Cell shape changes during gastrulation in Drosophila

Morphogenesis is the sum of the processes that create complex three dimensional forms out of simpler structures. One such process is the folding of flat, essentially two-dimensional epithelia into more complicated organs, for example during neurulation, formation of the eye cup, and during gastrulation. Epithelial folding has been studied in many organisms and various models have been proposed to explain the mechanics of cell and epithelial shape changes, but the molecular mechanisms of the processes are not understood. Molecules that are likely to play important roles in shape changes have been identified and analysed biochemically and their roles have been studied in vitro, but their functions in intact tissues or the whole organism have hardly been investigated. One way of studying function in vivo is by genetic analysis, and for this Drosophila is a particularly convenient organism. The first morphogenetic event during Drosophila gastrulation, the formation of the ventral furrow, is well suited to study morphogenesis genetically, because it is a relatively simple process and many of the genes that determine the fates of the cells involved are already known.

Before gastrulation begins, the Drosophila embryo consists of a single cell layer epithelium of about 5000 morphologically identical cells. On the ventral side of the embryo, a strip of cells invaginates to form a tube, the prospective mesoderm (Fullilove et al. 1978; Turner and Mahowald, 1977, see Fig. 1). The mesodermal tube and the overlying ectoderm (the germ band) elongate around the posterior end of the embryo in a process called germ band extension (Fig. 1D – H). At the end of germ band extension, the mesoderm forms a layer on the inside of the ectoderm. The major gastrulation movements and some aspects of cellular behaviour have been described previously (Poulson, 1950; Sonnenblick, 1950; Mahowald, 1963; Rickoll 1976; Turner and Mahowald, 1977; Fullilove et al. 1978; Campos-Ortega and Hartenstein, 1985), but to analyze the regulation and mechanics of epithelial folding during gastrulation and to understand the function of the molecules involved, it is necessary to know the details of the events to be investigated. Therefore we began our analysis of gastrulation by describing in detail the behaviour and shape changes of the cells involved in this process.

Because many of the genes that determine the fates of different cell populations in the embryo are known, it is possible to alter cell fates genetically and thereby analyse the role of the different populations in gastrulation. Fates along the dorsoventral axis of the embryo, including the ventral region that will form the mesoderm, are determined by the dorsal group of maternal effect genes (reviewed in Anderson, 1987). One of these, the gene dorsal, codes for a nuclear protein, whose uptake into the nucleus is regulated by the other dorsal group genes. The dorsal protein becomes localized in the nuclei on the ventral side, but remains in the cytoplasm on the dorsal side of the embryo (Roth et al. 1989; Rushlow et al. 1989; Steward, 1989). Mutations in dorsal group genes can alter the localisation of the dorsal protein such that either all or none of the nuclei in the embryo take up dorsal protein, or intermediate situations are created. As a result, all cells can take on dorsal or ventral fates, or the regions in which these fates occur can be increased or decreased in size (Roth et al. 1989). We use such embryos to test to what extent the behaviour of ventral cells during gastrulation depends on the fate of neighbouring cells and other regions of the embryo.

The localisation of dorsal protein determines which zygotic genes are activated or repressed along the dorsoventral axis. Two zygotic genes, twist and snail (Nüsslein-Volhard et al. 1984), known to be required for the development of the mesoderm (Simpson, 1983), are expressed in ventral cells with high nuclear concentrations of dorsal protein. Both code for proteins that are probably transcription factors (Thisse et al. 1988; Boulay et al. 1987) and thus control cell fate via the activation of other genes. We show that twist and snail independently control different aspects of ventral cell behaviour.

As wild-type flies we used homozygous white flies. The snail allele was sna^IIG, twist was twi^ID96 or Df(2R)twi^S60 (Simpson, 1983), the double mutant was sna^IIG Df(2R)twi^S60 and was kindly provided by Kavita Arora. The string allele was stg^7M. torso mutant embryos were derived from tor^XR1/tor^XRI mothers. The following alleles of maternal dorsalizing genes were used: dl¹/Df(2L)TW119, pll⁰⁷⁸/pll^RM8, Tl^5BRE/ Df(3R)ro^XB3, spz^RM7/spz¹⁹⁷ (completely dorsalized), spz⁶⁷/ spz^l97(weakly dorsalized at 18°C, Fig. 3), ndl⁰⁴⁶/ndl⁰⁹³, ea¹/ ea², tub as well as the ventralizing Toll allele T1^10B. Yash Hiromi provided the P-insertion line P336 in which the mesectoderm cells express β -galactosidase.

Mutant embryos were distinguished from wild-type or heterozygous embryos by the absence of twist protein in the case of twist mutants and double mutants. In snail embryos this criterion could not be used. All embryos expressed the twist gene, including the homozygous mutant snail embryos, which can be recognized by their phenotype at the extended germ band stage (see Fig. 4). We were initially unable to distinguish mutant embryos at the syncitial and early cellular blastoderm, but after sectioning it became clear that about 25 % of the embryos appeared different from wild-type embryos (see Results) and we concluded that these were snail mutant embryos. We later confirmed our identification of mutant embryos by using marked balancer chromosomes that allowed us to distinguish mutant from wild-type embryos at early stages.

The antibodies against twist and dorsal protein were gifts from Siegfried Roth (Roth et al. 1989), anti-zen from Chris Rush-low and anti-β -galactosidase from Ulrike Gaul. The second antibody was biotinylated goat anti-rabbit from Vector Labs. Embryos were stained according to standard protocols. The antibodies were detected histochemically using the Vectastain ABC Kit from Vector Labs.

Stained embryos were dehydrated through a methanol series followed by two 10min incubations in 100% ethanol. They were then transferred to dry acetone and finally to a 1:1 mixture of dry acetone and Araldite. Araldite was Durcupan ACM from Fluka and was prepared by mixing 100 ml of reagent A with 100 ml of reagent B, then adding 3.5 ml of reagent D and finally 2 ml of reagent C. This mixture was stored frozen until use. Embryos were left in the acetone:Araldite mixture in an open dish overnight until the acetone had evaporated. They were then oriented for sectioning, polymerized at 60 °C for 2 days and sectioned on a Reichert-Jung Supercut 2050 microtome. The sections were mounted in Araldite or permount and photographed on a Zeiss Axiophot with Nomarski optics.

The insert from the cloned DNA containing the snail gene was purified from agarose gels and labelled with Digoxigenin using the kit from Boehringer Mannheim, following the directions provided. The labelled probe was hybridized to embryos and detected as described by Tautz and Pfeifle (1989). Labelled embryos were treated for sectioning and photography in the same way as the antibody-stained embryos described above.

Fig. 1 shows whole mounts of embryos at successive stages of gastrulation. The embryos are stained with antibodies against the twist protein, a nuclear protein expressed in prospective mesoderm and endoderm cells. The mesoderm cells initially occupy the ventral surface of the embryo (Fig. IB) and invaginate during gastrulation to form the mesodermal cell layer underlying the ectoderm (Fig. 1H). The formation of the tube of mesoderm from the ventral epithelium takes about 15 min (Fig. 1B-1D), the conversion of the invaginated tube into the final mesodermal cell layer (Fig. 1D – 1G) just over an hour. To analyse the behaviour of the invaginating mesoderm cells in detail, we made transverse sections of gastrulating embryos.

Fig. 2A shows an embryo at the beginning of cellular-ization. The nuclei are fined up at the periphery of the egg in a thick layer of cortical cytoplasm which takes up about one-third of the diameter of the egg. The centre of the egg is filled with yolk. About 25 % of the periphery is occupied by nuclei that contain the twist protein and will later make the mesoderm. At the edge of the region of strongly stained nuclei there are one or two nuclei that are stained much more weakly. These are the future mesectoderm cells, which become joined at the ventral midline when the mesoderm has invaginated (see also Fig. 5S), and later form part of the nervous system (Campos-Ortega and Hartenstein, 1985). In Fig. 2A membranes have started to invaginate between the nuclei, this process being more advanced on the ventral side. This is the first morphologically visible difference between the dorsal and ventral sides of the embryo (apart from the overall egg shape) (Mahowald, 1963). This asymmetry is maintained until the completion of cellularization (Fig. 2B), so that the cells on the dorsal side do not take up all of the cytoplasmic layer and are shorter than the ventral cells. As soon as the cellular blastoderm has formed, the ventral cells become flat on their apical surfaces, while the surfaces of the other cells remain rounded (Fig. 2B,C). This flattening may be the beginning of the contraction of the ventral surface, which becomes clearly visible in Fig. 2D. Simultaneously, the nuclei of the central 8 – 10 cells begin to move inwards (Fig. 2C,D).

At this stage two subpopulations can be distinguished within the prospective mesoderm, a central and a peripheral population. The central population is about 8 – 10 cells wide. These cells contract apically and their nuclei move away from the periphery of the egg. We will refer to this population of cells as ‘central cells’. The other cells (which we will call ‘peripheral cells’), lying on either side of the central population in strips about four cells wide, show the opposite behaviour. Their nuclei remain close to the peripheral surface of the egg, their apical membranes expand (or are stretched), while their basal surface becomes very narrow. Both populations express the twist protein. The processes of apical narrowing and nuclear movement continue and the ventral epithelium forms an indentation (Fig. 2E,F). The central cells transiently become taller and narrower (Fig. 2F). In some strains bulbous projections from’the apical surfaces of the central cells appear at this stage (see Fig. 6 and Rickoll, 1976). The dorsal cells become tall and thin and their nuclei lose their regular alignment as the dorsal epithelium begins to form transverse folds (Fig. 2G,H and Fig. 1D – F). The ventral indentation increases and folds into the embryo as a tube (Fig. 2G – I). Figs 2G,H show embryos at approximately the same stage of development. The wide opening of the ventral side of the embryo in Fig. 2G is artificial. The live embryo develops inside a membrane (the vitelline membrane) which has to be removed for antibody staining and which holds the developing ventral fold together. In some cases (Fig. 2H) but not always (Fig. 2G) this state is maintained after fixation and removal of the vitelline membrane. After much longer and stronger fixations and in live embryos the open state of the ventral furrow is not seen. The fact that the ventral furrow can open up in this way shows that at this stage no intercellular bonds have been formed between the lips of the ventral furrow. In the tube that is formed by the invagination of the ventral furrow, the two subpopulations of the mesoderm can still be recognized (Fig. 2H). The central cells form the tube proper, while the two strips of peripheral cells form the stem. The most peripheral cells of the prospective mesoderm still lie on the surface of the embryo.

The tube now loses its typical epithelial appearance, flattens out (Fig. 2J) and disperses into single cells that undergo mitosis (Fig. 2K). The dispersal does not depend on mitosis, since it also occurs in string mutant embryos (see Fig. 7A), in which this and later cell divisions do not take place (Edgar and O’Farrell, 1989). The mesoderm cells begin to spread out laterally until they form a single cell layer on the inside of the ectoderm (Fig. 2L). The mesectoderm cells come to lie next to each other on the ventral midline, slightly invaginated. At this stage, the twist protein has disappeared from their nuclei.

In summary, the invagination of the mesoderm can be divided into two phases. During the first phase, the cells within the ventral epithelium change their shapes in such a way that the central part of the epithelium bends and is displaced towards the inside of the embryo. The more peripheral cells follow. During this phase all cells remain attached to each other within the epithelium. In the second phase, the epithelial character of the mesoderm is lost as the cells enter mitosis and spread out to form the second germ layer. It is probably in this second phase that cell-cell and cell-matrix interactions become important. For example, the Drosophila integrins are first expressed on the cell surface at this stage (see Fig. 7), supporting genetic evidence that at least this class of cell interaction molecules plays no role in the first phase of gastrulation (Leptin et al. 1989).

The following paragraphs describe the behaviour of cells in mutant eggs in which the determination and the differentiation of the mesoderm are affected. The maternal effect genes of the dorsal group are required for determining the different fates of cells along the dorsoventral axis (Anderson, 1987). Situations can be created genetically in which all cells have identical fates, either dorsal (dorsalized embryos) or ventral (ventralized embryos). We use four molecular markers (visualized with antibodies) for specific cell populations along the dorsoventral axis. First, the subcellular location of the dorsal gene product is a marker for fate along the dorsoventral axis: dorsal protein in the cytoplasm corresponds to dorsal fates, while dorsal protein in the nucleus determines ventral fates (Roth et al. 1989; Rushlow et al. 1989; Steward, 1989). Second, we use the twist gene product as a marker for ventral cells (Thisse et al. 1988); third, the product of the gene zen is a marker for dorsal cells (Rushlow et al. 1987). Finally, the expression of β -galactosidase under the control of an enhancer that directs expression in mesectoderm cells is used to mark these cells.

In Fig. 3 we show that mutant embryos in which all cells have either dorsal or ventral fates, all cells behave identically during cellularization. The cell membranes invaginate equally at all positions around the egg and at the cellular blastoderm stage all cells look alike. After the completion of cellularization, dorsalized and ventralized embryos begin to develop differently. In dorsalized embryos the nuclei of all cells remain at the periphery of the egg (Fig. 3A), as they do in dorsal cells in wild-type embryos (Fig. 1H). In contrast, in ventralized embryos (in Toll^10B mutant embryos), nuclei at many positions move towards the centre of the egg and the apical surfaces of most cells flatten (Fig. 3B). These are characteristics of the central population of ventral cells in wild-type embryos. Thus, the above aspects of the behaviour of individual dorsal and ventral cells do not require interactions with neighbouring cells with different fates but are part of an autonomous process of differentiation.

The phenotype of Toll^10B mutant embryos is variable, and often incompletely ventralized embryos show interesting abnormalities in ventral furrow formation. For example, although all cells around the periphery appear equally ventralized in the embryo shown in Fig. 3C, as judged by twist expression, there is still some dorsoventral asymmetry, and a furrow is formed ventrally. Thus, there are two types of twist-expressing cells in this embryo, and only those on the ventral side behave like the central population in wild-type embryos. This region is larger than in the wild type, more cells than usual are recruited into the furrow and, probably as a consequence, their shapes appear more elongated. These cells invaginate and generate a fairly normal looking tube of mesoderm, except that both the tube and the stem contain more cells than normal (Fig. 3D).

In incompletely dorsalized embryos, the mesoderm and the ventral furrow are reduced in size (Roth et al. 1989). Nevertheless, this narrow band of twist-expressing cells is able to invaginate completely (Fig. 3E,F; since the mesoderm develops only in the posterior part of these embryos, the ventral furrow is only seen in the part of the extended germ band that has folded around the dorsal side of the embryo, and not on the ventral, more anterior, part). These mutant phenotypes demonstrate that the overall concerted behaviour of the invaginating cell population does not depend on the size of the population.

The maternal genes determine fate along the dorsoventral axis by regulating (directly or indirectly) the transcription of zygotic genes. Two zygotic genes, twist and snail (Nüsslein-Volhard et al. 1984), are known to be required for the development of the mesoderm (Simp son, 1983). Both genes have been cloned and their sequences suggest that they may be DNA-binding proteins (Boulay et al. 1987 ; Thisse et al. 1988). Their products are localized in the prospective mesoderm (Thisse et al. 1988; this work).

Fig. 4 shows embryos mutant for twist, for snail and mutant for both genes (double mutants). All embryos were stained with twist antibodies. At the beginning of germ band extension, when the ventral furrow has formed in wild-type embryos (Fig. 1D), no invagination is visible in the three types of mutant embryos. Later, at the fully extended germ band stage, there are two cell layers in wild-type embryos (Fig. 1H; the mesoderm and the ectoderm), but only one in mutant embryos (Fig. 4, second row). However, in ventral views of mutant embryos small furrows can be distinguished (Fig. 4, bottom row and Simpson, 1983). In. snail embryos these folds are irregular and are neither centrally located nor parallel to the anterior-posterior axis. In twist mutants, the furrows are often quite deep and long and always run along the ventral midline. In double mutants, folds are rare and never very long.

In the following we describe how the absence of twist and snail function affects the behaviour of ventral cells as seen in sections. In all three classes of mutants, the first signs of shape changes in ventral cells occur much later than in wild-type embryos, beginning around the same time as germ band extension, when the ventral furrow is already invaginating in wild-type embryos (Fig. 5, A – D).

In twist embryos, the ventral furrow is formed by a strip 8 – 10 cells wide. These cells become cylindrical and invaginate. During this process, they produce membrane protrusions on their apical surfaces, like wildtype cells, but their apical surfaces do not flatten (Fig. 6). Nevertheless, a furrow forms, showing that the apical contraction is not absolutely required for the formation of a furrow. No specific changes are seen in the neighbouring more lateral cells, and their apical sides do not become stretched (Fig. 6). Although the ventral furrow in twist mutant embryos is quite deep, it is not stable. By the end of germ band extension it flattens out again and forms a continuous epithelium with the ectoderm. The region of uninvaginated mutant mesoderm is bordered by the two rows of mesectodermal cells, which would normally have become joined at the midline.

The first sign of a difference between ventral and other cells in snail mutant embryos is a thinning of the whole ventral epithelium (Fig. 5C). The ventral cells become shorter and more cuboidal and their nuclei remain very close to the apical side of the ventral cells. As the flattening becomes more extreme, the epithelium begins to buckle (Fig. 5G). The irregular folds do not necessarily appear along the midline of the epithelium, but can form anywhere. As the germ band elongates, the epithelium straightens out again and the folds disappear.

In embryos mutant for both twist and snail very few morphogenetic changes occur in the ventral epithelium. Occasionally, small transient furrows are formed. The example shown in Fig. 5H is one of the strongest cases we observed.

Twist, snail and double mutants have one feature in common: towards the end of germ band extension, the mutant mesoderm is still part of the outer cell layer of the embryo. Thus, the ventrolateral ectoderm (the neurectoderm) in these mutants cannot move ventrally as far as it does in wild-type embryos, and the more dorsal epithelium (which flattens and expands during the normal course of gastrulation) can also not spread as far ventrally. As a result, it forms two deep folds that invaginate where the dorsal ectoderm and the neurectoderm meet (Fig. 5J,K,L) This shows that the expansion of the dorsal epithelium does not exert enough pressure on the lateral epithelium to push it ventrally, and therefore the invagination of the ventral furrow in wildtype embryos cannot merely be a passive response to pressure from the lateral epithelium.

The shapes of cells in the ventral furrow of wild-type embryos suggest that the ventral epithelium consists of two subpopulations, both expressing the twist protein. This distinction could be a mechanical consequence of the way the furrow forms, but the appearance of the furrows in twist and in incompletely ventralized Toll^10B embryos makes this unlikely. Alternatively, the difference between central and peripheral cells could be determined by a gene specific to either of these populations. The twist gene cannot be this gene, since it is expressed in both populations (unless differences in twist protein level undetectable by our methods distinguish the two populations). Since the only other gene known to affect the mesoderm early is snail, we tested whether the snail gene was specifically expressed in either subpopulation. We find that the snail RNA is expressed in a strip of the same width as the twist protein (Fig. 7), that is in the whole prospective mesoderm, and is therefore probably not responsible for the subdivision of the mesoderm into central and peripheral cells.

In wild-type embryos the ventral furrow always forms along the ventral midline, over about 80% of the anterior-posterior length of the egg. Thus, the region of invagination is not co-extensive with twist expression along the anterior-posterior axis, since twist staining extends all the way to the anterior and posterior poles of the egg. It seemed possible that the region of invagination is dependent partly on the mechanical properties of the egg, which might not allow furrow formation close to the narrower poles of the egg. However, the following results indicate that the length of the invagination along the anterior-posterior axis is also genetically determined.

In torso mutant embryos the posterior and anterior poles take on more central fates (reviewed in Niisslein-Volhard et al. 1987; Klingler et al. 1988). In such embryos, the ventral furrow extends over the whole length of the embryo (Fig. 7). Thus, there are no mechanical constraints that prevent mesoderm invagination near the poles. Therefore, the cells near the poles that do not invaginate in wild-type embryos are probably distinguished genetically from the more central cells that do. This is confirmed by in situ hybridisations with a probe for the snail gene (Fig. 7). Only the region that invaginates is labelled with the snail probe, while the termini of the embryo do not express snail. In torso mutant embryos, which form a furrow along their whole length, the expression domain of snail extends to the poles (Fig. 7).

The description of wild type ventral furrow formation presented here was compiled as a basis for the analysis of the molecular mechanics of this process. A number of models and mechanisms have been proposed to explain epithelial folding (reviewed in Ettensohn, 1985). Some aspects of these models are inconsistent with the data presented here. Of the forces that have been considered to account for epithelial invagination - cell growth and division, cell adhesion, and cytoskeleton-mediated changes in cell shape - the first (cell growth and division) cannot be active in Drosophila ventral furrow formation, as changes in cell size are not apparent and cell division does not occur in the mesoderm until after it has invaginated.

It has been suggested (Gustafson and Wolpert, 1967) that an increase in cell adhesiveness could cause neighbouring cells in an epithelium to increase their area of contact. If this occurred only at the apical end of the cells, this ‘zippering-up’ would cause an invagination. However, as the ventral furrow forms, the areas of membrane contact between neighbouring ventral cells do not increase noticeably. Instead, spare plasma membrane is protruded in bulbous villi from the apical surfaces of cells that change shape. These phenomema are inconsistent with shape change through increased membrane adhesion.

Cytoskeleton mediated cell shape change is the mechanism best supported by histological and experimental evidence (reviewed in Ettensohn, 1985). Apical contraction of the actin filament system has been suggested as a driving force in neural tube formation (Bumside, 1973), and computer modelling (Odellet al. 1981) shows that a wave of apical contractions of cells in an epithelium can cause an invagination in the epithelium. Here, however, such a wave of contractions does not seem to occur. Rather, the 8 – 10 ventralmost cells go through all their morphogenetic changes within a short period in an apparently random order. One often sees sections in which lateral cells have narrow apical ends, while closer to the midline they still have their original large apical surface area. (Such a stochastic sequence of cell shape changes in an invaginating epithelium is also observed in gastrulating amphibian embryos; Hardin and Keller, 1988.) The only cells of the prospective mesoderm that are ever seen to have expanded apical surfaces are the outer three or four on each side. We cannot exclude the possibility that a wave of contractions passes through the epithelium very quickly (between the stages that we show in Fig. 2B,C) and that later events are a consequence of this. However, in over 200 embryos we sectioned we never found one that would represent an intermediate in this process.

The cell shape changes in the central population of the prospective mesoderm are probably driven by more than one mechanism. The two main events occurring in these cells are the movement of the nuclei away from the apical surface of the cells and the contraction of the apical surface. Neither the sections we show here nor observations of live embryos indicate whether these events are causally related. However, the way the ventral furrow forms in twist mutant embryos suggests that they are not. In twist embryos, the nuclei of the ventral cells move away from the apical surface and the cells redistribute their contents such that they become more cylindrical, but the apical surface does not flatten or contract. The cells invaginate later and in a less orderly way than in wild-type embryos, but they do form a deep furrow. Thus, apical contraction and the displacement of nuclei are largely independent, and apical contraction is not an essential component of furrow formation. The combination of several mechanisms producing shape change might be a way of ensuring the reproducible formation of a stable furrow and of protecting the process from random external disturbances.

From the observations in Toll^10B, twist and snail mutant embryos, we conclude that the formation of the ventral furrow is an autonomous action of the ventral epithelium and does not depend on the behaviour of lateral or dorsal cells. As in wild-type embryos, the dorsal and lateral epithelia in twist and snail mutants flatten and expand during gastrulation. This expansion, however, does not exert enough force to push the ventral epithelium in snail or snail twist double mutant embryos into a furrow. The small transient folds in snail mutants might be a response to some pressure from the lateral epithelium (and also show that the epithelium is capable of folding), but they are much smaller than wild-type furrows. In incompletely ventralized Toll^10B embryos, the dorsal epithelium does not flatten or expand at all, because it has been transformed towards a more ventral fate. Nevertheless, a ventral furrow forms and invaginates completely. Therefore, the expansion of the dorsal and lateral ectoderm is neither sufficient nor required for ventral furrow formation and the only force driving the invagination must be the cell shape changes in the prospective mesoderm.

In amphibians, it has been shown by explanting parts of gastrulating embryos into tissue culture that many cell movements and shape changes can be carried out autonomously even by cells separated from their normal environment (reviewed in Gerhart and Keller, 1986). A similar situation is created genetically in Drosophila embryos in which all cells have identical fates. Ventralized embryos show that the apical flattening of ventral cells and the movement of their nuclei towards the interior does not depend on interactions with the lateral or dorsal epithelia. Thus, the nuclei are not squeezed inwards as a passive response to the action of forces from neighbouring cells (no inward movement is seen in dorsalized embryos), but are probably moved by forces from within the cells. At the very least, there must be a specific and ventral cell autonomous event that releases the nuclei to move inwards, or a specific dorsal mechanism that holds them at the periphery.

In mutant eggs in which the region that expresses ventral-specific genes is changed in size or shape, the ventral furrow is changed correspondingly. Our results suggest that the region of invagination along the anterior-posterior axis correlates more closely with the region of expression of the snail gene than that of the twist gene. Unlike the ventral furrow, the twist domain extends to the anterior and posterior poles of the embryo, whereas the snail expression domain is confined to the region that forms the ventral furrow and later the mesoderm. In torso mutant embryos, the snailexpressing region extends to the poles of the egg and so does the ventral furrow.

Also along the dorsoventral axis (i.e. the right-left axis in the mesoderm), the twist-expressing region consists of populations of cells showing different types of behaviour. In wild-type embryos, two subpopulations can be distinguished in the ventral furrow, a central and a peripheral population. Judging only from the description of the wild type, one might contribute this merely to mechanical reasons. It is possible that all cells expressing twist have a propensity to change their shape in the way the central cells do, but that cells at the edge of the field are subject to mechanical influences from cells outside the twist-expressing region which prevent this shape change. However, this seems very unlikely in view of the ventral furrows in incompletely ventralized Toll^10B embryos. In these embryos, all cells express the twist protein, but only a band of cells on the ventral side (and only ever on the ventral side) invaginate. Therefore, not all twist-expressing cells are determined to go through the cell shape changes characteristic of the central cells. The invaginating region in Toll^10B is wider than the ventral furrow in wild-type embryos, showing that there are no mechanical reasons why the whole twist-expressing region in wild-type embryos should not invaginate. It seems likely that the region that invaginates in Toll^10B corresponds to the wild-type central population, while the remaining cells correspond to the peripheral population, and that these regions are genetically distinct.

We do not know which gene might be responsible for the subdivision of the twist-expressing region, since both twist and snail are expressed in all cells. Differences in levels of twist protein might play a role, but even in very weakly stained embryos, cells on either side of the border between the central and peripheral cells are stained with equal intensity. The visible gradient of twist protein expression begins only in the most lateral part of the peripheral region. The region of peripheral cells could be determined by nonautono-mous influences from genes expressed in more lateral cells, or the ventral region could be subdivided by an as yet unknown gene.

Twist and snail both affect the same process, ventral furrow formation and mesoderm development. One might have expected that they would act at successive steps in a hierarchy of gene activity, and that one gene is under the control of the other. However, this cannot be the case, since the phenotype of the double mutant is more severe than that of either single mutant. Their phenotypes in terms of the behaviour of the mutant ventral cells differ, which means that they themselves control different aspects of cellular behaviour. It is important to point out that we cannot infer from the cell behaviour in twist and snail mutant embryos what aspects of cell shape change these genes control in wildtype embryos. The twist and snail gene products probably act by regulating the transcription of other genes, and do not affect cell shape directly. Therefore, the shape changes in ventral cells must require the transcription of at least two zygotic genes (one controlled by twist, the other by snail) in the ventral region, whose products interact with the pre-existing cellular machinery that mediates cell shape changes. Understanding how the ventral furrow is formed depends not only on identifying these downstream genes, and the postulated gene that distinguishes the central from the peripheral region of the ventral furrow, but also on determining how these genes affect the cytoskeleton to induce cells to change their shape.

We are very grateful to Siegfried Roth for discussions, suggestions, antibodies and some of the embryos used in this work. We thank Kavita Arora for providing the twist snail double mutant stock, Chris Rushlow and Ulrike Gaul for antibodies, Yash Hiromi for the mesectoderm-marker lines, Audrey Alberga for a clone of the snail gene, Rolf Reuter for the suggestion to look at string mutants and Nancy Hopkins, Daniel St. Johnston, Wolfgang Driever, Rudi Winklbauer, Helen Doyle, Manfred Frasch, Gos Micklem and Christiane Niisslein-Volhard for comments on the manuscript.