Fibroblast growth factor signalling controls successive cell behaviours during mesoderm layer formation in Drosophila

The development of tissues and organs depends on coordinated cell movements. Some tissues migrate as epithelial structures, maintaining apical-basal polarity, whereas others undergo a transition to a mesenchymal state (EMT) prior to migration (Biname et al., 2010). Mesenchymal cells migrate individually or as a collective, in which cells can move relative to each other while engaging and disengaging cell contacts (Weijer, 2009). It is not fully understood which cell behaviours contribute to collective migration and how intrinsic and extrinsic signals coordinate collective cell migration.

The formation of the mesoderm in Drosophila is a well-studied example of collective cell movement and we have extensive knowledge of its genetic regulation (Leptin, 1999). The mesoderm primordium invaginates and is internalised as an epithelial monolayer (Costa et al., 1993; Leptin and Grunewald, 1990). The cells then undergo EMT and establish a multilayered aggregate apposed to the basal surface of the neuroectoderm epithelium. This multi-layered mesoderm aggregate spreads out between the neuroectoderm and the centrally located yolk cell and eventually forms a transient monolayer (Fig. 1A,B). Mesoderm layer formation depends on signalling by the FGF8-like ligands Thisbe (Ths) and Pyramus (Pyr) through the FGF receptor Heartless (Htl) (Beiman et al., 1996; Gryzik and Müller, 2004). Htl is expressed in the mesoderm cells while the ligands exhibit a dynamic expression pattern in the neuroectoderm (Klingseisen et al., 2009; Shishido et al., 1993; Stathopoulos et al., 2004).

FGF signalling is also implicated in mesoderm formation in vertebrate gastrulation. In mouse embryos, FGF signalling is required for EMT and ingression of epiblast cells to form the mesoderm layer (Ciruna and Rossant, 2001; Sun et al., 1999). One mechanism by which FGF signalling initiates EMT is through maintaining expression of Snail-family transcription factors, which downregulate epithelial gene networks (Nieto et al., 1994). In chick gastrulation, FGFs play an important role in directing the collective migration of the mesoderm cells after ingression, whereas their function in EMT in the primitive streak is less clear (Chuai and Weijer, 2009a).

During Drosophila mesoderm layer formation, cells at the dorsal edge of the mesoderm aggregate extend protrusions pointing in the direction of their migration (Schumacher et al., 2004). These protrusions depend on localised expression of Pyr, but not Ths, indicating that Pyr in the dorsal ectoderm might provide directional information (Klingseisen et al., 2009). The overall dynamics of cell movement during mesoderm spreading in Drosophila have recently been analysed by live imaging (McMahon et al., 2008; Murray and Saint, 2007). Cell movements were deduced from determining the positional fate of selectively labelled groups of cells using photo-activatable GFP (Murray and Saint, 2007). These analyses suggested that cells in contact with the ectoderm move dorsally during spreading. Presumably, the more internally located cells either intercalate radially or overtake cells at the dorsal edge. A second study tracked mesoderm cell nuclei labelled with GFP (McMahon et al., 2008). The authors suggested that the mesoderm spreads by synchronous waves of cell intercalation and that a stable leading cell population migrates directionally. These studies have provided important insight into the global movement of the cell collective. However, the underlying behaviour of individual cells that is responsible for the movement cannot be deduced from these studies. To understand the basis for the cell movements it is necessary to study this behaviour directly and determine how it is controlled by the FGF receptor and its ligands. We have therefore set up an experimental system that allows us to follow cell shape changes in live-imaging experiments. We find that mesoderm cells change their behaviour several times, and that these successive cell behaviours have differential requirements for the two FGF ligands and possibly differential downstream signalling events.

Fly stocks were maintained under standard conditions. The following mutant alleles and transgenes used in this study have been described previously: mys¹ (Wright, 1960), mys¹¹ (Wieschaus et al., 1984), stg^7M53 (Jürgens et al., 1984), htl^AB42 (Gisselbrecht et al., 1996), dof¹ (Vincent et al., 1998), twi::CD2 (Dunin-Borkowski and Brown, 1995), UAS::Cdc42[N17] (Luo et al., 1994), Cdc42³, Cdc42⁶ (Genova et al., 2000) and shg² (Tepass et al., 1996).

Embryos were collected, fixed and stained using standard methods (Müller, 2008). Fixation was 40 minutes in 3.7% formaldehyde/PBS following manual devitellinisation. The following primary antibodies were used: rabbit-anti-Twist (1:1000), mouse-anti-Neurotactin [Developmental Studies Hybridoma Bank (DSHB); 1:50], mouse-anti-CD2 (1:500; Serotec), rat-anti-E-cadherin (1:50; DSHB) and rabbit anti-β-galactosidase (1:5000, Cappel). Secondary antibodies were conjugated to Alexa488, Alexa647 (Molecular Probes) or Cy3 (Jackson), or enzyme conjugated. Images were collected using a Leica SP2 laser scanning confocal microscope and an Olympus BX61 wide-field microscope. Homozygous recessive mutants were recognised by the absence of transgenic β-galactosidase on balancer chromosomes. Histological cross-sections were prepared as described elsewhere (Klingseisen et al., 2009). For staging, the extent of mesoderm spreading in wild type relative to the extent of germ band elongation was used. The distance from the posterior of the embryo to the posterior end of the mesoderm was measured and divided by the total embryo length. Embryos giving values of 44-53% of embryo length were considered to be in the tube-flattening phase, whereas embryos giving values from 56-59% were considered to be in the dorsal migration phase. For determination of protrusion numbers (Figs 6, 7), all protrusions were measured in a defined length of the anterior part of the germ band excluding the area where the germ band begins to curve dorsally. Positions of the tip of each protrusion and the centre of the nearest mesoderm nucleus were determined using Volocity (Improvision), with reference to images sectioned in three different planes (XY, XZ, YZ) to incorporate three-dimensional information. The distance between these positions was calculated and protrusions less than 5 μm in length were excluded. Protrusion numbers are normalised per 200 μm distance along the anteroposterior axis of the embryo.

For measurement of relative intensities of E-cadherin staining, sections were selected including areas of ectoderm and mesoderm tissue (determined by twi::CD2 staining). After measuring and subtracting the mean background staining intensity, regions of interest were drawn covering the tissues to be measured. Pixels representing the cell surfaces were selected by thresholding and mean intensities measured. Paired t-tests were employed to compare data from the different tissues within the same measured embryos. All measurements were carried out using ImageJ (NIH).

The twi::GFP-actin5C was generated by cloning 3 kb genomic DNA upstream of the transcriptional start site of the twist gene using PCR. The EGFP-actin5C fusion was generated using pEGFP (Clontech). Both fragments were cloned into pCaSpeR 4 and transgenic lines were generated using standard protocols.

Embryos expressing twi::GFP-actin5C were prepared for imaging by a method published elsewhere (Davis, 2000). Embryos were collected, aged to stage 5, manually dechorionated, washed and oriented on agarose. They were stuck onto a cover slip coated with heptane glue, dried briefly over silica gel and covered with halocarbon 700 oil (Sigma). Images were collected using a 40× 1.3 NA oil immersion objective on an inverted Nikon Eclipse TE2000-U microscope attached to a Biorad Radiance 2100 multiphoton scanning system. A chameleon Titanium Sapphire femto-second laser (Coherent) was tuned to 895 nm to excite GFP. Signal was collected using a non-descanned light path with no emission filter to maximise collection of emitted light. The scan head was rotated so that the vertical (y) dimension of the captured images represents the anteroposterior axis of the embryo. In some experiments, the stage was moved along this dimension to maintain the same field of view as cells move posteriorly during germ band elongation. Scanning was unidirectional at a speed of 166 lines per second with no averaging. Laser power was adjusted to give an adequate signal-to-noise ratio while allowing normal spreading and cell division behaviour. Embryos that have been imaged under our conditions survive to hatch as viable larvae.

In 4D datasets, individual nuclei were tracked manually in order to follow cell movements for both the generation of movies and measurements of migration rates. Before tracking, the image stacks were processed in ImageJ (NIH) using a macro that creates stacks from projections of the original data along any of the original image axes. Using projected data in this way improves the signal-to-background ratio and allows more structures from large cells to be included (e.g. the nucleus and a cellular protrusion). At each time point, the position of the centre of each nucleus was determined manually with reference to projected data in the original (xy) orientation and also the two re-sliced orientations xz and xy, approximately representing transverse and saggital sections, respectively. Projections used for tracking and represented in the movies and figures were made by averaging four original image slices. Nuclear positions were recorded using the MtrackJ plug-in for ImageJ (Erik Meijering). Three-dimensional rendered images were created using Volocity (Improvision). For 3D reconstructions of two-photon data from live embryos, the autofluorescence signal from the vitelline membrane was removed by creating a binary image mask from the original data based on a defined threshold value. Masks were edited to remove any pixels representing signal internal to the embryo then dilated, to ensure that the entire signal from the membrane was covered. Masks were then blurred using a Gaussian filter and subtracted from the original image data before importing into Volocity and rendering.

The mesoderm primordium invaginates by the ventral furrow forming an epithelial tube within the interior of the embryo. Reminiscent of a classic EMT, the entire tube then loses its epithelial structure and the cells adopt a mesenchymal morphology. However, unlike classic EMT, the mesoderm cells maintain an intimate contact with each other while they spread out to form a transient monolayer (Fig. 1A,B). The dynamic cell shape changes that occur during these tissue rearrangements have not been determined because of the lack of suitable imaging methods.

To monitor the dynamics of mesoderm cell morphology we used two-photon microscopy to image expression of a GFP-Actin5C fusion gene driven by the mesoderm-specific twist promoter. We are able to visualise the most superficial mesoderm cells, as well as a short distance into the cell layer lying beneath the superficial layer. GFP-Actin is abundant in cellular protrusions, allowing cell shape dynamics to be observed. Nuclei appear as dark areas, which can be used to track individual cells. To visualise different stages of the spreading movements, we adopt different protocols, varying the direction from which images are captured as well as the zoom and depth of imaging (see Fig. S1 in the supplementary material). We observe mitoses in time-lapse experiments when the reporter protein enters the nucleus upon breakdown of the nuclear envelope. Mesoderm cells divide in three waves of nearly synchronous mitoses during gastrulation, the first occurring shortly after invagination and the second at approximately the time when a cell monolayer is established, at stage 9 of embryonic development (Fig. 1A). The third division occurs later in stage 10 and coincides with segmentation of the mesoderm and its adopting a multilayered structure (Campos-Ortega and Hartenstein, 1997). We record images between the first two mitoses, allowing observation of successive changes of cell behaviour that contribute to monolayer formation.

Following invagination and EMT, mesoderm cells divide and the tissue flattens rapidly against the basal ectoderm surface. The first cellular activity of mesoderm flattening is the formation of contacts of the basal side of the mesoderm tube with the ectoderm (Fig. 1B) (Schumacher et al., 2004). It is not known whether flattening is a passive process caused by differential adhesion or whether mesoderm cells actively move towards the ectoderm.

Mesoderm-ectoderm (ME) attachment is associated with actin-rich protrusions from the mesoderm cells directed towards the ectoderm (Fig. 2). The initial region of ME contacts at the base of the tube expands during and after the period when the cells undergo EMT and divide (Fig. 1A,B). The more lateral mesoderm cells attach progressively to the ectoderm and protrude radially in between adjacent ectoderm cells. The flattening movement can thus be described as a ‘zippering-up’ of the mesoderm cell aggregate onto the ectoderm (Fig. 2A,B; see Movies 4, 5 in the supplementary material). Mesoderm cell protrusions intercalate between the basolateral surfaces of the ectoderm and are visible as concentrations of GFP-Actin at the interface between the germ layers (Fig. 2A; see Movie 4 in the supplementary material). These cell shape changes are associated with cell movement. Cells located at the dorsolateral edges of the mesoderm (dorsal edge cells, DECs) move radially outwards from central towards more peripheral positions (Fig. 2B; see Movie 5 in the supplementary material). Large radial protrusions (19.0 μm±2.7 (s.d.; n=7 from three embryos) are seen in these cells and also at more ventral positions (Fig. 2C; see Movie 6 in the supplementary material). We conclude that the major initial movement of mesoderm cells is radial and is an active process, probably driven by actin-rich protrusions extending from the mesoderm cells.

The discovery that the Drosophila mesoderm forms an intimate connection with the ectoderm following internalisation is unexpected. During gastrulation in vertebrates, cell signalling events trigger repulsion between mesoderm and ectoderm, and an extracellular matrix is established at the germ layer interface (Chuai and Weijer, 2009b; Winklbauer et al., 2001).

It has been proposed that integrin-mediated contacts are responsible for high-affinity ME interactions in Drosophila (McMahon et al., 2010; Murray and Saint, 2007). β1-myospheroid (mys) integrin complexes become concentrated at the ME interface in the late stages of spreading (Leptin et al., 1989; McMahon et al., 2010). It has recently been reported that embryos maternally and zygotically (M/Z) mutant for the mys¹ allele exhibit defects in monolayer formation (McMahon et al., 2010). mys¹ M/Z mutant embryos show abnormal monolayer formation, reduced dorsal mesoderm differentiation and fewer DEC protrusions (Fig. 3B,D,E). The molecular lesion in mys¹ is described as a chromosomal deletion, but the extent of this chromosomal rearrangement has not been determined (Bunch and Brower, 1992). As the molecular lesion of the mys¹ allele is not unambiguous, we repeated these experiments with the protein null allele mys¹¹ (Bunch and Brower, 1992; Jannuzi et al., 2002; Leptin et al., 1989). mys¹¹ M/Z mutants exhibit normal mesoderm migratory morphology, normal monolayer formation and dorsal mesoderm specification (Fig. 3A,C,E). These data indicate that the mys¹ allele is associated with other mutations on the chromosome that alone or in combination produce mesoderm-spreading defects. We conclude that β1-integrin function is not required for mesoderm spreading.

As integrin-mediated adhesion is dispensable for ME attachment we reasoned that direct cell-cell contacts may be responsible. It has been suggested that ME separation represents a classic example of tissue separation via differential expression of cadherins (Oda and Tsukita, 1999). Consistent with this model is the observation that E-cadherin transcription is repressed by Snail and N-cadherin transcription is upregulated by Twist (Oda and Tsukita, 1999). However, N-cadherin protein accumulates in the mesoderm only at later stages during monolayer formation (Oda and Tsukita, 1999). We find that throughout mesoderm spreading, the maternally provided E-cadherin remains present in mesoderm cells (Fig. 3F-I). Measurements of the fluorescence intensity of E-cadherin immunolabeling revealed that the intensity at ectoderm/ectoderm cell contacts was not significantly different from that of mesoderm/mesoderm contacts [P=0.39 (n=5); paired t-test]. By contrast, the level of E-cadherin at the mesoderm/ectoderm interface is elevated, in comparison with cell-cell contacts within either the mesoderm [P=0.00088 (n=5); paired t-test] or the basolateral domain of ectoderm cells [P=0.00067 (n=5); paired t-test] (Fig. 3F,G). E-cadherin is also accumulated at sites where mesodermal protrusions penetrate between ectoderm cells (Fig. 3H,I; see Movie 7 in the supplementary material). The high E-cadherin staining intensity at these sites is not simply a consequence of the closely apposed membranes of the protrusion, as the raised level of E-cadherin is also seen at protrusions where the membranes can be distinguished from each other (Fig. 3H). A similar accumulation it is not seen for other membrane proteins, such as neurotactin (not shown). Consistent with a role of E-cadherin in mesoderm layer formation, embryos mutant for the E-cadherin loss-of-function allele shg² (shotgun) exhibit various defects in mesoderm morphogenesis and dorsal mesoderm differentiation (see Fig. S2 in the supplementary material). These observations indicate that E-cadherin is recruited to ectoderm/mesoderm contacts and suggest that the intercalation of mesoderm cells into the ectoderm involves direct cell contacts mediated by E-cadherin.

In the first phase of mesoderm spreading, EMT and mesoderm flattening increase the number of cells in contact with the ectoderm; however, not all mesoderm cells have made contact with the ectoderm at the end of this phase. Various mechanisms have been proposed to account for the mesoderm layer formation (Fig. 1C-E) (Kadam et al., 2009; Klingseisen et al., 2009; McMahon et al., 2008; Murray and Saint, 2007; Wilson et al., 2005; Winklbauer and Müller, 2011). We analysed the behaviour of DECs of the mesoderm as we have shown before that protrusions at the dorsal edge are sensitive to the availability of FGF8-like ligands.

Live observations of DECs reveal that their movement follows a stereotypic biphasic pattern. During flattening, DECs initially protrude radially with little net dorsal movement. During a 15-20 minute period after the first mitotic wave, all DECs (n=12; three embryos) extend protrusions in a radial direction between ectoderm cells (Fig. 2A-C; Fig. 4A). As the cells move radially outwards, these radial protrusions become shorter and eventually disappear (Fig. 4A; see Movie 8 in the supplementary material). DECs then repolarise and switch from radial movement to a persistent dorsal migration during the following 30-40 minutes. Thereafter, they undergo the second wave of mitotic divisions. Tracking nuclear positions of DECs over time confirms a biphasic migration of DECs, first in a radial direction and then in a dorsal direction (Fig. 4D).

Repolarisation is associated with remodelling of the protrusive activity of DECs. Half of the DECs we imaged (6/13; three embryos) formed new protrusions, oriented in the direction of spreading, as they began their dorsal migration (Fig. 4A,B; see Movie 8 in the supplementary material). Imaging DECs at higher resolution reveals a polarised morphology that is indicative of directionally migrating cells (Fig. 4B; see Movie 9A in the supplementary material). DECs extend and polarise along the direction of migration and form lamellipodia and filopodia. The distance from the front edge of the nucleus to the tip of the lamellipodium can extend up to 17 μm, whereas the cell body diameter (distance between the centres of adjacent nuclei) is 7.1 μm (±1.1; n=6). Filopodia are believed to contribute a sensory function during migration by sampling the environment for directional cues (Millard and Martin, 2008). We observe dynamic filopodia extending from the leading edge cells (Fig. 4C; see Movie 9B in the supplementary material). The polarised morphology of DECs and the presence of filopodia at their leading edge suggest that these cells migrate directionally towards a dorsal attractant.

Although DECs retract their radial protrusions as they repolarise and migrate dorsally, the more ventrally located cells maintain radial protrusive activity throughout the period of dorsal spreading. Large, dynamic protrusions extend deep into the ectoderm, disappearing only as cells round up at the next wave of mitosis as the monolayer forms (Fig. 4E; see Movie 10 in the supplementary material). These results demonstrate continuous intercalation behaviour of mesoderm cells and a dynamic interaction with the basal-lateral domain of the ectoderm throughout the spreading process. The dorsal movement of DECs also reduces space constraints for inner cells and might thereby promote their movement to the ectoderm surface (Fig. 1C). Indeed, tracking of mesoderm nuclei in live embryos has suggested that the majority of inner mesoderm cells reach the surface by this route (McMahon et al., 2008), and we observe this cell behaviour directly in our time-lapse experiments (see Movie 11 in the supplementary material). These results demonstrate that, after flattening, mesoderm cells maximise their contact to the ectoderm surface by two major movements: directional migration of dorsal edge cells and radial intercalation movements of inner cells.

Mesoderm layer formation requires FGF signalling, and we have shown previously that ME attachment prior to EMT depends on Htl (Schumacher et al., 2004), Having identified several successive cellular behaviours during mesoderm layer formation, we wished to determine how these behaviours are affected by FGF signalling.

As FGF is known to be important for EMT in vertebrates, we first analysed the role of FGF in the transition from the epithelial invaginated mesoderm to a mesenchymal state. During formation of the blastoderm epithelium, E-cadherin accumulates at sub-apical adherens junctions. In the mesoderm, the early adherens junctions are disassembled and re-established at the extreme apical site (Kölsch et al., 2007), where they are maintained until after formation of the mesoderm tube (Fig. 5). As EMT occurs, E-cadherin loses its apical concentration and redistributes over the mesoderm cell surfaces in a non-polarised fashion (Fig. 5). At the same time, cells begin to divide, but the loss of cell contact is not exclusively due to cells rounding up in mitosis, as it is also seen in embryos mutant for string (stg), in which post-blastoderm mitoses are blocked (Fig. 5A,B) (Leptin and Grunewald, 1990). Thus, EMT can occur independently of mitotic divisions, implying that other processes remodel cell polarity and epithelial junctions.

One possibility is that FGF signalling contributes to EMT. It is not possible to test this simply by disrupting FGF signalling, as mitosis still occurs and cells therefore lose their contacts. To test for a possible role for FGF signalling in EMT it is necessary to interfere with FGF signalling in a situation where mitosis does not occur. We thus created embryos that were simultaneously mutant for stg and the FGF receptor Htl or the FGF signal transducer Dof. In htl stg or dof stg double mutants, mesoderm cells maintain their epithelial state and show persistent polarised E-cadherin distribution and maintenance of junctions after invagination (Fig. 5A,B; data not shown). The block of EMT is transient, but results in strongly delayed spreading and monolayer formation (Wilson et al., 2005) (see Fig. S3 in the supplementary material). This result implies a role for FGF signalling in the remodelling of polarised E-cadherin adhesion during the onset of EMT in the mesoderm.

ME attachment and mesoderm flattening are impaired in the absence of FGF signalling (Schumacher et al., 2004; Wilson et al., 2005). Embryos lacking Htl develop fewer and shorter mesodermal protrusions (see Fig. S3 and Movie 12 in the supplementary material), and there is little radial cell movement (data not shown). The absence of early pre-EMT protrusions at the base of the tube in htl and dof mutant embryos suggests functions of FGF signalling other than the control of EMT (Wilson et al., 2005). To determine the role of the Htl ligands Pyr and Ths for protrusive activity of mesoderm cells after EMT, we studied mutants lacking either one of the ligands. The number of protrusions was determined in tightly staged embryos, fixed during mesoderm flattening or during dorsal migration of DECs (Fig. 6A-D).

Although embryos lacking ths function have the normal number of protrusions during flattening, pyr mutant embryos show a reduction in protrusions, both in the dorsal and in ventrolateral cells (Fig. 6C). The phenotype is not as severe as in embryos lacking Htl, which suggests redundancy between Pyr and Ths functions. We conclude that FGF signalling has independent functions in the transition to a mesenchymal state and the formation of protrusions associated with the orderly flattening of the mesoderm onto the ectoderm surface.

DEC morphology during later mesoderm spreading suggested that after flattening, cells migrate dorsally in response to directional cues. An obvious candidate for a dorsal attractant is Pyr. Its expression becomes largely restricted to a dorsal domain of the ectoderm when dorsal spreading occurs (Gryzik and Müller, 2004; Stathopoulos et al., 2004) and it is necessary for the formation of DEC protrusions (Fig. 6B,D). We asked whether reduced numbers of dorsal protrusions in pyr¹⁸ mutants are reflected in a difference in the migratory behaviour of DECs. Wild-type DECs move dorsally with a mean rate of 52.7±3.0 μm/h (n=13; three embryos). This is comparable with the migration speeds of other Drosophila cells that move in collectives, e.g. ovarian border cells [21-60 μm/h (Bianco et al., 2007; Prasad and Montell, 2007; Tekotte et al., 2007)] and male-specific SGPs [(12.3 μm/h) (Clark et al., 2006)]. The mean migration rates of DECs were significantly lower in pyr¹⁸ homozygous mutants (21.4±4.1 μm/h; n=9; two embryos), showing that Pyr is necessary for the directional dorsal migration of DECs (Fig. 6F; see Movie 13 in the supplementary material). These experiments demonstrate that Pyr represents a dorsal cue providing a significant influence on the direction and the speed of dorsal mesoderm cells.

Throughout the dorsal migration phase, the ventral mesoderm cells exhibit a dynamic radial protrusive activity (Fig. 4E). We found that Ths, and to a lesser extent also Pyr, are required for the normal number of these radial protrusions during dorsal migration (Fig. 6D). However, in neither of the single mutants are radial protrusions completely abolished, suggesting that Ths and Pyr contribute together to this activity.

To explore the functional importance of the radial protrusions for mesoderm spreading, we sought to identify a way to interfere selectively with their formation. As filopodial protrusions often depend on the activity of the small Rho GTPase Cdc42, we examined protrusion formation in embryos in which we had interfered with Cdc42 function by expressing a dominant-negative construct, Cdc42[N17], in the mesoderm. Formation of radial protrusions was strongly compromised upon expression of Cdc42[N17], whereas the number of dorsal protrusions did not change significantly (Fig. 7A). These data provide evidence that dorsal and radial protrusions are controlled by distinct molecular pathways. Consistent with a role for Cdc42 in mesoderm morphogenesis, embryos in which Cdc42 activity is reduced exhibit defects in mesoderm layer formation and dorsal mesoderm differentiation (Fig. 7B). To investigate whether the reduction of radial protrusions in mesoderm cells expressing Cdc42[N17] correlates with defects in redistribution of E-cadherin at the germ layer interface, we examined E-cadherin localisation in those embryos (Fig. 7C). In contrast to our findings in the wild type, where E-cadherin is enriched at the ectoderm/mesoderm interface, in embryos expressing Cdc42[N17] we found no difference in E-cadherin immunoreactivity at the mesoderm/ectoderm interface compared with either mesoderm/mesoderm [P>0.1 (n=3); paired t-test] or ectoderm/ectoderm [P>0.05 (n=3); paired t-test] contacts. These data implicate Cdc42 function in the recruitment of E-cadherin to the mesoderm-ectoderm interface.

In vertebrate gastrulation, FGF signalling has been implicated in EMT, ingression and guidance of mesoderm cells away from the primitive streak (Chuai and Weijer, 2009a; Thiery et al., 2009). We have identified successive cellular activities and behaviours during Drosophila mesoderm layer formation. Our results now allow the integration of the previously described dynamic observations of tissue movements with the shape changes of individual mesoderm cells. We find that the two FGF ligands, Pyr and Ths, have multiple, partially overlapping functions in directing this morphogenetic behaviour.

The rapid flattening of the mesoderm onto the ectoderm surface has been attributed to decreased adhesion between mesoderm cells as a result of the mitotic division that follows internalisation (Murray and Saint, 2007). An alternative view proposed that the mesoderm actively contact the ectoderm involving mesoderm cell protrusions (Schumacher et al., 2004). We now find that the mesoderm extends actin-rich protrusions towards the ectoderm as the tissue flattens. The formation of these protrusions depends on FGF signalling, which suggests a role for the FGF pathway in controlling the dynamics of the actin cytoskeleton.

Mesoderm flattening occurs by a zippering motion, with progressive attachments that commence in the most ventral region and propagate to more dorsolateral positions (Fig. 8A). We propose that the region in which protrusions are formed expands dorsally, because flattening of ventral parts of the mesoderm exposes more dorsally located cells to the influence of FGF expressed by the neuroectoderm. This propagation model of mesoderm flattening helps to explain on a cellular level how cells from defined initial positions follow apparently stereotypical paths (McMahon et al., 2008; Murray and Saint, 2007). Such a mechanism would provide for an orderly association of the germ layers, ensuring that mesoderm cells are symmetrically distributed about the ventral midline (Fig. 8A).

It has been proposed that mesoderm spreading may be driven by differential adhesion (Murray and Saint, 2007; Wilson et al., 2005). A propensity for the mesoderm to maximise contact with the ectoderm would follow from mesoderm cells that exhibit a higher affinity for the ectoderm than for each other. Consistent with earlier studies (Oda et al., 1998), we find that although E-cadherin transcription is repressed by Snail in the mesoderm, maternal E-cadherin protein levels do not rapidly decrease upon EMT. During EMT, E-cadherin distributes over the whole cell surface of the mesoderm cells. As contact with the ectoderm is made, E-cadherin accumulates at the germ layer interface, including the sites where mesodermal protrusions penetrate the ectodermal layer. E-cadherin mutants exhibit mild defects in dorsal mesoderm morphogenesis, but we do not yet understand the function of E-cadherin in differential adhesion or in promoting mesoderm-ectoderm attachment and spreading. To address this issue, it will be necessary to establish a system for tissue-specific conditional interference with E-cadherin.

It is also possible that molecules other than E-cadherin mediate adhesion between the ectoderm and the mesoderm germ layer. Although the prime candidate for this mechanism is integrin-mediated adhesion, we find no evidence for a role of integrins in mesoderm spreading, confirming earlier studies (Bunch and Brower, 1992; Jannuzi et al., 2002; Leptin et al., 1989). We therefore favour the model that E-cadherin has a major role in establishing adhesion between the germ layers.

FGF signalling contributes to a switch in the state of E-cadherin leading to the redistribution of polarised E-cadherin during EMT in the invaginated mesoderm cells. A similar switch in E-cadherin function occurs during border cell migration in Drosophila oogenesis (Niewiadomska et al., 1999). Although the cytoplasmic domain of E-cadherin contains a conserved function necessary for cell migration (Pacquelet and Rorth, 2005), it is unclear how the E-cadherin in migrating cell collectives is linked to the cytoskeleton to allow it to transmit the forces required for movement. The cell contacts involved in these movements need not be stable adherens junctions, but are perhaps rather dynamic interactions. Our study identifies Cdc42 as an important determinant of both protrusion formation and E-cadherin accumulation at the ectoderm/mesoderm interface. Further studies will have to address the mechanisms by which Cdc42 is controlled and functions upstream of E-cadherin localisation.

Our study has revealed that dorsal edge cells undergo successive changes in protrusive behaviour (Fig. 8B). The biphasic movement of DECs and repolarisation of protrusive behaviour correlate in timing with the switch in pyr mRNA distribution to a more restricted expression in the dorsal ectoderm (Gryzik and Müller, 2004; Stathopoulos et al., 2004). We show that Pyr is indeed required for the normal migratory behaviour of cells at the dorsal edge. We propose that as DECs migrate dorsally, opportunities are created in their wake for the intercalation of inner mesoderm cells into the monolayer more ventrally (Fig. 8B).

The relevance of the radial protrusive activity for mesoderm spreading is less easy to understand. Pyr and Ths are both required for E-cadherin redistribution and radial protrusion formation during mesoderm flattening. The patterns of the paths derived by tracking all mesodermal cells have hinted at intercalation as an important mechanism of mesoderm layer formation (McMahon et al., 2008). The radial protrusive activity indicates a continuous attraction of mesoderm protrusions towards the ectoderm. We propose that a main function of FGF signalling on a cellular level is to direct protrusive activity into two overall directions: dorsal and radial. We have shown earlier that dorsal protrusions depend on the Rac pathway (van Impel et al., 2009), whereas we show here that radial protrusions are particularly sensitive to loss of Cdc42 function. These results suggest that FGF signals might be differentially transduced within the migrating collective or that radial protrusive activity uses distinct molecular pathways.

Based on the evidence presented here and in other studies, we propose that FGF signalling performs three key functions in controlling mesoderm cell behaviour (Fig. 8C): (1) FGF triggers actin-dependent protrusive activity during flattening; (2) FGF induces modulation of E-cadherin distribution during EMT; (3) FGF acts as an attractant for dorsal migration. Therefore, the key cellular processes that depend on FGF are the remodelling of E-cadherin adhesions and the guidance of directional protrusive activity. Although the molecular details of the signalling pathways remain to be discovered, our data suggest that distinct small GTPase pathways, such as Cdc42 and Rac, play crucial roles in determining the specificity of the FGF signalling responses that direct cell behaviours during mesoderm layer formation.

We thank Sabine Schumacher for construction of the twi-GFP-actin5C transgenes and Sam Swift and Paul Appleton and the staff of the Centre for High resolution Imaging at the College of Life Sciences (Dundee) for support and discussions. We thank Nick Brown and the Bloomington stock centre for fly stocks, Ryan Webster for expert technical assistance and Sirin Otte for her contribution to the EMT study. We thank David Finnegan for hosting IBNC during the final stages of this work. We thank Rudi Winklbauer, Kees Weijer and members of the H.-A.J.M laboratory for discussions and critical comments on the manuscript. Flybase was used as a reference source throughout this work. This work was funded by a MRC Non-Clinical Senior Fellowship (GO0501679) and MRC project grant (GO0901020) to H.-A.J.M. and by funds from the College of Life Sciences, University of Dundee. Deposited in PMC for release after 6 months.