Induction of identified mesodermal cells by CNS midline progenitors in Drosophila

The Drosophila embryo is initially subdivided by dorsal/ventral patterning genes into several discrete domains (e.g. reviewed in Anderson, 1987; Roth et al., 1989; Govind and Steward, 1991; St Johnston and Nüsslein-Volhard, 1992). The ventral domain of the blastoderm stage embryo represents the mesodermal anlagen. The ventrolateral domain on either side forms the neuroectoderm, which is separated from the mesoderm by a single row of cells, called mesectoderm (Poulson, 1950; Thomas et al., 1988; Crews et al., 1988). The early specification of both the mesectodermal and the mesodermal domains requires the transcriptional regulators TWIST and SNAIL (Rushlow and Arora, 1990; Kosman et al., 1991). Upon gastrulation, twist expression is confined to the mesoderm. It is required for the expression of other mesodermal genes (Thisse et al., 1987; Bodmer et al., 1990; Leptin, 1991) and probably acts as a key factor in initiating subsequent mesodermal differentiation (for reviews see Bate, 1993; Baylies and Bate, 1996).

First patterning of the mesoderm, establishing the primordia of the somatic and visceral muscles, the heart and the fat body, is mediated by genes like tinman, bagpipe and serpent. Their restricted expression in the mesoderm appears to involve the function of segmentation genes (Azpiazu et al., 1996). During early stages at the onset of gastrulation, however, mesodermal cells invaginating through the ventral furrow do not yet seem to be committed to form particular mesodermal derivatives (Beer et al., 1987). Thus, the specification of the various mesodermal cell fates could result from an autonomous self-organizing capacity of the mesoderm itself and/or it could be nonautonomously provided by other tissues during early stages.

The latter is supported by transplantation experiments, which suggest that the positions occupied by the mesodermal cells following invagination and subsequent dorsal movement across the inner surface of the ectoderm are critical for their determination (Beer et al., 1987). Position-dependent differentiation of mesodermal cells could be evoked by inductive signals from the ectoderm. Indeed, recent data have provided evidence for ectodermal signals controling various aspects of mesodermal patterning (Staehling-Hampton et al., 1994; Baylies et al., 1995; Frasch, 1995; Baker and Schubiger, 1995; Bate and Baylies, 1996; Ranganayakulu et al., 1996).

An alternative source for inductive signals patterning the mesoderm could be the mesectodermal cells, which give rise to the CNS midline cells (Klämbt et al., 1991; Bossing and Technau, 1994). During gastrulation the two rows of mesectodermal cells become juxtaposed along the ventral midline while still being in direct contact with cells of the mesodermal anlage. The gene single minded (sim) acts as a master regulatory gene of CNS midline development, and in sim mutant embryos midline cells fail to differentiate (Thomas et al., 1988; Crews et al., 1988; Nambu et al., 1991). In addition sim is required for proper patterning of the ventral ectoderm. Since sim is not expressed in this tissue the establishment of ventral ectodermal cell fates appears to require an inductive signal originating from the CNS midline cells (Kim and Crews, 1993). Based on its characteristic ventral cuticle phenotype, sim has been placed together with sichel, rhomboid, Star, pointed and spitz in the so-called spitz group (Mayer and Nüsslein-Volhard, 1988). Mutations in all spitz group genes also lead to defects in CNS midline development (Klämbt et al., 1991; Sonnenfeld and Jacobs, 1994). Ventral ectodermal cell fates are specified by differential activation of the Drosophila EGF receptor homologue (DER) through specific ligands (Raz and Shilo, 1993; Schweitzer et al., 1995a). spitz encodes such a DER-activating ligand, and recent experiments suggest that the midline cells can provide SPITZ protein required for the correct patterning of the ventral ectoderm (Golembo et al., 1996b).

To further investigate the question of whether the ventral CNS midline cells also induce aspects of mesodermal patterning, we have employed a combination of experimental and genetic tools. We have focused our analyses on the formation of a special set of mesodermal cells, the dorsal median (DM) cells. The DM cells (described as transverse nerve exit glia by Gorczyca et al., 1994) are arranged as a pair in each of the three thoracic and in the first seven abdominal segments medially on the dorsal surface of the ventral nerve cord and close to the segmental boundaries (Fig. 3A). They send cell processes laterally to muscle attachment sites, which pioneer and enwrap a set of motoneuronal axons exiting from the dorsal midline of the CNS to form the transverse nerve on either side (Osborne, 1964; Beer et al., 1987; Chiang et al., 1994; Gorczyca et al., 1994; Dunin Borkowski et al., 1995). In the thoracic segments, during larval development, the DM cells lose their lateral extensions to form one dorsal neurohemal organ per thoracic segment (Gorczyca et al., 1994; R. Cantera, personal communication), which expresses the hormone allatotropin (Zitman et al., 1993) and becomes innervated by fibers containing FMRFamide-like peptide (White et al., 1986; Schneider et al., 1991).

We show that in the early embryos a single mesodermal progenitor cell adjacent to the CNS midline progenitors gives rise to a segmental pair of DM cells. The formation of these mesodermal cells is controled by inductive signals from the CNS midline cells. Overexpression of spitz in all midline cells induces the formation of supernumerary DM cells by triggering additional mitoses of the DM progenitor cells. Transplantation of midline cells, which express high levels of SPITZ into the midline of wild-type embryos, results in an even more prominent induction of extra DM cells. Our experiments suggest that patterning of the mesoderm requires the presence of gradients, and the steepness of the gradient eventually regulates the formation of the correct number of DM cells.

A reporter strain harbouring the P[btn 1.85-lacZ] constuct was used to trace the fate of the DM cells (Chiang et al., 1994; kindly provided by P. Beachy). The following mutant lines have been used: sim^H9 (Hilliker et al., 1980; kindly provided by S. Crews); btn^r3 (Chiang et al., 1994; kindly provided by P. Beachy); flb^IK35 (Schejter and Shilo, 1989); spitz^IIA14 (Nüsslein-Volhard et al., 1984), Star¹²⁶ (Kolodkin et al., 1994); pnt^rM254 ; pnt^Δ114 (Klämbt, 1993); rho^del1 (Sturtevant et al., 1996; kindly provided by E. Bier); Df(1)KA14 uncovering otd (Finkelstein and Perrimon, 1990); cycA³ (Lehner and O’Farrell,1989; kindly provided by C. Lehner). Homozygous mutant embryos were identified by the absence of ‘blue balancer’ chromosomes. For ectopic expression of sspi we employed the GAL4 system (Brand and Perrimon, 1993). The following GAL4 inducer and UAS lines were used: sim-GAL4 (Scholz et al. 1997); rho-GAL4 (kindly provided by M. Levine); UAS-sspi4a (Schweitzer et al., 1995a; kindly provided by B. Shilo).

Embryos were mounted ventral side down on coverslips, dessicated, covered with fluorocarbon oil and placed on an inverted microscope equipped with a micromanipulator (Leica). DiI labelling as well as processing and photoconversion of flat preparations were performed as previously described (Bossing and Technau, 1994; Bossing et al., 1996). Individual mesodermal primordial cells were labelled with DiI during gastrulation, shortly before the mesectodermal cells meet at the ventral midline (stage 6). Mesodermal cells that were among the latest to invaginate into the ventral furrow were labelled. Subsequently, upon completion of gastrulation (stage 7/8), their position relative to the midline progenitors (which are clearly identifiable at this stage, see below) was determined under the fluorescent microscope. Only those labelled cells that were clearly located adjacent to the midline cells (see Fig. 1D) were further analyzed.

Embryos were mounted ventral side down, as described for DiI labelling. A capillary (tip diameter 6–8 μm) was moved along the surface of the coverslip towards and in line with the ventral furrow of an embryo. The capillary was introduced through the vitelline membrane midventrally from the posterior side at the place where the vitelline membrane is attached to the coverslip. 15–25 minutes after the onset of gastrulation (stage 7/early stage 8; stages according to Campos-Ortega and Hartenstein, 1985) the two rows of mesectodermal midline cells are juxtaposed at the ventral midline, and their cell diameters are significantly increased, as they will divide shortly afterwards (see Bossing and Technau, 1994). Thus cell ablations were performed at this stage, when the midline progenitor cells were clearly distinguishable in vivo from neighbouring neuroectodermal or mesodermal cells by their position, size and shape (magnification used: 50×10). The capillary was carefully moved along the ventral midline and all midline progenitors within 1-2 segmental anlagen were selectively removed, leaving their neuroectodermal and mesodermal neighbours in place. Upon the removal of midline cells it was verified that the neighbouring cells remained intact. At late embryonic stages (stage 16) specimens were stained for various markers and processed as described below.

Donor and host embryos were mounted ventral side down. Donors were labelled with HRP, as previously described (Prokop and Technau, 1993). Midline cells were selectively removed with a capillary (tip diameter 12 μm) from donors at stage 7, as described above. Depending on the type of experiment one or several cells were transplanted into the midline of hosts at the same stage. In one experiment hosts were at a later stage, as indicated in the text. For implantation the capillary was oriented perpendicular to the midline of host embryos to minimize spatial distortions. Precise position and number of implanted cells were determined under direct optic control. At late embryonic stages (stage 16) hosts were stained and processed as described below.

In addition to the P[btn 1.85 lacZ] transformant strain (Chiang et al., 1994; kindly provided by P. Beachy), antibodies against glutactin (Olson et al., 1990) or collagen IV (Murray et al., 1995; both kindly provided by L. Fessler and J. Fessler) were used for tracing DM cells, and BP102 antibody (kindly provided by C. S. Goodman) for staining CNS fiber tracts. Anti-β-galactosidase antibodies (Dianova) or X-Gal were utilized for staining embryos of the P[btn 1.85 lacZ] strain or to detect blue balancer chromosomes. Antibody, HRP and X-Gal single or double stainings were performed on flat preparations as previously described (Schmidt-Ott and Technau, 1992: Bossing et al., 1996).

The pattern of the somatic muscles and other mesodermal derivatives has been described in detail (for a review, see Bate, 1993). However, except for an assignment of mesodermal cells to a metamerically repeated anteriorposterior subdivision (Dunin Borkowski et al., 1995), a fate map of the mesodermal anlage is still lacking. Assuming that possible inductive signals from the mesectoderm would most significantly affect mesodermal cells close to the midline, we attempted to trace the fate of mesodermal cells that are in direct contact with the midline cells.

Single mesodermal primordial cells were labelled with DiI during gastrulation, just before they move into the ventral furrow. Labellings were performed within the region of segment anlagen T2 to A3. Upon completion of ventral furrow formation, when the mesectodermal cells can be clearly identified at the ventral midline, the position of the labelled cell with respect to the midline cells was checked (Fig. 1D). The progeny of mesodermal cells that were located adjacent to the midline were further traced into late embryonic stages (stage 16) and subjected to photoconversion to analyze the structures formed by these cell clones in fixed preparations (n=29). These ‘first row’ mesodermal cells gave rise to three types of structures (Fig. 1A-C): ventral somatic muscles (n=16), cells associated with the salivary glands (presumably fat cells, n=8) and the dorsal median (DM) cells (n=5). Most of the muscle clones obtained from the ‘first row’ mesodermal progenitors belonged to the ventral longitudinal group and ventral oblique group (nomenclature according to Bate, 1993). Clonal progeny cells participated in the formation of up to five muscles in different combinations.

DiI labellings indicate that each segmental pair of DM cells derives from one mesodermal progenitor cell. This agrees with previous data obtained from lineage analysis on mesodermal cells using HRP as a lineage tracer (Beer et al., 1987). Furthermore, the clonal analysis reveals that, in addition to the DM cells, the progenitor cell gives rise to further progeny, which participate in the formation of ventral somatic muscles.

In one case, in which we simultaneously labelled a midline progenitor and an adjacent mesodermal progenitor cell, we obtained a clone of ventral unpaired median (VUM) neurons (an interneuron and a motoneuron; see also Bossing and Technau, 1994) in the CNS midline, and a mesodermal clone consisting of DM cells and a group of ventral somatic muscles (Fig. 1C). This shows that the DM cell progenitor is adjacent to the midline and suggests that it maps near the VUM progenitors (for the arrangement of midline progenitors, see Klämbt et al., 1991; Bossing and Technau, 1994).

In embryos that lack the function of cyclin A mitosis is arrested during the 16th cell cycle (Lehner and O’Farell, 1989). Nevertheless, cells in this mutant are able to differentiate and DM cells carrying bilateral cytoplasmic extensions are clearly identifiable in late embryos (Fig. 5C). In cyclin A mutants there is only one DM cell per segment (Figs 4, 5C) suggesting that in the wild type the DM cells appear in the lineage after the second postblastodermal mitosis. To reveal the spatiotemporal pattern of DM cell development in some more detail we used a reporter strain harbouring the P[btn 1.85 lacZ] construct, which confers β-galactosidase expression specifically in the DM cells (Fig. 2; Chiang et al., 1994). Expression of the marker is first detectable at early stage 11 in a cluster of 2-4 cells at the midline on top of the CNS (Fig. 2A; see also Chiang et al., 1994). One of these cells expresses the marker significantly more strongly than the others. A short time later (stage 11) this cell appears to divide (Fig. 2B), and the strongly labelled progeny subsequently differentiate into the DM cells (Fig. 2C,D). During their entire development these cells stay close to the CNS midline. In contrast, the weakly labelled mesodermal cells migrate laterally during stage 13 (Fig. 2C) and subsequently participate in the formation of ventral somatic muscles (Fig. 2D). These cells probably correpond to those sharing the same lineage with the DM cells (see above).

The single minded gene encodes a basic HLH transcription factor and acts as a master regulator of CNS midline development (Nambu et al., 1991). In sim mutant embryos the midline progenitor cells fail to divide and do not migrate in the CNS layer where they normally differentiate into neurons and glia (Nambu et al., 1991). Furthermore, tracing their fate with cellspecific markers suggested that the midline cells do not survive in single minded mutant embryos (Sonnenfeld and Jacobs, 1994).

To see whether the DM cells are affected in sim mutants we used the P[btn 1.85 lacZ] reporter element. In sim^H9 mutant embryos (sim^H9 represents a null sim allele, Nambu et al., 1991), DM cells are generally missing (except in 8 out of 130 segments; Fig. 3A,B). This was confirmed by staining mutant single minded embryos with antibodies against glutactin and collagen IV (Olson et al., 1990; Murray et al., 1995) as independent markers for the DM cells. Thus, expression of sim is required for the formation of the DM cells.

sim expression is not strictly confined to the midline cells but is also found in ventral mesodermal progenitor cells (Lewis and Crews, 1994). Therefore, failure of DM cell formation in sim mutants could be a non-autonomous as well as an autonomous effect. To distinguish between these possibilities we selectively removed midline progenitors from embryos carrying the P[btn 1.85 lacZ] transgene (n=22). Midline cells were removed under direct optic control with a capillary shortly after gastrulation, when they are clearly identifiable and distinguishable from neighbouring cells in the living embryo by their position, shape and size (see Materials and Methods). Up to 18 consecutive midline cells were removed, corresponding to the entire population of 2–3 segmental anlagen (number of midline progenitors per segment: 6-9; Bossing and Technau, 1994). In addition to producing a local phenocopy of the CNS defects described for sim mutants (absence of commissural axon tracts, fusion of the longitudinal axon tracts and reduced nerve cord condensation; Thomas et al., 1988), we find a clear effect on the formation of the DM cells (Fig. 3C). This effect depends on the number of consecutive midline cells removed from the embryos. Removal of fewer than ten cells had no effect on the formation of DM cells and resulted in only a weak local phenocopy of the sim mutant CNS phenotype (n=15; Fig. 3D). Upon removal of 12-18 midline progenitor cells one pair of DM cells was lacking (n=6), and in one case two pairs of DM cells in consecutive segments were lacking. These data indicate that the midline cells are required for the formation of the DM cells.

To test whether the midline cells are sufficient for the induction of the mesodermal DM cells we performed a series of transplantation experiments. In the first experiment, 3–6 midline progenitors were transplanted from wild-type donors into the midline region of sim mutant hosts carrying the P[btn 1.85 lacZ] construct (stage 7). Following further development (stage 16) the hosts were stained for β-galactosidase expression. In all cases the implanted wild-type cells led to a local rescue of the sim phenotype in that they induced the formation of 2–6 DM cells (n=11). These are located dorsally on the ventral cord, although their paired arrangement and lateral processes are disturbed (Fig. 3F). Furthermore, the implanted cells also had rescuing effects on CNS fiber tract formation in otherwise mutant single minded embryos (data not shown).

In a second experiment we implanted about ten wild-type midline cells in the CNS midline of wild-type hosts carrying the P[btn 1.85 lacZ] construct. These additional midline cells led to the formation of supernumerary DM cells (n=13; Fig. 3E). Since transplantation of fewer than five cells was insufficient to induce the recruitment of additional DM cells (n=30), we conclude that a certain threshold in the level of a midlinederived inductor is required.

The CNS midline cells induce the formation of the DM cells. Do all midline cells share similar inductive capabilities? To address this question we took advantage of mutations only affecting specific cell types within the developing CNS midline. In contrast to sim, mutations in other spitz group genes most prominently affect the midline glial cell lineage. This is most clearly seen in mutant pointed embryos. pointed encodes an ETS transcription factor, which in the CNS is expressed only in glial cells (Klämbt, 1993). In pointed mutant embryos, these glial cells fail to differentiate properly (Klaes et al., 1994).

DM cell formation is affected by all spitz group mutations (Fig. 4). In mutant pointed embryos, reduced numbers of DM cells (2–12; compared to 20 DM cells in wild-type embryos) form. In spi mutant embryos either no DM cells (in about 30% of the cases) or reduced numbers of 4–12 DM cells per embryo are formed, as revealed with an anti-glutactin or anti-collagen IV antibody. In Star mutant embryos DM cell numbers varied between 2 and 6 per embryo, and in rhomboid mutant embryos between 6 and 10. These data suggest that the midline glial cells perform an important function during the induction of the DM cells.

To assay whether the CNS midline neurons contribute to the inductive signals governing the formation of the DM cells, we analysed mutant orthodenticle (otd) embryos. otd encodes a homeodomain protein that is expressed in midline neurones. In otd mutant embryos most of the VUM neurons and probably also the median neuroblast (MNB) degenerate (Finkelstein et al., 1990). In addition, the midline glial cells fail to migrate posteriorly, but this may be an indirect effect due to the absence of the VUM neurons (Klämbt et al., 1991). In all otd embryos inspected, DM cell numbers corresponded to wild type (n=18; Fig. 4). Thus, midline neurones appear to be dispensable with regard to DM cell formation. However, in otd mutants not all neuronal midline lineages degenerate. It is thus possible that the midline neurons contribute to the inductive signal required for DM cell formation at earlier developmental stages.

Cell-fate specification in the ventral ectoderm requires the graded activation of DER (flb) by secreted SPITZ. The RHOMBOID and STAR proteins may be required for the production or processing of the SPITZ precursor to generate the active, secreted, form (Schweitzer et al., 1995a). It has been recently shown that the midline can be the source of active SPITZ protein required for patterning in the entire ventral ectoderm (Golembo et al., 1996b). The phenotypes described above for spi, S, rho and pnt mutant embryos as well as for DER/flb mutants (see Fig. 4) are compatible with the hypothesis that the DER signalling pathway may also be involved in the induction of ventral mesodermal cell fates, like the DM cells.

We thus asked whether overexpression of secreted SPITZ in the midline would result in a recruitment of additional DM cells. Towards this goal we have employed the GAL4 system (Brand and Perrimon, 1993) in combination with cell transplantations. In embryos expressing secreted SPITZ in all cells of the ventral midline (sim-GAL4/UAS-spitz), the number of DM cells is increased to about the same extent as upon implantation of additional wild-type midline cells in the wild type (see above; Fig. 5A).

The increase in DM cell numbers was much more dramatic when we isotopically transplanted midline cells from simGAL4/UAS-spitz donors (stage 7) into the midline of wild-type hosts (same stage; n=17). The implantation of only one or two midline cells from these donors is sufficient for the induction of a large number of additional DM cells (Fig. 5B,D).

The supernumerary DM cells found upon transplantation of individual SPITZ-expressing midline cells could result either from conferring the DM fate onto a larger number of progenitor cells within the mesodermal primordium and/or from inducing individual DM cell progenitors to perform additional mitoses. To distinguish between these possibilities we transplanted midline cells overexpressing spitz into the midline of embryos that lack the function of cyclin A. In such mutant hosts the spitz-overexpressing midline cells were not able to induce additional DM cells (n=13; Fig. 5C). This suggests that in the wild-type background overexpression of spitz in the midline cells or implantation of supernumerary midline cells leads to an increase in DM cell numbers by triggering additional mitoses and not by recruiting additional DM cell progenitors.

No additional DM cells are induced when spitz-overexpressing midline progenitors from stage 7 donors (about 20 minutes after onset of gastrulation) are heterochronously transplanted into wild-type hosts in which the midline cells have already divided (about 50 minutes after onset of gastrulation; stage 8). At this stage rearrangement of mesodermal cells has not yet taken place and they still contact the midline (Hartenstein, 1993). Therefore, the ability of mesodermal progenitors to respond to the SPITZ signal appears to have already ceased after the early gastrula stage.

In the above experiments we noted a difference in the number of DM cells formed upon overexpression of secreted SPITZ in all midline cells versus overexpression of secreted SPITZ in only a single midline cell. The generation of a local source of secreted SPITZ protein within the midline by cell transplantation appeared to be most efficient in inducing DM cell formation. To address the question of whether general expression of secreted SPITZ in the midline as well as in the mesoderm might level out inductive gradients and subsequently block the formation of extra DM cells, we used a modified rho-promotor-GAL4 line (kindly provided by M. Levine; Klaes et al., 1994). The expression pattern conferred by this GAL4 line mimics the rhomboid expression pattern, i.e. expression in the CNS midline and the ventralmost neuroectodermal region (Bier et al., 1990). However, since SNAIL binding sites have been destroyed in the rhomboid promotor fragment used, GAL4 expression is also found in the mesodermal domain (data not shown).

In contrast to the DM cell phenotype of sim-GAL4/UAS-spitz embryos, we do not find an increase in DM cell numbers in embryos in which expression of secreted SPITZ is driven by rho-GAL4. This result is in line with the above-mentioned model of DM cell formation. To control for secreted SPITZ expression in rho-GAL4/UAS-spitz embryos we transplanted one or two midline cells from rho-GAL4/UAS-spitz donors into the midline of wild type (n=18). In these experiments we observed the same DM cell phenotypes as upon transplantation of midline cells taken from sim-GAL4/UAS-spitz donors (see above). Thus we conclude that DM cell formation indeed does not depend on the absolute amount of the SPITZ signal, but rather on local differences in its concentration.

In summary we have presented evidence that the midline cells activate DM cell formation via the generation of local inductive gradient(s). Beside providing instructive signals to the mesoderm, the midline cells appear to be capable of influencing cell divisions in the mesoderm.

We have studied mechanisms underlying the formation of a pair of identified mesodermal cells, the so-called DM cells. Genetic and experimental data provide evidence that their formation requires inductive signals from the CNS midline cells. A combination of cell transplantation experiments and missexpression studies suggest that the DER signalling pathway is required at two steps of DM cell formation: firstly it is required to recruit DM progenitor cells from the pool of mesodermal cells, and secondly DER signalling appears to control the subsequent cell divisions of the DM cell progenitor.

In a series of DiI single-cell-labelling experiments we traced the fate of the mesodermal progenitors that are located adjacent to the midline. These lateral mesodermal cells were found to give rise to three cell types: ventral somatic muscles, a group of cells located close to the salivary glands (presumably fat cells) and the dorsal median cells (DM cells).

The mesodermal origin of the DM cells has been previously documented (Beer et al., 1987) and they do not express known glial or neuronal markers (e.g. repo, AA142, pointed, elav; data not shown). Beside their characteristic morphological properties, the DM cells were shown to be molecularly distinct mesodermal cells, as they express and require the homeobox gene buttonless (btn) (Chiang et al., 1994) in addition to a number of other markers (neuroglian, Bieber et al., 1989; laminin, Montell and Goodman, 1989; glutactin and collagen IV, Olson et al., 1990; Murray et al., 1995).

The DM cells develop from the lateral mesoderm, which is in direct contact with a distinct set of neural progenitors, the later CNS midline cells. By transplantation and cell ablation experiments we showed that the CNS midline provides an inductive signal leading to DM cell formation and that this signal acts on mesodermal progenitors early in the gastrula (stage 7). Genetic data, as well as ectopic expression experiments, suggest that the inductive signal is conveyed via the DER signalling pathway. This is also corroborated by the fact that midline cells overexpressing secreted SPITZ fail to induce DM cells when isotopically transplanted into DER/faint little ball (flb) mutant hosts (n=9; data not shown). However, in spitz, rhomboid and Star mutants in which the production or processing of SPITZ is impaired (Schweitzer et al., 1995a), the effect on DM cell formation is significantly lower compared to single minded. Thus, the midline may contribute other signals beside SPITZ involved in DM cell formation. A candidate for such a factor could be VEIN, which has been shown to be expressed in overlapping domains with SPITZ and to act synergistically with SPITZ to achieve the required level of DER activation for normal development of ventrolateral ectodermal cells (Schnepp et al., 1996). On the other hand, since DER/flb mutant embryos also show a weaker DM cell phenotype compared to sim (Fig. 4), wecannot exclude that additional signalling pathways contribute to DM cell development.

In invertebrate as well as in vertebrate development, cells located at the midline of the central nervous system often act as organizing centres. The cells located at the ventral midline of the neural tube are a set of morphological distinct cells called floor plate cells (Kingsbury, 1930). They are involved in dorsoventral patterning of the neural tube by contact-dependent mechanisms as well as diffusible molecules (Yamada et al., 1993; Hynes et al., 1995) and participate in controlling the patterning of somitic derivatives (e.g. Pourquie et al., 1996; Münsterberg and Lassar, 1995).

Similarly, the Drosophila CNS midline cells exert a number of inductive functions. The midline cells have been implicated in the patterning of the ventral epidermis (Kim and Crews, 1993) and recent experiments suggest that the CNS midline cells are able to express secreted SPITZ protein, which is required for ventral ectoderm pattern formation (Golembo et al., 1996b). Here we have shown that an inductive signal, originating from the CNS midline, governs the development of the mesodermal DM cells.

In addition, the CNS midline is also required for establishing the normal pattern of the ventral somatic muscles. In sim mutant embryos the pattern of the ventral somatic muscles is disturbed, in that many of them cross above the ventral cord and attach to the body wall on the opposite side of the embryo (Fig. 3B; see also Lewis and Crews, 1994). Removal of midline progenitors from the wild type leads to a phenocopy of this pattern, and implantation of wild-type midline cells into the midline of sim mutant embryos leads to a partial rescue of the phenotype (Fig. 3C,F), which indicates that the muscle pattern defects are non-autonomous and due to the absence of midline cells.

Thus in all systems analysed, the CNS midline cells appear to act as an important bidirectional source of inductive signals.

Inductive processes have been analysed thoroughly in many other systems, such as the induction of photoreceptor cell development in the Drosophila eye, where all ommatidial cells are recruited by a reiterative activation of the DER signalling pathway (Freeman, 1996). In cells competent to respond to the recruitment signal (SPITZ), DER signalling is activated and the cell adopts its appropriate fate. A fine tuning of DER activation is achieved by the interplay of two antagonising proteins, SPITZ and ARGOS. Although ARGOS antagonises SPITZ functions in many systems (Schweitzer et al., 1995b; Freeman, 1996; Golembo et al., 1996a; Okabe et al., 1996), we have found no evidence for a similar function of ARGOS in DM cell formation.

In other developmental contexts EGF-receptor signalling controls cell division (for a review, see Marshall, 1995). This is also the case in the developing eye, where DER signalling is required for apparently all cell divisions before its function is reused during ommatidial recruitment (Xu and Rubin, 1993).

Similar to eye development, secreted SPITZ appears to initiate DM cell formation by the recruitment of an already competent cell and to control its subsequent proliferation.

In the mesoderm the presumptive DM cell progenitors appear to be routed towards this particular developmental fate by the action of twist, segment polarity genes and tinman (Azipiazu and Frasch, 1993; Chiang et al., 1994; Gorczyca et al., 1994; Azipiazu et al., 1996; our own observations). Although only one mesodermal primordial cell gives rise to DM cells, we cannot discriminate to date whether the initial competence to develop as DM progenitor is conferred to two or only one mesodermal cell per segment. However, if two such cells are selected (flanking the mesectodermal cells, one on either side of the mesodermal anlage) a subsequent cross-talk between these cells must be postulated. Notch and other neurogenic genes might be candidate genes for this process and, indeed, approximately twice the number of DM cells were observed in Notch mutant embryos (Hartenstein et al., 1992). Directing the competent mesodermal cell towards the DM cell fate requires signals derived from the CNS midline. Secreted SPITZ is likely to represent this midline-derived signal, since overexpression of secreted SPITZ in the midline leads to additional DM cells. We observed a less efficient induction of DM cells when secreted SPITZ was expressed in all midline cells compared to expression in only few midline cells. This could indicate that the DM progenitor cell requires graded DER activation. Once the DM cell progenitor is specified, interactions with neighboring primordial mesoderm cells might ensure that the DM cell fate is maintained and that other mesodermal cells refrain from taking over the same fate. In summary, the midline cells of the Drosophila CNS, which in many respects are similar to the floorplate cells of the neural tube, act as an important signalling centre controlling ectodermal as well as mesodermal patterning. Future work is required to determine whether the underlying molecules have also been conserved.

We thank Jose Campos-Ortega, Lisa Meadows and Gerald Udolph for comments on the manuscript, Christof Rickert for assistence with the artwork and Phil Beachy, Ethan Bier, Stephen Crews, Lisa Fessler, John Fessler, Corey Goodman, Christian Lehner, Michael Levine, Ben-Zion Shilo and Olav Vef for providing fly stocks and antibodies. This work was funded by the DFG (grants to C.K. and G.M.T.) and the EC (grant to G.M.T.).