Real-time imaging of cell-cell adherens junctions reveals that Drosophila mesoderm invagination begins with two phases of apical constriction of cells

Invagination or folding of an epithelial cell sheet is seen in a variety of morphogenetic events in metazoan development. Such epithelial morphogenesis often begins with constriction of the apices of specific cell populations. For example, avian neural tube formation and amphibian bottle cell formation, which have been analyzed extensively, initially exhibit cells with small apices in defined regions of the epithelia (Schoenwolf and Franks, 1984; Hardin and Keller, 1988). Constriction of cell apices appears to play an important role in the initiation of epithelial invagination or folding.

Invagination of the prospective mesoderm in Drosophila gastrulation is a well-studied morphogenetic event that begins with constriction of cell apices. At about 3 hours of development, a Drosophila embryo has about 6000 cells, all of which, with the exception of the germ cells, are essentially uniform in size and shape, and are organized in an epithelial layer. In the cellular blastoderm, a band of cells corresponding to the ventral one-quarter of the embryo (approximately 18 cells in width) is destined to be the mesoderm. The prospective mesoderm begins morphogenetic movement earlier than all other regions of the embryo, and is brought into the interior of the embryo within 20 minutes at 25°C. Although there have been many detailed studies of the morphology of the mesodermal cells during the invagination (Turner and Mahowald, 1977; Fullilove et al., 1978; Leptin and Grunewald, 1990; Kam et al., 1991; Parks and Wieschaus, 1991; Sweeton et al., 1991; Costa et al., 1993; Oda et al., 1998), most were performed with fixed embryos. Detailed observations using scanning electron microscopy and epi-illumination video microscopy (Sweeton et al., 1991) suggested that the process of mesoderm invagination is divided into four phases: (1) flattening of the cell apices and initial random apical constrictions; (2) a shallow groove stage associated with increased constriction and cell elongation; (3) cell shortening and the subsequent inward movement; (4) closing over of the invagination and germband extension. In addition, examination of living embryos using a combination of injection of fluorescent reagents that visualize cell outlines and nuclei, and time-lapse three-dimensional fluorescence microscopy, supported the idea of initial random initiation of cell shape changes and the subsequent groove formation stage, although the data sets were only acquired at 1-minute intervals (Kam et al., 1991). It was also suggested that cell-cell communication mediated by Folded gastrulation may involve a rapid, orderly progression of constriction initiations from the center to the periphery of the mesoderm (Costa et al., 1994). These earlier observations implied that the mesodermal cell behavior during gastrulation is too dynamic to be thoroughly described without examining living embryos at much higher time resolutions.

The classic cadherin-based cell-cell adhesion system has been shown to play a fundamental role in controlling cell behavior during morphogenesis of both vertebrates and invertebrates (Tepass, 1999; Steinberg and McNutt, 1999). In epithelial cells of embryos or tissues, E-type classic cadherin, a membrane-spanning protein, constitutes cell-cell adherens junctions together with many other components including β- and α-catenin (Yap et al., 1997; Steinberg and McNutt, 1999; Provost and Rimm, 1999). The adherens junctions, linking apical edges of epithelial cells, interact with cortical actin bundles (Provost and Rimm, 1999), and are thought to contribute mechanically to the constriction of cell apices in epithelial invagination or folding. In the Drosophila cellular blastoderm, adherens junctions are formed in the apical areas of the lateral surfaces of the cells (Tepass and Hartenstein, 1994). Drosophila E-cadherin (DE-cadherin) is present in the adherens junctions together with Armadillo and Dα-catenin (Oda et al., 1993; Oda et al., 1994; Müller and Wieschaus, 1996). In the ventralmost prospective mesodermal cells that have just begun to constrict their apices, adherens junctions are observed at the apical poles of the lateral cell surfaces (Oda et al., 1998). As the invagination proceeds, the adherens junctions appear to be disrupted in parallel with transition of the mesodermal cells from the epithelial to the mesenchymal state. Contraction of a cortical actin-myosin network may mediate apical constriction at the start of mesoderm invagination (Young et al., 1991; Leptin et al., 1992). Concertina (G protein α-subunit), DRhoGEF2 (guanine nucleotide exchange factor) and Rho1 (GTPase) may be involved in the Folded gastrulation (Fog) signaling pathway, which coordinates cell shape changes (Parks and Wieschaus, 1991; Costa et al., 1994; Barrett et al., 1997; Häcker and Perrimon, 1998).

In this study, we expressed DE-cadherin tagged with green fluorescent protein in early Drosophila embryos to enable real-time imaging of cell-cell adherens junctions before and during gastrulation. Low-light fluorescence video microscopy, with which we recorded images every second, directly visualized the onset of mesoderm invagination in addition to the rest of this event in wild-type embryos. Our observations revealed dynamic aspects of cell behavior, some of which might be difficult to recognize clearly in low-time resolution analyses. We found that Drosophila mesoderm invagination began with slow synchronous and subsequent fast stochastic phases of cell apex constriction. Furthermore, we observed defective cell behavior of the prospective mesoderm in living fog mutant embryos, directly demonstrating the importance of Fog-mediated signaling in the two phases of cell apex constriction.

cDNA containing the open reading frame and the 3′ untranslated region for DE-cadherin was modified by addition of a BspHI site at the translational start site and of a NcoI site in front of the stop codon, and cloned into the NcoI/XbaI site of pUP2M, which contains the promoter region of the ubiquitin gene (Lee et al., 1988). The resulting plasmid was named pUP-DECH-Nc. A red-shifted mutant GFP cDNA (pQBI25, excitation peak 473 nm; emission peak 509 nm, Quantum Biotechnologies) was inserted into the NcoI site of pUP-DECH-Nc, and then the KpnI-XbaI fragment was transferred to pCaSpeR4-SV40 (Oda and Tsukita, 1999) to produce pCaSpeR-ubi-DE-cad-GFP. Transgenic flies were produced as described previously (Robertson et al., 1988). Several independent homozygous lines were established, and examined with a Zeiss Axiophot II microscope or an Edge direct-view 3-D microscope. In all the lines, cell-cell adherens junctions were clearly visible. Although the selected line of ubi-DE-cad-GFP laid many unfertilized eggs (about 50%) for as yet unknown reasons, most fertilized eggs from the flies developed normally.

For western blotting, anti-DE-cadherin (Oda et al., 1994), anti-α-spectrin (Pesacreta et al., 1989) and anti-GFP (Clontech) antibodies were used at dilutions of 1:50, 1:1000 and 1:1000, respectively.

Canton-S was used as the wild-type Drosophila melanogaster strain. Df(2R)E2, shg^R69 and fog^4a6 were described previously (Uemura et al., 1996; Godt and Tepass, 1998; Costa et al., 1994). The recombinant, Df(2R)E2, ubi-DE-cad-GFP, did not complement Df(2R)E2, Df(2R)D17, insc^P49 or insc^P72 (Uemura et al., 1996; Kraut and Campos-Ortega, 1996), but did complement shg², shg^g317 and shg^E17B in addition to shg^R69 (Uemura et al., 1996; Tepass et al., 1996).

Genomic DNA was isolated from adult flies with the indicated genotypes, and digested with HindIII. Digoxigenin (DIG)-labeled nucleotide probes were generated using a PCR DIG Probe Synthesis Kit (Boehringer Mannheim), and used for hybridization. Signals were detected using a combination of peroxidase-conjugated anti-DIG antibody (Boehringer Mannheim) and the ECL detection system (Amersham Pharmacia Biotech).

ubi-DE-cad-GFP embryos at 2.5 hours of development (25°C) were dechorionated, placed on coverslips in the appropriate orientation, set in place with vacuum silicone grease (Beckman) and mounted in halocarbon oil 700 (Sigma) on special slides with an oxygen-permeable membrane (Yellow Spring Instrument) as described (Girdham and O’Farrell, 1994). The embryos were then viewed under an inverted fluorescence microscope (Carl Zeiss, Axiovert 135) equipped with a highly sensitive SIT video camera (C2741-08, Hamamatsu Photonics) using a 100× NA1.4 Plan-Apochromat oil immersion objective (Carl Zeiss) and a filter set for GFP (Chroma, 41017). Excitation light generated from a 100 W mercury arc fluorescent-light source was attenuated to 1.5-6% using 50, 25 and 12% ND filters. Images were subjected to real-time averaging by Argus 20 (FRM=32, Hamamatsu Photonics), and recorded every second on a laser disc using a laser videodisc recorder (LVR-3000AN, Sony). During the 30-60 minute period of observation of embryos at 23±1°C, the focus was manually adjusted to the level of the embryo surface. The observed embryos were further incubated to check that they developed normally to hatching. Some images were transferred to a Power Macintosh, and processed using Adobe Potoshop 4.0J, NIH image 1.61 and Adobe Premiere 4.2.1J to produce the figures and the QuickTime movies. Measurement of area was performed using Adobe Photoshop 4.0J.

To observe fog mutant embryos, a confocal microscope (Bio-rad, MRC1024) was used. Images were taken at 10 second intervals with 3% of the maximum laser power. This condition did not affect gastrulation and germband extension in wild-type embryos. Embryos from fog^4a6/FM7 and Y/FM7; ubi-DE-cad-GFP were randomly taken and observed. During and after the observations, we ascertained whether the individual embryos exhibited characteristic fog phenotypes as described previously (Costa et al., 1994).

To visualize the adherens junctions in living Drosophila embryos, we constructed a chimeric gene encoding DE-cadherin tagged at its C terminus with green fluorescent protein (DE-cadherin-GFP), and used it to transform flies. The constructed gene, designated as ubi-DE-cad-GFP, contained the ubiquitin promoter (Lee et al., 1988) (Fig. 1A), which allowed ubiquitous expression of the gene product throughout development. We established several independent homozygous lines, each of which had two copies of ubi-DE-cad-GFP in addition to two copies of the endogenous DE-cadherin gene, encoded in the shotgun (shg) locus (Uemura et al., 1996; Tepass et al., 1996), and designated these flies as ubi-DE-cad-GFP. Of these, we selected one line not only bearing the transgene on the second chromosome, where shg is located, but also expressing DE-cadherin-GFP at a level comparable to that of endogenous DE-cadherin (Fig. 1B), and used it throughout this study. When living ubi-DE-cad-GFP embryos were examined by fluorescence microscopy, cell-cell adherens junctions, which form three-dimensional networks linking apical edges of epithelial cells, were clearly visible in blastoderm cells and later epithelial cells (Fig. 2).

We next examined whether ubi-DE-cad-GFP can function in development. Zygotic mutations in shg lead to severe defects in morphogenetically active epithelia of the embryo (Uemura et al., 1996; Tepass et al., 1996). To perform rescue assays of shg mutants with ubi-DE-cad-GFP, we generated a chromosome containing ubi-DE-cad-GFP and a deletion, Df(2R)E2, which removes at least shg and a nearby essential gene, inscuteable (insc) (Uemura et al., 1996; Kraut and Campos-Ortega, 1996). Flies with this chromosome were crossed with those carrying a shg null allele, shg^R69 (Godt and Tepass, 1998), to produce embryos with the zygotic genotype of Df(2R)E2, ubi-DE-cad-GFP/shg^R69. These embryos developed into morphologically normal adult flies, the genotype of which was confirmed by Southern blotting (Fig. 1C). In addition, the progeny of Df(2R)E2, ubi-DE-cad-GFP/shg^R69 flies were maintained for more than 5 generations, and homozygotes for shg^R69, ubi-DE-cad-GFP developed normally. These results indicated that ubi-DE-cad-GFP complemented the shg null allele, and that DE-cadherin-GFP can function normally even in the complete absence of intact DE-cadherin. Also, it was suggested that the behavior of DE-cadherin-GFP, visible as fluorescence in the living flies, is the same as that of intact DE-cadherin. ubi-DE-cad-GFP, therefore, could be an ideal tool not only for visualizing the adherens junctions but also for analyzing the dynamics of the functional cadherin molecule.

To follow the changes in shape and position of the prospective mesodermal cells before and during gastrulation, we examined living ubi-DE-cad-GFP embryos. A highly sensitive video camera enabled us to detect sufficient levels of GFP signals continuously with excitation light largely attenuated by ND filters. Exposure of early embryos to the excitation light for 30-60 minutes did not affect gastrulation or later development. Blastoderm embryos were mounted on slides so that their ventral side, corresponding to the prospective mesoderm region, faced the objective lens (Fig. 3A). Since the mounted embryos were slightly flattened by compression, a larger part of the embryo surface was set in focus than in intact embryos. This deformation did not essentially affect cell behaviors in the developing embryos. The observed embryos on the slides developed normally to hatching, suggesting that they had undergone no serious damage while being observed. Images were recorded every second from before the onset of gastrulation to the start of germband extension, at 23±1°C. We obtained four sets of sequential images from different embryos, and found that they showed consistent results with only small variations. Of the four, a representative set of images is shown in Fig. 3B. We traced back the cells positioned at the site of ventral closure, corresponding to the mesectodermal cells, and determined which cells were destined to form the mesoderm. The prospective mesodermal cells, positioned more ventrally than the mesectodermal cells, showed dynamic changes in shape and position. Onset of movement in the prospective mesoderm was detected around a time point corresponding to the third panel of Fig. 3B in a quantitative analysis described below (Fig. 4), and this time point was therefore designated time 0.

The process of mesoderm invagination was roughly divided into the following four stages. First, the majority of cells in the centralmost (or midventral) region, corresponding to about two thirds of the total mesoderm, constricted their apices, while the apices of more lateral mesodermal cells, designated as the lateral mesodermal cells, were elongated along the mediolateral axis (Fig. 3B, 0-7 minutes). DE-cadherin-GFP was strongly concentrated at the reduced apical zones of cell-cell contact. This constriction stage was subdivided into slow and fast phases on the basis of the analysis described below. Second, the midventral cells gradually moved apart from the tangential plane to form a shallow furrow, called the ventral furrow, along the anterior-posterior axis (Fig. 3B, 7-14 minutes). Many of the lateral mesodermal cells were still in the exterior, while their apices were further elongated. Third, the lateral mesodermal cells rapidly migrated toward the ventral midline, and then entered the interior of the embryo, presumably pushing the midventral cells inward (Fig. 3B, 14-20 minutes). Finally, the prospective mesectodermal cells coming from both sides met at the ventral midline, and maximized their contacts, where DE-cadherin-GFP was accumulated (Fig. 3B, 20-25 minutes). This sealing of the ventral midline by the mesectodermal cells was accompanied by the start of germband extension.

Most published images of fixed embryos at various gastrulation stages could be correlated with the images obtained here. Our observations, therefore, provide an example of time scales for cell shape changes and cell movements during mesoderm invagination, although the timing of the cellular behaviors varied slightly between embryos. The stages defined above may be comparable to those defined previously using a combination of epi-illumination video microscopy and scanning electron microscopy (Sweeton et al., 1991).

The prospective mesodermal cells in the cellular blastoderm were essentially uniform in shape and area, with only small variations. We found no cells that increased or decreased their apical surface areas at a significant rate in an irreversible manner before the onset of mesoderm invagination. After only 5 minutes, however, the apical surfaces of the mesodermal cells exhibited large differences in shape and area, as they began to change their shapes (Fig. 3B). To examine the relationship between the position of the mesodermal cells relative to the medio-lateral axis and the timing of their shape changes, we identified four adjoining groups each containing ten cells in mediolateral order as shown in Fig. 4A, and followed the temporal changes in the apical surface area of each group (Fig. 4B). The most ventral group, Group 1, began to show a reduction of the apical area earlier than all the other groups. The adjacent, more lateral group, Group 2, showed a rapid reduction in apical area with a delay of about 2.5 minutes. In contrast to Groups 1 and 2, Group 3 showed a significant increase in apical area followed by a rapid reduction in the area. The most lateral group, Group 4, did not decrease but increased its apical area from the onset of apical constriction, eventually reaching a plateau. These observations indicated that apical constrictions start first in the center of the mesoderm followed by lateral propagation of apical constrictions, while the most lateral cells (about 3 cells in width) are forced to enlarge their apical surfaces.

The behavior of the cells around the ventral midline, including those in Group 1, involved two phases of apical constriction. During the first phase, lasting for about 2 minutes, the apical area of Group 1 decreased slowly with only slight differences between the individual cells (Fig. 4A,B). Most cells in the central region reduced their apical areas. No cells were found that showed rapid constriction of their apices compared to their surrounding cells. During the second phase, the apical area of Group 1 decreased about 2.5-fold faster than in the preceding phase, and the differences between the cells became progressively apparent (Fig. 4B,C). Many, but not all, cells around the ventral midline showed accelerated constriction of their apices. Many rapidly constricting cells were also found in Group 2 and in similar positions, with fewer in positions corresponding to Group 3 during the second phase (Fig. 4A). In contrast to the rapidly constricting cells, the remaining cells failed to accelerate their constriction, and some were slightly widened, coinciding with constriction of nearby cells (Fig. 4C), although they soon began again to reduce their apical areas. In the middle of the second phase, cells with small and others with relatively large apices were randomly arranged, and the individual embryos showed different patterns. In all the embryos observed, cell behaviors characteristic of both the slow and fast phases were observed. These observations suggested that the onset of mesoderm invagination was followed by slow synchronous and subsequent fast stochastic phases of cell apex constriction.

Some previous observations of living and fixed embryos suggested that mesoderm invagination begins with random initiation of apical constriction of individual cells (Sweeton et al., 1991; Kam et al., 1991). However, we found no cells showing prominently distinct behaviors from those of surrounding cells just after the onset of mesoderm invagination. In the initial 2 minutes, the cell shape changes were slight, and might be difficult to recognize in fixed embryo preparations or in low-time resolution analyses. Our observations suggested that the random apical constrictions observed previously (Sweeton et al., 1991) corresponded to the second phase of apical constriction defined in this study. It was previously shown that basal migration of the nuclei of the mesodermal cells also occurs in a random manner at the beginning of mesoderm invagination (Kam et al., 1991), but the relationship between apical constriction and nuclear migration was not analyzed in this study.

We introduced ubi-DE-cad-GFP into fog mutant embryos, in which ventral furrow formation was previously shown to be disordered (Costa et al., 1994), and observed the behavior of the prospective mesodermal cells in the living mutant embryos. We took images at 10 second intervals using time-lapse confocal microscopy, and present some of the images obtained from a fog mutant embryo together with those from a normal embryo in Fig. 5. In fog mutant embryos, the onset of mesoderm invagination was recognized when some cells randomly positioned in a central 8-cell-wide region corresponding to Groups 1 and 2 began to constrict their apices very slowly (Fig. 5A, arrows). The apices of many other cells in the central region remained large with no initial constriction (Fig. 5A, asterisks). The temporal changes in the apical areas of the cell groups were measured as was done for the normal embryo (Figs 4D, 5A). Even when the apical areas of the central mesodermal cells became varied, only slight decreases in the apical areas of Groups 1 and 2 were detected. Several minutes after the onset of mesoderm invagination, a random arrangement of cells with apices of various sizes became more apparent (Fig. 5A, 4-7 minutes). This mode of cell shape change was in contrast to that seen in wild-type embryos, in which stochastic acceleration of constriction of cell apices took place. Although the numbers of cells with relatively small apices slowly increased, the cell shape changes were reversible in fog mutant embryos. Some of the constricted cells enlarged their apices again later (Fig. 5, arrows). In addition, the apices of the lateral mesodermal cells were little enlarged in at least the first 6 minutes (Figs 4D, 5A), indicating that the total contraction force generated in the central mesodermal cells may be markedly weak. Although the initiation of folding of the mesodermal cell sheet was delayed in fog mutant embryos, when it did suddenly take place the apices of most central mesodermal cells were not fully constricted. During the folding, some cells around the ventral midline were forced to expand their apices (Fig. 5, asterisks). Thus, the prospective mesoderm of fog mutant embryos lacked the normal two constriction phases seen in wild-type embryos, and instead exhibited weak, asynchronous cell shape changes.

Fog is a secreted protein that may function as a signal to induce apical constriction of neighboring cells (Costa et al., 1994; Morize et al., 1998). Our high-temporal resolution observations of wild-type and fog mutant embryos support the idea that Fog-mediated signaling enables synchronous initiation of the contraction forces exerted in central populations of the prospective mesodermal cells and rapid propagation of this initiation of contraction to more lateral cells. Without Fog activity, some prospective mesodermal cells did initiate apical constrictions, although their cell shape changes were feeble. The constrictions in fog mutants were likely to be caused by relatively weak contraction forces continuously or discontinuously generated in the cells in a rather independent manner. In fog mutant embryos, failure to minimize the apical areas of the central mesodermal cells resulted in an abnormal mode of folding of the mesodermal sheet.

Some previous studies stressed independent initiation of apical constriction in the individual cells (Sweeton et al., 1991; Kam et al., 1991). However, our observations suggest the importance of a Fog-mediated mechanism by which the central mesodermal cells synchronously initiate contraction forces to drive apical constriction at the onset of invagination. These contraction forces may give rise to high levels of tension among the central mesodermal cells. The apical area of each cell may depend on the balance between the force generated in the cell and that pulled from the surrounding cells. The irregular changes in cell shape observed in the second phase of apical constriction may result from imbalance of the elevated contraction forces among the individual cells rather than from random initiations of apical constriction. In other words, the synchronous activation of the contractile cortex by Fog-mediated signaling may cause competition among the individual mesodermal cells for apical constriction. During the second phase, lateral progression of initiation of apical constriction was detected by quantitative analysis (Fig. 4B). The changes in the apical areas of the grouped cells partially mimicked the situation proposed in the ‘purse-string’ model, in which a wave of contractions propagating from cell to cell is postulated (Odell et al., 1981).

This study enabled real-time imaging of cell-cell adherens junctions by making flies with ubiquitous expression of GFP-tagged DE-cadherin. We used these flies to examine dynamic cell behavior in Drosophila mesoderm invagination in wild-type and mutant embryos. The new tools are easily applicable to phenotypic analysis of other Drosophila mutants that have defects in epithelial cell polarity and morphogenesis. This kind of analysis may allow one to detect the earliest defects in such mutants, contributing to a better understanding of the molecular and cellular mechanisms controlled by the genes.

We thank D. Branton, J. Lee, U. Tepass and E. Wieschaus for providing the antibody, plasmids and flies, V. Verkhusha for discussion, and Y. Akiyama-Oda for critical reading of this manuscript. We are also grateful to S. Okajima and M. Okubo for technical assistance.