Mutations in a novel gene, myoblast city, provide evidence in support of the founder cell hypothesis for Drosophila muscle development

The larval muscles of Drosophila melanogaster form a complex but highly regular array on the body wall of the larva. The muscles are formed during embryogenesis by the fusion of myoblasts, cells of the somatic mesoderm. Midway through development, small syncytia containing two or three nuclei appear at defined points in the somatic mesoderm (Bate, 1990). These syncytia are termed ‘muscle precursors’, as each one gives rise, by a continuing process of fusion and extension/migration, to a larval muscle, whose identity is defined by its size, position, orientation on the body wall of the larva and pattern of innervation.

Interestingly, fusion does not take place randomly among myoblasts, but always begins at particular points in the mesoderm in close contact with the ectoderm (Bate, 1990). A model for muscle development has been proposed in which muscle differentiation begins with the segregation of founder cells – single cells in the somatic mesoderm – which seed the fusion process and which may in some way confer identity on the developing muscle (Bate, 1990; Dohrmann et al., 1990; Bate et al., 1993).

This view is supported by the observation that genes such as S59, apterous (ap) and vestigial (vg) are each expressed in a small number of larval muscles, and that earlier, myoblasts expressing these genes behave like the predicted founder cells. Expression of these genes begins in the mesoderm in a small number of cells before fusion. These cells fuse with neigh-bouring cells, which then begin to express S59, ap or vg themselves, and the resulting fused groups go on to form specific muscles (Dohrmann et al., 1990; Bourgouin et al., 1992; Bate, 1993; Bate et al., 1993). S59 and ap contain homeobox sequences and so are implicated in DNA binding and transcriptional control (Dohrmann et al., 1990; Cohen et al., 1992). vg is related to a family of proteins involved in protein-protein interactions (Williams et al., 1991). All three proteins are localised to the nucleus. The function of such genes in muscle development is not known, but it is possible that they are involved in specifying the identity of the muscle precursors in which they are expressed. The best evidence for such a function is provided by ap. In the absence of ap function there is variable loss of one or more of the ap expressing muscles and strikingly, when ap is expressed ubiquitously under heat-shock control, extra muscles develop in a similar location and position to some of the wild-type ap-expressing muscles, while the rest of the muscle pattern is apparently normal (Bourgouin et al., 1992).

The idea of the muscle founder cell in Drosophila was preceded by the observation of pioneering muscle cells in the grasshopper embryo. Muscle pioneers in the grasshopper stretch out and span the future territory of the muscle they prefigure. The pioneers form a scaffold with which mesodermal cells fuse to form the mature musculature (Ho et al., 1983). Superficially, the muscle pioneers of the grasshopper are different from the muscle founder cells of Drosophila, because the pioneers extend processes to span their territory before fusion takes place, while the proposed founder cells first initiate fusion and form multinucleate muscle precursors. It is these precursors in Drosophila that resemble the muscle pioneers of the grasshopper in that they start to explore their territory and form attachments to the epidermis. This difference in timing may simply reflect the more rapid process of embryogenesis in Drosophila as compared with the grasshopper. While the existence of founder cells in Drosophila has yet to be demonstrated, they have been proposed as a model to explain early events in the formation of the muscle pattern.

Here we describe the embryonic phenotype of lethal mutations in a gene called myoblast city (mbc). The principal characteristic of these mutations is lack of myoblast fusion, and we use this phenotype to examine the founder cell hypothesis in Drosophila. We present evidence in support of the founder cell hypothesis and show that, in the absence of fusion, two classes of myoblasts exist. The first class resembles muscle pioneers in that these cells stretch out and span muscle territories, while the second class appears to constitute a pool of myoblasts with which the pioneers would normally fuse.

All marker mutations and balancer chromosomes used are described in FlyBase (1994).

(a) Deficiencies in the 95 region were created as follows: The P{ry⁺, lacZ}A189.2F3 line (Bellen et al., 1989) was obtained from the Bloomington Stock Center. It carries an enhancer trap element at 95A, and homozygotes are phenotypically wild type. Males homozygous for P{ry⁺, lacZ}A189.2F3 were irradiated with 4000 rads of X rays. Approximately 26,000 chromosomes were screened for the loss of the ry⁺ marker. A single deficiency, Df(3R)mbc-R1, with breakpoints in 95A5-7 and 95D6-11 was recovered in this screen.

A parallel mutagenesis utilised an insertion of the plasmid H{ry⁺har1} at 95A (Blackman et al., 1989). This hobo-associated ry⁺ insertion (referred to as TnM2) is homozygous lethal, with no obvious muscle abnormalities. TnM2-bearing males balanced with TM6B (ry⁺) were irradiated with 4000 rads of X-rays and crossed to MKRS (ry)/TM2(ry) virgin females. Approximately 44,000 chromosomes were examined for loss of the ry⁺ insertion. A single deficiency, Df(3R) mbc-30, with breakpoints in 95A5-7 and 95C10-11 was recovered.

(b) Four alleles of mbc were identified in several screens, on the basis of failing to complement mbc deletions for embryonic viability: Homozygous red e males were fed 25 mM EMS in 5% sucrose (Lewis and Bacher, 1968). Mutagenized third chromosomes were recovered over the TM3 Sb balancer chromosome and test crossed against the Df(3R)mbc-R1 chromosome. Three independent lethal alleles of mbc (mbc^C1, mbc^C2 and mbc^C3) were recovered from a total of 1,440 chromosomes tested.

In a parallel analysis, isogenic ru st e males were fed 30 mM EMS in 10% sucrose. Mutagenized third chromosomes were recovered over a TM3 (Sb, ry) balancer and tested for complementation of Df(3R) mbc-30. One lethal allele of mbc (mbc^S4) was recovered from a total of approximately 2400 chromosomes tested. Independent recombination mapping localised this allele to cytological region 95.

Eggs were laid and allowed to develop on yeasted apple-juice agar plates at 25°C. Embryos were fixed for 20 minutes in heptane and 4% paraformaldehyde in phosphate buffer pH7.0, and devitellinized in a 1:1 mixture of heptane and methanol with vigorous shaking. Embryos were blocked prior to incubation with antibody, in phosphate-buffered saline, 0.3% Triton-X (PBS-TX) + 0.5% bovine serum albumin (Sigma). Anti-myosin, used at a dilution of 1:1000 was a gift from Dan Kiehart (Kiehart and Feghali, 1986). Anti-S59 was produced in collaboration with Fernando Jimenez and Mary Baylies, using an S59 cDNA clone kindly sent by Manfred Frasch (Dohrmann et al., 1990). Rabbit anti-S59 was used at a dilution of 1:5000. Anti-VG was provided by Sean Carroll and used at a dilution of 1:100 (Williams et al., 1991). Anti-Connectin was provided by Rob White and used at a dilution of 1:10 (Meadows et al., 1994). Anti-Groovin was provided by Talila Volk and used at a dilution of 1:3. Anti-Peroxidasin was provided by J. Fessler and L. Fessler and used at a dilution of 1:1000.

Incubation with antibody took place overnight at 4°C. All washes were in PBS-TX. Incubation in biotinylated secondary antibody (Vector Labs) was followed by Vectastain ABC Elite enhancement, and the staining was developed using 0.5% diaminobenzidene, 0.03% nickel chloride and 0.02% hydrogen peroxide. For the double stains, embryos were incubated first in anti-Peroxidasin, anti-Groovin, anti-S59 or anti-VG and developed in the presence of nickel chloride (giving a blue-black stain), then incubated in anti-myosin and developed in the absence of nickel chloride (for a pale brown stain).

The anti-myosin-stained embryos shown in Fig. 1 were aged by making one-hour egg-lays and allowing the embryos to develop for the appropriate length of time at 25°C. All other embryos were staged using morphological criteria.

Embryos were examined and photographed using a Zeiss Axiophot microscope.

Alleles of mbc were recovered in a screen designed to make mutations in the 95 region of the third chromosome. This region includes the gene nautilus (nau), a Drosophila homologue of vertebrate MyoD. nau is expressed from an early stage in developing muscles (Michelson et al., 1990). Results for nau will be published elsewhere; here we concentrate on the phenotype of mbc.

A total of 4 alleles of the mbc locus were isolated: mbc^C1, mbc^C2, mbc^C3 and mbc^S4. Deletion mapping using deficiency chromosomes (described in Materials and Methods) maps the mbc locus between 95A5-7 and 95C10-11 on the right arm of the third chromosome. Deficiencies that remove the nau coding sequence complement the mbc mutations (Erickson and Abmayr, unpublished results), thus mbc is not nautilus.

All known mbc alleles are recessive and cause embryonic lethality. Epidermal development and differentiation appear normal: cuticle preparations of mbc mutant embryos reveal no abnormalities in segmentation or patterning (data not shown). The embryos lie motionless in the vitelline membrane and fail to hatch. Examination of the mbc mutant embryos with a polarised light microscope (Drysdale et al., 1993) shows a striking lack of differentiated muscle.

The same muscle phenotype was consistently found in embryos homozygous for Df(3R)mbc-R.1, Df(3R)mbc-30 and all four mbc alleles, and also in embryos carrying any combination of deficiency and point mutation. This suggests that all mutations generated so far have a null phenotype.

In order to make a detailed analysis of the muscle phenotype of embryos mutant for mbc, we used an antibody against Drosophila muscle myosin to reveal the muscles during development (Kiehart and Feghali, 1986). In wild-type embryos, muscle myosin is first expressed approximately 9 hours after egg laying (AEL). At this time, myoblast fusion is under way and every muscle is represented by a precursor (Bate, 1990). Myosin expression begins in a small number of precursors lying ventrally and laterally and rapidly extends to all the pre-cursors and some of the cells that are about to fuse with them. By 13 hours AEL, the muscle pattern is complete and muscle attachments are forming on the epidermis (Fig. 1A).

In embryos mutant for mbc, myosin expression begins on schedule at 9 hours AEL, in cells that appear by their position to belong to the somatic mesoderm. Because of their position and muscle myosin expression, we identify these cells as myoblasts. As in the wild-type, expression begins in a few cells ventrally and laterally, and by 10 hours, apparently extends to most of the cells of the somatic mesoderm. The myosinexpressing cells are clearly organised into an array which resembles the pattern of wild-type muscle precursors, with a segmentally repeated arrangement of ventral, lateral and dorsal cell clusters (Fig. 1B). Strikingly, with only a few exceptions, these myoblasts fail to fuse. Initially, all the cells are rounded, but from 11 hours AEL, a subset of mononucleate cells become slightly elongated, lying in positions and orientations similar to the multinucleate muscle precursors in wild-type embryos (Fig. 1C,D,F).

The elongated myoblasts first appear at about 11 hours AELand increase in number until about 13-15 hours AEL. After this time some myoblasts become much longer, some now spanning distances two or three times the length of normal muscles. These myoblasts occasionally have more than one nucleus, indicating that a small number of cell fusions occur. At 13 hours AEL, there is still a large number of rounded myosin-expressing myoblasts, but from 14 hours onward, this population diminishes. Some cells may simply lose myosin expression, as many faintly stained and unstained rounded cells can be seen using Nomarski optics. However, cell death is also involved, as myosin-positive cells engulfed by macrophages can be seen in preparations which have been doubly stained for myosin and peroxidasin (Nelson et al., 1994; Fig. 2).

Groovin is expressed in the epidermis at muscle attachment sites (Volk and VijayRaghavan, 1994). To investigate whether the elongated myoblasts were indeed finding their correct attachment sites, embryos were double-labelled using anti-myosin and anti-Groovin. Fig. 3 shows myosin-positive cells stretching out and making contact with Groovin-positive epidermal cells, which appear normal in mbc mutant embryos. We conclude that at least some of the elongated myoblasts succeed in forming attachments with appropriate cells in the epidermis, and that myoblast fusion is not required for the normal patterning of muscle attachment sites on the epidermis.

In summary, there appear to be two populations of myoblasts within embryos mutant for mbc. Both populations are first visible by myosin expression at 9 hours AEL, as rounded cells, and one population remains rounded as long as they express myosin until at least 17 hours AEL, when cuticle formation prevents further antibody staining. Initially these cells are only slightly elongated, but from about 15 hours AEL they stretch and send out long processes (Fig. 1C). These two myoblast populations and their behaviours are reminiscent of the two myoblast types postulated to exist during wild-type myogenesis. The rounded myoblasts in mbc mutants might constitute the pool of myoblasts available for fusion, whereas the elongated cells might represent the founder cells, whose special properties are revealed in this mutant because fusion fails to occur.

To explore this idea, we stained mbc mutant embryos using antibodies against S59 and VG, the products of two genes that are expressed in putative founder cells (Fig. 4).

The early expression of S59 is identical in wild-type and mbc mutant embryos. S59 expression begins in a consistent pattern of a small number of cells in the somatic mesoderm. Here we focus on the sequence of events in a single abdominal hemisegment. In both wild-type and mbc mutant embryos, expression begins in a single ventral cell between 6 and 7 hours AEL, which divides to give rise to two cells, known collectively as are detectable, decreasing in number from about 13 hours AEL. The other type of cell, in contrast, can be distinguished as elongated cells from about 11 hours AEL, often situated in positions and orientations recognisably like those of wild-type muscle precursors, and these cells continue to Group I. Four cells posterior and slightly ventral to Group I begin expression at about 7 hours AEL; these are known as Group II. Group III is the last to appear, at about 8 hours AEL, consisting of two cells lying dorsally and at the same level in the anteroposterior axis as the cells of Group II.

Subsequently, S59-expressing cells undergo movements and pattern refinements and this rearrangement is identical in mbc mutants and wild-type embryos, as follows. After Group I has divided into two cells, cell Ib remains in the same place, while cell Ia migrates across the segment border in both mbc mutants and the wild-type. The behaviour of Group II in mbc mutants appears at first sight to be different from that of the wild-type, in that only one of the original four cells maintains S59 expression after germ band retraction (Fig. 4B,D). The original observations of Dohrmann et al. (1990) suggested that in the wild-type, all the cells of Group II maintain expression, contributing eventually to muscle 27 (muscle nomenclature according to Crossley (1978). However, more recent observations (Carmena, Bate and Jimenez, unpublished data) show that in fact, only one of the cells of Group II does so. The other cells gradually lose S59 expression and each contribute to separate muscles. The loss of S59 expression in three of the four cells of Group II in mbc mutants, therefore, exactly follows the sequence of S59-expression in wild-type embryos. Group III also maintains S59 expression in only one cell in embryos mutant for mbc (Fig. 4B,D). Once again it is probable that this corresponds to events in the wild-type as it is likely that the two cells of Group III give rise to two separate muscles, only one of which (muscle 18) expresses S59 (Carmena, Bate and Jimenez, unpublished data). Oddly, cell Ib does behave abnormally in mbc. In wild-type embryos, cell Ib continues to express S59 and gives rise to muscle 25. In mbc mutants, however, cell Ib loses expression (Fig. 4B,D). This is the more curious, in that we do see a putative founder cell for muscle 25 in mbc mutant embryos which have been stained for myosin.

In embryos mutant for mbc, therefore, the initial pattern of expression of S59 is almost identical to wild-type, suggesting that the segregation and movement of founder cells is normal in these mutants. There is, however, a dramatic difference in S59 expression between wild-type and mbc mutant embryos, in that muscle fusion fails to occur in mbc mutants and there is no concomitant increase in the number of nuclei positive for S59 (Fig. 4B,D). Thus, fusion is required for the recruitment of cells to express S59, as predicted by Dohrmann et al. (1990). The S59-positive cells remain as single cells, lying in approximately the same position as the muscle they would normally have given rise to, and continue to express S59 at least until 17 hours AEL – the limit of our ability to detect protein by antibody staining. In 3% of cases (n=200), two cells can be seen instead of one and we assume that this is caused by the rare fusion events that occasionally take place in mbc mutant embryos.

Like S59, vg is expressed in a small number of mesodermal cells (Bate, 1993; Bate et al., 1993). In the wild-type embryo, these cells contribute to ventral muscles 6, 7, 12 and 13 and dorsal muscles 1, 2, 3 and 4. Early development of the dorsal muscles is difficult to examine because vg expression in the epidermis overlies and partly conceals the dorsal mesodermal vg expression. Here we concentrate on ventral mesodermal vg expression. Ventrally, vg expression in wild-type and in mbc mutants begins in one cell per abdominal hemisegment during the extended germ band stage of development and soon increases to three or four cells. By 10 hours AEL, the cells lie in a small cluster in the posterior of the segment. So far, the pattern of vg expression is identical in wild-type and mbc mutant embryos. In wild-type, the vg-expressing cells further increase in number and resolve into four ventral longitudinal muscles (Fig. 4E). In mbc mutants, however, the vg-expressing cells do not increase in number, but otherwise behave in a similar fashion to the wild-type ones. The four cells separate and align themselves in a dorsoventral pattern in the positions normally taken by the ventral longitudinal muscles (Fig. 4F). Thus we conclude, firstly, that as for S59 expression, increase in the number of vg-expressing cells is due to fusion and recruitment. Secondly, the vg-expressing cells contain the necessary information to migrate to their correct positions in the segment.

The above results suggest that founder cells are segregated normally in mbc mutants and behave normally in every respect save that of fusion. To confirm that these cells correspond to the stretched myoblasts seen in the myosin-stained preparations, mbc mutants were double stained using antibodies to myosin and VG or S59 (Fig. 5). In mbc mutants vg or S59 positive nuclei are clearly seen in stretched myosin-expressing cells which span the territory that in wild-type is occupied by an S59 or vg-expressing muscle. We have never seen a vg or S59-expressing nucleus with a rounded cytoplasm. It is not always possible to see the cytoplasm of these cells owing to the many rounded myoblasts which surround them. However where the cytoplasm can be distinguished, the orientation of the cell is consistent with the orientation of the wild-type muscle which it represents. For example, S59-expressing muscle 27 was examined in 192 segments, and in 155 of these segments it was possible to distinguish a myosin-stained process with an S59-expressing nucleus. Of these, 138 (89%) were correctly oriented, running from ventral-anterior to dorsal-posterior. S59-expressing muscle 18 was harder to distinguish from surrounding myoblasts, with clear processes visible in only 121 out of 220 segments examined. However, of these 121 processes, 115 (95%) were correctly oriented. We conclude, therefore that myoblasts which express S59 or vg contain information which enables them to find their correct position and orientation.

A feature of normal Drosophila development is specific innervation of particular muscles by particular motor neurons. The Connectin protein is expressed on the surface of a subset of developing motor neurons and muscles and may be involved in mediating homophilic adhesion between them, prior to synapse formation (Nose et al., 1992; Meadows et al., 1994). We stained mbc mutant embryos with an antibody to the Connectin gene product and showed that a subset of myoblasts in the appropriate parts of the segment express Connectin on their surface, while the surrounding myoblasts do not. Moreover, Connectin-expressing myoblasts are contacted specifically by Connectin-expressing nerves (Fig. 6). This strongly suggests that these Connectin-expressing myoblasts have an identity which is recognised by motor neurons and which is not shared by the surrounding pool of myoblasts.

In this paper we describe the phenotype of mutants of a newly identified gene, mbc, which is required for normal myogenesis in Drosophila. In normal Drosophila embryos, muscles form by fusion of adjacent myoblasts. Each muscle is a unique element in a distinctive pattern and each has its own position, size, orientation, attachment sites and innervation. In mbc mutant embryos, myoblasts fail to fuse and no multinucleate muscles are formed, yet a subset of myoblasts retains the characteristics of position, orientation and specific innervation.

Two possible models could explain how fusion of myoblasts generates individual muscles each with its own identity and characteristics. In the first model, myoblasts are specified as a group in which all the cells contain the information as to which muscle they are about to form. According to this model, each myoblast is specified to form a particular muscle and no myoblast in a group is unique. If fusion were to fail in this system, one would expect to see all the unfused myoblasts behaving in the same way, perhaps all sending out processes to span the territory of the muscle they would normally form. In the second model, a single cell is specified to become a particular muscle and this cell is capable of seeding the process of fusion in the surrounding cells, which then take on the identity of the cell with which they have fused. In this second model, we might expect to see a mass of myoblasts without identity or distinguishing characteristics and a small population of myoblasts that have some of the characteristics of the muscles they would normally form. These characteristics might include the expression of certain genes, and exploration of the territory normally covered by that muscle.

In support of the second model, there are indeed, in mbc mutants, two different populations of myoblasts, one that remains rounded throughout embryogenesis and one that becomes elongated. This apparent subdivision could be a result of random behaviour of the myoblasts, as a consequence of failure of fusion or some other aspect of the mbc mutant phenotype. However, in embryos mutant for mbc we can identify S59 and vg-expressing cells with the stretched myosinexpressing cells, in preparations that have been double-stained. For example, some vg-expressing cells have processes that span the region which in wild-type would be spanned by vg-expressing muscles. The orientation of these single-cell ‘muscles’ is not always accurate, but this is most likely because these myoblasts explore their surroundings later than wild-type muscle precursors, and the surfaces over which they migrate may be expressing different proteins. Moreover, the unfused myoblasts themselves may make the terrain confusing to exploring cells. It is perhaps the more surprising therefore that so many of the founder cells we see are in the correct orientation.

In mbc mutants, S59 or vg-expressing cells appear at the correct time and place, and migrate correctly, but fail to recruit surrounding cells to S59 or vg expression, showing that the S59 and vg-expressing cells are a distinct population of myoblasts with their own identity. Clearly, as predicted by the founder cell hypothesis, neighbouring myoblasts cannot acquire this identity in the absence of fusion.

We have shown in mbc mutant embryos that Connectin is expressed on the surface of a subset of unfused myoblasts and on nerves making contact with these myoblasts. We argue therefore that these particular myoblasts have an identity which can be recognised by the outgrowing motor axons and which is not shared by the surrounding myoblasts. This observation is also consistent with the founder cell hypothesis, as it shows that only a subset of myoblasts are able to specify a characteristic pattern of innervation.

Many genes may be required for fusion to occur successfully. There are already several reports of genetic loci, which have not so far been completely characterised, but whose mutant phenotypes show partial or complete loss of fusion. These include two P-element induced mutations, rolling stone and P-20 (Paululat et al., 1994a,b), several first-chromosome deficiency lines, and runt (Drysdale et al., 1993). Furthermore, myoblast fusion fails when a constitutively active form of Drac1, the Drosophila homologue of Rac, is expressed in the mesoderm. The function of Drac1 is not known, but it is expressed in the mesoderm from about 6 hours AEL. Embryos that express the mutated Drac1 have a number of features in common with embryos mutant for mbc, including stretched myosin-expressing cells and a Connectin expression pattern which appears identical to that of mbc mutants (Luo et al., 1994). This suggests that the myoblast morphology described for mbc mutants is indeed a secondary effect due to lack of fusion and not a phenomenon restricted to mbc mutants.

To summarise, therefore, we suggest that the stretched cells in mbc are founder-cell like. They constitute a special population of cells which in wild-type cannot be distinguished morphologically because by the time muscles are stretching and extending processes, the cells are part of a syncytium (Bate, 1990). In mbc mutants however, the founder cells are revealed because of the lack of myoblast fusion.

There is a clear analogy between the cells described here and the muscle pioneer cells of the grasshopper (Ho et al., 1983). Both are large, distinctive cells which extend processes to explore their surroundings and both stretch out to span the territories of the future muscle. A prediction from the founder cell hypothesis is that if a founder cell is removed, the muscle it should have formed would be missing. This prediction was tested in the grasshopper. When individual muscle pioneers were ablated, no muscle subsequently formed, though the myoblasts that normally contributed to it were still present (Ball et al., 1985).

We therefore believe that myogenesis is essentially identical in grasshopper and Drosophila. Both require a special class of cells that initiate fusion. Drosophila’s more rapid development may explain the different sequence of events, in that founder cells form a syncytium before spanning their territories, while the grasshopper pioneers are visible first as single, stretching cells. This rapid development obscures the essential similarity of the founder cells to the pioneer cells, and this similarity is revealed only in a mutant where fusion does not take place. Under these special conditions, a particular subset of cells is revealed, which resemble grasshopper muscle pioneer cells, and we identify these cells with the founder cells that have been proposed to explain the process of muscle development in Drosophila.

There are some puzzling aspects of the phenotype of mbc mutants, some of which reveal unexpected features of normal myogenesis. Firstly, the observation, as revealed by myosin staining, that the population of stretched cells in mbc mutants does not emerge until some time after muscles have spanned their territories in wild-type embryos. It certainly is not the case that muscle development in general is delayed in mbc mutants, since muscle-specific proteins such as S59, VG and myosin are all expressed on schedule, and the S59 and vg-expressing myoblasts variously migrate, divide and lose expression at the correct times. It may be that normal muscle growth and extension occurs in two phases. At first growth may be passive, as a result of fusion of myoblasts into the syncytium. Once fusion is complete, further extension must take place actively by the process of sending out muscle growth cones. Possibly it is this late stage of active extension that is revealed in embryos mutant for mbc.

Secondly, in mbc mutants, rounded myoblasts decline in number from about 14 hours AEL onwards, as detected by antimyosin staining. It appears that this is due in part to cells losing myosin expression, and in part to cell death, as detected by the presence of macrophages that have engulfed myosin-expressing cells. In either case, it seems that fusion is a requirement in non-founder cell myoblasts to maintain viability and/or a muscle fate. Founder cells, on the other hand, appear to contain the information necessary to sustain myosin expression and other aspects of muscle differentiation. It could be that the distinctive characteristics of founder cells depend at least in part on contact with the epidermis. Three observations support this idea: firstly, that S59 and vg-expressing cells arise in close contact with the ectoderm and remain in contact (our own observations). Secondly, that the stretched cells in mbc mutants appear to contact the epidermis, and thirdly, that in neurogenic mutants, lack of epidermis leads to premature loss of expression of such genes as S59 and vg (Bate et al., 1993).

We cannot explain the curious observation that in mbc mutants, S59 expression is lost from cell 1B. As described above, Group I of the S59-expressing cells is first seen as one cell, which divides into two. Cell Ia migrates across the segment border, loses S59 expression in both wild-type embryos and mbc mutants and in wild-type embryos becomes incorporated into muscle 5. Cell Ib migrates a couple of celldiameters posteriorly and fuses with its neighbours to form S59-expressing muscle 25. The loss of S59 expression from cell Ib in mbc mutants is therefore a mystery and the only light we can shed on the puzzle comes from the observation that in mbc all the cells that retain S59 expression are lying in contact with the anterior margin of the engrailed- (en) expressing stripe in the ectoderm (our unpublished observations). It may be that this is in some sense a ground state for mesodermal S59 expression. All the mesodermal S59-expressing cells arise in contact with the en stripe, and cell Ib is the only S59-expressing cell in mbc mutants to lose contact with it (Dohrmann et al., 1990; our observations). It may be that in the wild-type, some aspect of fusion overrides the need for contact of cell Ib with en-expressing cells.

To conclude, we here present evidence in support of the founder cell hypothesis of larval muscle development in Drosophila. This model suggests that single cells of the somatic mesoderm are selected and set aside to initiate the process of muscle fusion. In addition, these founder cells contain the information that gives the future muscle its identity. This information is required from a very early stage, as shown by the patterns of gene expression and cell migrations of the single-cell muscle founders. It is also required later in development to determine the position and orientation of the muscle and its innervation pattern. We have shown here that single cells of the somatic mesoderm are capable of displaying the above characteristics even in the absence of fusion and we therefore identify these cells as muscle founder cells.

We thank Jacob Harrison for technical assistance, and Mary Erickson, Ana Carmena and Fernando Jimenez for communicating results prior to publication. We also thank Mary Baylies, Kendal Broadie, Andreas Prokop, Helen Skaer and Mike Taylor for critically reading earlier versions of this manuscript and making many helpful suggestions. This work is funded by grants from The Wellcome Trust (E. R., R. D. and M. B.) and the National Science Foundation (S. A.). A. M. is an Assistant Investigator of the Howard Hughes Medical Institute.