992) using the hybridisation conditions of Rosen and Beddington (1993). The length of proteinase K treatment varied; 5.5-6.5 dpc (5 minutes), 7.0-8.5 dpc (10 minutes), 9.0 dpc and older (15 minutes.) Embryos were postfixed in 4% paraformaldehyde, 0.1% glutaraldehyde in PBS. RNA in situ hybridisation of cryosections were performed according to Myat et al. (1996). Double in situ hybridisation was achieved as follows. Synthesised riboprobes incorporating DIG-UTP or fluorescein-UTP were detected with alkaline phosphatase (AP)-conjugated anti-DIG or anti-fluorescein antibodies. The substrate for AP was either BCIP/NBT (blue) or BCIP/INT (brown; Boehringer Mannheim). Following the AP reaction in the first instance, the precipitate was fixed with 4% PFA in PBS, the AP activity destroyed with 0.1 M glycine in water, pH 2.2 (2×5 minutes, cryosection; 2×15 minutes, whole mount). The embryos or cryosections were blocked again prior to incubation with the next antibody.

After whole-mount RNA in situ hybridisation, embryos were processed for paraffin sectioning by dehydration through an ethanol series, clearing in Histoclear (National Diagnostics) and embedding in paraffin wax (Histoplast, m.p. 56°C). Sections were dewaxed in Histoclear (5 minutes) and mounted under coverslips in DPX mountant (BDH). Embryos for cryosectioning were fixed with 4% PFA/PBS at 4°C overnight, cryoprotected with 30% sucrose/PBS at 4°C overnight.

10.5-13.5 dpc mice were injected with 100 μg BrdU per gram body weight according to Miller and Nowakowski (1988) for 30-120 minutes. Embryos were fixed in 4% PFA/PBS and cryosectioned as above. RNA in situ hybridisation was performed on the cryosections using the Fast Red fluorescent alkaline phosphatase substrate (Boehringer Mannheim). After staining, slides were washed in PBS and processed for BrdU immunodetection (Biffo et al., 1992). Islet-1/2 protein was detected as follows. Following RNA in situ hybridisation and staining using the Fast Red substrate slides were washed in PBS and blocked. Anti-BrdU (1/50; Beckton Dickinson) and antiislet-1/2 (1/50; Developmental Studies Hybridoma Bank maintained by the Department of Pharmacology and Molecular Sciences, Johns Hopkins University School of Medicine, MD 21205, and the Department of Biological Sciences, University of Iowa, Iowa City, IA 52242, under contract N01-HD-2-3144 from the NICHD) were detected using FITC-conjugated goat anti-mouse secondary antibody (1/40; Cappel). Composite fluorescent images were obtained using a Bio-Rad MRC600 confocal microscope.

Dll3 cDNA was cloned into CS2+ vector (Turner and Weintraub, 1994). Capped RNA for injection was generated according to Kreig and Melton (1984). Capped lacZ RNA was generated from pSP6nucbgal (Smith and Harland, 1991). Synchronous embryos were obtained by artificial fertilisation, maintained in 10% Normal Amphibian Medium (NAM: Slack, 1984) at 14-18°C and staged according to Nieuwkoop and Faber (1975). Embryos for microinjection were transferred to 75% NAM containing 4% Ficoll Type 400 (Sigma). 200 pg of Dll3 and 50 pg of lacZ RNA was injected in a volume of 10 nl into a single blastomere at the 2- to 4-cell stage. RNA was injected into an animal blastomere on the prospective dorsal side of the embryo in order to direct expression of the RNA into the neural tissue. 2 hours after injection embryos were returned to 10% NAM.

At neural plate stage (stage 13-15), the vitelline membrane was removed and embryos fixed in MEMFA (Harland, 1991) for 1 hr at 4°C. Embryos were washed 3×20 minutes in wash buffer and stained at 37°C for 1 hour. X-gal wash and stain solutions were as described in Beddington et al. (1989). Embryos were postfixed in MEMFA at 4°C overnight and dehydrated. Whole-mount RNA in situ hybridisation was performed essentially as described by Harland (1991).

Following DNA sequencing and whole-mount RNA in situ hybridisation analysis of clones hybridising to the PS-(Ect+End) subtracted probe, several cDNAs corresponding to genes expressed in mesoderm and the primitive streak during gastrulation were identified. Of the six novel genes that were identified in this screen, one proved to be a homologue of the Drosophila gene Delta.

Dll3 cDNA consists of 2243 bp and contains an open reading frame (ORF) of 1755 bp (Fig. 1). This ORF encodes a protein of 585 amino acids and contains several features that identify it as a homologue of Drosophila Delta. At the N terminus, a stretch of 32 hydrophobic amino acids indicate a signal sequence. Another hydrophobic region is located between amino acids 489-513 representing a membrane-spanning region of the protein. Like Drosophila Delta, the extracellular domain contains EGF-like repeats and a cysteine-rich domain, the DSL (Tax et al., 1994), which is required for binding to the Notch receptor (Fehon et al., 1990). There are six EGF-like repeats encoded by the Dll3 cDNA with a gap consisting of 29 amino acids placed between the first and second repeats. This represents fewer repeats than are present in Drosophila Delta (9 repeats) (Kopczynski et al., 1988) and Dll1 (8 repeats) (Bettenhausen et al., 1995). The Dll3 protein also contains a rotamase signature between amino acids 56 and 71 (PROSITE database, PDOC00426), a feature that is absent from Drosophila Delta and Dll1 (Fig. 2). Although polyadenylation signals have not been identified in the 3′UTR, this sequence is derived from a full-length transcript as a poly(A) tail was present at the 3′ end of the cDNA. Northern blot analysis using a 3′UTR-derived probe of the Dll3, indicated that the gene produced a predominant transcript of 2.2 kb and a minor transcript of 0.8 kb. The 2.2 kb transcript is detected in the embryo (11.5 dpc), to a much lesser extent than in undifferentiated ES cells and is absent from the placenta (11.5 dpc) (Fig. 2).

An amino acid sequence comparison of Dll3 with Drosophila Delta and the mouse Dll1 (Fig. 3A) indicates that, overall, Dll3 shares 29% and 36% identity with Drosophila Delta and Dll1, respectively. Regions of greater identity exist between the EGF-like repeats, the greatest occurs between repeat 3 of Dll3 and repeat 5 of Drosophila Delta (47%), and between repeat 4 of Dll3 and repeat 6 of Dll1 (63%). The intracellular domains of these three proteins are highly dissimilar, the one from Dll3 being shorter. Fig. 3B illustrates a comparison of the DSL of Dll3 with the DSL of vertebrate and invertebrate Delta-like proteins. The DSL domain of Dll3 shows considerable divergence although some conservation of cysteine spacing is evident.

In Xenopus, primary neurogenesis gives rise to three longitudinal stripes of neurons either side of the dorsal midline. These neurons can be identified by the expression of neural-specific type II β-tubulin referred to as N-tubulin (Oschwald et al., 1991; Chitnis et al., 1995). Ectopic expression of a constitutively active form of Notch-1 prevents the production of these primary neurons as judged by the loss of N-tubulin-expressing cells in the neural plate (Chitnis et al., 1995). Likewise, ectopic expression of X-Delta-1 blocks neuron production indicating that it binds to and activates the Notch receptor (Chitnis et al., 1995). Given that Dll3 has such a divergent DSL and a reduced number of EGF-like repeats, it was possible that it might not be able to activate the Notch receptor. In order to test this, Dll3 and lacZ RNA were coinjected into a single Xenopus blastomere at the 2- to 4-cell stage and the embryos allowed to develop until the neural plate stage (stage 13-15). Whole-mount RNA in situ hybridisation revealed a reduction or complete loss of N-tubulin expression in the neural plate on the injected side of the embryo, showing that ectopic expression of Dll3 can also inhibit primary neuron production. Even though Dll3 is derived from the mouse, we observed inhibition of primary neurogenesis in 84% (16/19) of injected embryos, which is comparable with 79% (15/19) when X-Delta-1 RNA is injected (Fig. 4).

RNA in situ hybridisation either to intact embryos (5.5-13.5 dpc) or to cryosections (9.5-15.5 dpc) was used to determine the localisation of Dll3 transcripts. Dll3 transcripts are detected immediately prior to primitive streak formation (5.5-6 dpc) and are localised to the epiblast (data not shown). A low level of transcripts persist throughout the epiblast following the onset of gastrulation and up to mid-primitive streak stages (Fig. 5A,B). Later, at the full length primitive streak stage, epiblast expression is only detected in the cells that are adjacent to the primitive streak (Fig. 5B,C). Transcripts are also localised to the primitive streak itself and extend along its entire length but are absent from the node. Expression of Dll3 is also apparent in the nascent mesoderm as it emerges from the primitive streak (Fig. 5C). At first, this expression occurs throughout the entire wing of mesoderm but by the late primitive streak stage only the distal mesoderm which is fated to become paraxial mesoderm (Parameswaran and Tam, 1995) contains Dll3 transcripts (Fig. 5D). At all stages, expression in the epiblast appears considerably lower than that in the mesoderm.

Dll3 transcripts continue to be localised to the primitive streak throughout the latter stages of gastrulation and also persist in the tail bud (Fig. 6A,F). At early somite stages the highest level of Dll3 transcripts is seen in the paraxial mesoderm (Fig. 6A). Transverse sections through the posterior trunk confirm that the transcripts are localised to the paraxial mesoderm and are not present in axial (notochord), intermediate or lateral mesoderm (Fig. 6B). However, expression in paraxial mesoderm is only observed in presomitic mesoderm and nascent somites, and appears to cease as somites mature (Fig. 6C). The level of transcript accumulation is highest in the presomitic mesoderm, lower in the somite that is in the process of forming and in the most immature somites expression is detectable only at the anterior margin (Fig. 6C). This pattern of expression is maintained throughout somitogenesis indicating that Dll3 may be involved only in the initial formation of somites and not their subsequent differentiation.

Dll3 expression in the presumptive brain region is first detected at 8.25 dpc at a low level (Fig. 6A). High levels of transcript are seen only in a punctate pattern, and are first seen in the presumptive midbrain at 8.75 dpc (Fig. 6D), which then extends posteriorly into the hindbrain and spinal cord regions (Fig. 6E,F). This peppered expression pattern appears to coincide with the formation of neurons which, in the mouse, arise in the brain (8.5 dpc) before the spinal cord (9.5 dpc) (Mastick and Easter, 1996; Nornes and Carry, 1978). By the 25-somite stage, the anterior neuropore has closed and expression in the midbrain region has extended laterally. In addition, Dll3 transcripts are particularly prominent in the nasal pits, sensory ganglia (V, VII, IX, X), the sympathetic chain and the dorsal root ganglia (Fig. 6E,F).

Neural precursors proliferate within a pseudostratified columnar epithelium (ventricular zone) and their nuclei migrate between the basal and lumenal surfaces in a cell-cycle-dependent manner. S-phase nuclei lie basally while mitosis occurs at the lumenal surface. As the neural tube closes, cells at the basal surface start to lose their cytoplasmic connection with the lumenal surface and become postmitotic. These neuroblasts reside in the newly established intermediate layer (marginal zone) and will differentiate into neurons. Neurons migrate laterally into the mantle layer before elaborating axons. As neural differentiation progresses the mantle layer broadens initially in the ventral half of the neural tube (the basal plate) and then dorsally in the alar plate. Dll3 is expressed in the spinal cord along its dorsoventral extent (Fig. 7A) until 15.5 dpc when transcripts are no longer detected. This loss of expression is preempted by reduced expression at 14.5 dpc (data not shown). In order to clarify the proliferative status of the Dll3-expressing cells, RNA in situ hybridisation was performed in conjunction with BrdU labelling. This showed that Dll3-expressing cells are postmitotic since they do not incorporate BrdU (Fig. 7B). In general, the nuclei that had incorporated BrdU lie adjacent to the lumen, while those that express Dll3 lie lateral to this (Fig. 7C). Since the cells that express Dll3 are postmitotic, we determined whether they had started terminal differentiation. Islet-1 is a LIM homeobox gene expressed by motor neurons soon after they have left the cell cycle and by dorsal ipsilateral interneurons (Ericson et al., 1992). An antibody that detects islet-1 and islet-2 (also present in motor neurons; Tsuchida et al., 1994) was used in double labelling experiments and shows that virtually all cells that express Dll3 are devoid of the islet-1/2 protein (Fig. 7E), although coexpression was detected in occasional cells adjacent to the basal plate (Fig. 7F). Taken together these data show that Dll3 gene expression occurs after neuroepithelial cells have ceased proliferating but generally before terminal differentiation.

It has been demonstrated in the chick (Henrique et al., 1995; Myat et al., 1996) and we have confirmed in the mouse (data not shown) that C-Delta-1- and Dll1-expressing cells of the spinal cord are also postmitotic. Therefore, in the mouse, Dll1 and Dll3 may be expressed in the same cells. RNA in situ hybridisation shows that the distribution of Dll3 and Dll1 transcripts differs within the spinal cord (Fig. 8A,B), such that Dll1-expressing cells are localised to the ventricular zone while Dll3-expressing cells lie more laterally. Double RNA in situ hybridisation confirms this and shows very little overlap in gene expression (Fig. 8C). In three small patches, some overlap of expression is observed and this normally corresponds with the expression of Dll3 extending towards the lumen and ‘invading’ the Dll1 domain. While a more or less continuous rim of Dll3-expressing cells extends dorsoventrally, the expression of Dll1 is discontinuous and creates two gaps adjacent to the lumen that are devoid of Delta gene expression. These gaps in Dll1 expression coincide with the expression of Serrate-1, another Notch ligand (data not shown). In the hindbrain, similar domains of coincident and separate expression exist for Dll3 and Dll1 (Fig. 8D,E). Furthermore, whole-mount RNA in situ hybridisation demonstrates that Dll3 is only expressed in a subset of the regions of the hindbrain containing Dll1 transcripts. Thus at 8.5 dpc, only Dll1 appears to be expressed in patches in the lateral midbrain and in the forebrain while both Dll3 and Dll1 are co-localised in the dorsal midbrain (Fig. 8F). By 9.5 dpc, Dll3 expression is observed in the forebrain but unlike Dll1, in a very defined patch of cells in the ventral telencephalon (Fig. 8G). In addition at this stage, Dll3 transcripts appear to outline rhombomere 4 and mark the ventral aspect of rhombomere 2.

Both Dll3 and Dll1 transcripts are abundant in paraxial mesoderm but double RNA in situ hybridisation revealed differences in their patterns of expression. Within the presomitic mesoderm, the levels of Dll1 transcripts are not constant resulting in two regions where expression is relatively higher. In contrast, Dll3 transcripts were more uniformly distributed along the presomitic mesoderm but extended more rostrally than Dll1 and this rostral limit appeared to correspond to the anterior of the somite just about to form (Fig. 8H). During formation of a somite, a broad band of Dll3 expression was evident anteriorly while Dll1 was restricted to a posterior domain. In the most recently formed somite, these bands were refined to give a thinner stripe of Dll1 expression at the posterior and a comparable anterior stripe of Dll3 (Fig. 8I,J). The anterior stripe of Dll3 expression was only evident in the most recently formed somite whereas Dll1 expression in the posterior persisted in mature somites.

The mouse Delta homologue, Dll3, which we identified from a subtractive cDNA screen, is significantly divergent from Drosophila Delta and its other homologues (Fig. 3A). In the DSL region, the spacing of cysteine residues, which is conserved in all other Delta homologues, is only partly conserved in Dll3. Since the DSL is required for the binding of Delta to Notch (Fehon et al., 1990), this divergence suggests that Dll3 may preferentially activate a different Notch receptor to Dll1/Delta-1. The identification of a rotamase signature indicates that Dll3 has the potential to bend proteins (Fischer and Schmid, 1990), suggesting additional protein interactions.

Ectopic expression of X-Delta-1 or a constitutively active form of X-Notch-1 inhibits primary neuron formation in Xenopus (Chitnis et al., 1995). Despite the divergence of Dll3, it is also capable of inhibiting primary neurogenesis (Fig. 4). This strongly suggests that Dll3 can bind and activate Notch-1 or another Notch receptor and therefore is a true functional homologue of Drosophila Delta.

One function of the Delta/Notch signalling pathway is to control the release of individual cells from an initial population where cells are equivalent. A putative stem cell population resides within the primitive streak and its descendant, the tail bud, in the mouse (Tam and Beddington, 1987; Lawson et al., 1991; Wilson and Beddington, 1996; Wilson and Beddington, unpublished) which contributes to the supply of mesoderm required for axial elongation. Since Dll3, Dll1 and Notch-1 are all expressed in the primitive streak and tail bud (Figs 5C, 6A,F) (Bettenhausen et al., 1995; Franco del Amo et al., 1992; Reaume et al., 1992), they may influence the balance between the stem cell pool and the rate at which progeny for differentiation are supplied. One might predict that loss-of-function mutations in members of the Delta/Notch signalling pathway would cause a decrease in the size of the stem cell population and hence axis truncations. However, such a phenotype has not been observed for either Notch-1, Dll1 or RBP-Jκ(a component of the Notch signal transduction pathway) null mutant embryos (Swiatek et al., 1994; Conlon et al., 1995; Hrabe de Angelis et al., 1977; Oka et al., 1995), but this failure to disrupt gastrulation could be due to compensation by other family members. Although speculative, the uniform expression, albeit apparently low, of Dll3 in pre- and early gastrulation epiblast may help to maintain pluripotency of this tissue.

Segmentation of paraxial mesoderm proceeds in a rostrocaudal sequence with the newly generated epithelial somite forming at the anterior of the presomitic (unsegmented) mesoderm. A periodicity is apparent in the presomitic mesoderm in the form of somitomeres; whirls of mesenchyme thought to presage the formation of epithelial somites (Meier, 1979; Tam et al., 1982). The distribution of Dll3 and Dll1 transcripts in the presomitic mesoderm clearly indicates that this tissue has explicit anterior-posterior polarity (Fig. 8H). Rotation of the presomitic mesoderm through 180° in the chick results in the original sequence of somite formation being maintained so that they now form in a caudal-to-rostral sequence (Menkes and Sandor, 1977), arguing that this expression profile reflects a stable ‘prepattern’ within the presomitic mesoderm. However, if the cells are disaggregated prior to grafting into a host, then a normal rostrocaudal sequence results (Menkes and Sandor, 1969; Stern and Keynes, 1986). It is possible that the Delta genes play a role here by maintaining cell-cell interactions required for stable anterior-posterior polarity.

Ultimately the production of a somite requires the formation of a boundary and the final transition from mesenchyme to epithelium necessitates an increase in cellular adhesion (see Keynes and Stern, 1988). Dll3 and Dll1 (Bettenenhausen et al., 1995) are expressed in the presomitic mesoderm and during the birth of nascent somites. Despite this overlap of expression in the paraxial mesoderm, Dll1 null mutant embryos display a somite phenotype (Hrabe de Angelis et al., 1997). This suggests that Dll3 cannot compensate for the absence of Dll1 and that Dll1 and Dll3 do have different functions during somitogenesis. Mutant analysis shows that Dll1 is required for the formation of epithelial somites and in establishing their anterior-posterior polarity. Notch signalling is also implicated in somitogenesis since null mutations for both Notch-1 and RBP-Jκ exhibit a somite phenotype (Swiatek et al., 1994; Conlon et al., 1995; Oka et al., 1995). In contrast to Dll1 mutants, Notch-1 and RBP-Jκmutants form epithelial somites, but these are irregular in size and shape. Despite this irregularity, there is evidence that anterior-posterior and dorsalventral pattern is established.

Dll3 and Dll1 are expressed in nascent somites as well as in the somite that is in the process of forming. In the nascent somite, Dll3 is expressed at the anterior boundary while Dll1 is expressed at the posterior boundary. Given the expression of Dll3 and Dll1 on either side of the nascent somite boundary, it is possible that one of their roles may be to reinforce segmentation by homotypic protein interaction. In Drosophila, Delta can bind to itself as well as to Notch (Fehon et al., 1990) and therefore one could envisage an adhesive function whereby the two mutually exclusive Delta molecules serve to maintain nascent somite integrity. This also assumes that Dll1 and Dll3 do not bind to one another, which is supported by the fact that they are considerably divergent. That like and unlike somite halves exist is demonstrated by grafting experiments in the chick, which show that when the anterior (A) of one somite and the posterior (P) of another somite adjoin the integrity of the somite is maintained. But when two like halves (A and A or P and P) abut then cell mixing occurs and the boundary is lost (Stern and Keynes, 1987). In addition, it is also possible that Dll3 and Dll1 are involved in the formation of the somite boundary itself. As the somite forms, a broad band of Dll3 expression is evident in the anterior half. Thus the anterior boundary of the somite in the process of forming (which expresses Dll3) abuts the posterior aspect of the somite that has just formed (which expresses Dll1). Perhaps formation of the somite boundary is analogous to the establishment of the dorsoventral boundary during Drosophila wing development (see Brook et al., 1996). Notch is required to set up the dorsoventral wing boundary but activation of Notch requires two different ligands expressed in mutually exclusive compartments. In the ventral compartment, Notch is activated by Serrate expressed on the surface of dorsal cells. In the dorsal compartment, Notch is activated by Delta which is expressed on the surface of ventral cells. Thus, at the prospective somite boundary, Dll3 may activate Notch on one side and Dll1 may activate Notch on the other. Serrate-1 may also play a role since, in the mouse, it is expressed in a thin stripe of cells corresponding to the site of somite boundary formation (Mitsiadis et al., 1997). However, it remains to be resolved whether the activation of Notch by different ligands (Dll3 and Dll1) produces a differential signal or whether Notch is activated by different ligands simply because the ligands are differentially expressed due to prior patterning constraints. It may be relevant that during Drosphila wing development, Notch-expressing cells in a given compartment have different responses to Delta and Serrate (Doherty et al., 1996). Therefore, during wing development in Drosophila, appropriate signalling via Notch does appear to be dependent on activation by two different ligands. A similar sensitivity to different ligands may exist in mouse somitogenesis although it is also possible that different Notch receptors may be activated on either side of the somite boundary.

Dll3 and Dll1 are extensively expressed during neurogenesis in both overlapping and distinct domains. For example, in the midbrain overlapping gene expression occurs dorsally with exclusive Dll1 expression extending more laterally (Fig. 8F). Likewise, Dll1 is expressed throughout much of the forebrain while Dll3 is restricted to a small ventral patch (Fig. 8G). Over-lapping expression is evident in the dorsal root ganglia and much of the CNS extending from the midbrain to the caudal spinal cord. In the CNS, the cells that express Dll1 and Dll3 are postmitotic cells (Fig. 7B,C,D). These genes are however not coexpressed, as double RNA in situ hybridisation reveals almost mutually exclusive domains of gene expression (Fig. 8C-E). The Dll1-expressing cells lie within the ventricular zone while cells expressing Dll3 reside more laterally but are negative for islet-1/2 protein, a very early marker of motor neuron differentiation. Therefore the Dll3-expressing cells appear to represent a population on the verge of differentiating.

The different domains of Dll1 and Dll3 expression could represent two distinct postmitotic populations destined for different neuronal fates. However, while it is clear that different neuronal cell types emerge from different dorsoventral levels of the CNS, there is no evidence for distinct mediolateral neuronal populations (see Tanabe and Jessell, 1996). Therefore, it is more likely that the different mediolateral expression profiles of Dll1 and Dll3 represent sequential gene expression and reveal an additional layer of complexity in the sequence of events leading to the birth of a neuron. That the occasional cell can coexpress Dll3 and islet-1/2 (Fig. 7E,F) supports this hypothesis since coexpression of Dll1 and islet-1/2 was not observed (data not shown).

In Drosophila, distinct basic helix-loop-helix (bHLH) proteins contribute to the specification of different neurons (Jarman et al., 1993; Skeath and Doe, 1996). In vertebrates, non-overlapping dorsoventral domains of neurogenin, Mash-1 and Math-1 (Ma et al., 1996; Lo et al., 1991; Akazawa et al., 1995) gene expression exist within the ventricular zone of the spinal cord (see Tanabe and Jessell, 1996). These genes probably act at the head of the neurogenic patterning program and lie upstream of other bHLH genes such as NeuroD (Lee et al., 1995; Ma et al., 1996). In a comparable fashion, the progression from Dll1 to Dll3 may represent increasing commitment to terminal differentiation. If neuronal diversification involves the sequential activation of bHLH proteins, a control mechanism is required to release cells from one bHLH-expressing state to the next. This is where lateral inhibition via Delta/Notch signalling may again have a role. The utilisation of more than one Delta ligand in a stereotyped spatiotemporal sequence, as cells leave the progenitor pool and approach differentiation, may allow for finer control of neuronal supply. Therefore the progression from Dll1 to Dll3 expression may reflect a stepwise increase in commitment to terminal differentiation of distinct neuronal types during the relatively prolonged process of vertebrate neurogensis.

We would like to thank Dr Paul Thomas for his help in screening cDNA library clones, Drs Tim Mohun and Branko Latinki ć for their assistance with the Xenopus experiment, Ajay Angris for cryosections and Dr Achim Gossler for sharing unpublished results. S. L. D is supported by and R. S. P. B. is an International Scholar of the Howard Hughes Medical Institute, D. H. holds a European Science Foundation long term fellowship.