Mesoderm migration in Drosophila is a multi-step process requiring FGF signaling and integrin activity

Controlled cell migration is an essential aspect of development in which cells relocate to respond to chemical signals and form structures (Lecaudey and Gilmour, 2006; Rorth, 2007; Montell, 2008; Ilina and Friedl, 2009). Aberrant migration, by contrast, can lead to diseases such as metastatic cancer (Deisboeck and Couzin, 2009; Friedl and Gilmour, 2009). As a result, studying the molecular and physical mechanisms that control migration is crucial for understanding both development and disease. Several models have been developed for examining different types of cell migration in vivo, such as the border cells in Drosophila melanogaster and the lateral line in Danio rerio for studying small group migrations, the neural crest cells in vertebrates for studying streaming, and wound healing for understanding sheet migration (Friedl and Gilmour, 2009; Rorth, 2009; Weijer, 2009). Here, we study the migration of the mesoderm during gastrulation in Drosophila melanogaster embryos as it is a tractable model for the collective migration of hundreds of mesenchymal cells that can be characterized by quantitative analysis (McMahon et al., 2008; Supatto et al., 2009).

Mesoderm migration in Drosophila involves several movements that transform a tube of cells into a monolayer; the completion of this migration is important for muscle and heart development (Leptin and Grunewald, 1990; Wilson and Leptin, 2000). First, the mesoderm invaginates by apical constriction to form an epithelial tube within the embryo. The mesoderm then undergoes an epithelial-to-mesenchymal transition (EMT) and collapse of the tube follows. Next, the collapsed cells spread dorsally along the ectoderm. Lastly, the mesoderm transforms from a multi-layer to a monolayer. This sequence of events has been described previously, but it was not known if these migratory actions were distinct or overlapping events. Furthermore, it has not been established whether particular biochemical signals are required to coordinate each event.

The most well-characterized molecular action during mesoderm migration is fibroblast growth factor (FGF) signaling (Wilson et al., 2005; Murray and Saint, 2007; McMahon et al., 2008; Kadam et al., 2009; Klingseisen et al., 2009). FGF signaling is essential in animals for both differentiation and migration (Thisse and Thisse, 2005). The FGF receptor (FGFR) Heartless (Htl) has been studied extensively in the context of mesoderm migration and differentiation (Beiman et al., 1996; Gisselbrecht et al., 1996) and has recently been shown definitively to control organized collapse of the mesodermal tube onto the underlying ectoderm during Drosophila gastrulation (McMahon et al., 2008). This organization helps maintain the collective behavior of the mesoderm, as the absence of Htl results in two behaviorally distinct cell populations. However, it remains unclear how the two ligands for Htl, the Fgf8-like Pyramus (Pyr) and Thisbe (Ths) proteins, contribute to this process.

In the Drosophila system, two different models have been presented regarding how Pyr and Ths activate the Htl receptor during mesoderm migration. The first model proposes that the ligands function redundantly and provide robustness, and the second suggests that the ligands activate the receptor differentially (Kadam et al., 2009; Klingseisen et al., 2009). These previous studies, which include our own previous work, addressed the role of Pyr and Ths ligands by extrapolating their functions during the dynamic process of migration through examination of fixed tissues. Thus, it had yet to be determined definitively whether both Pyr and Ths are required for mesoderm migration during Drosophila gastrulation and, furthermore, whether the ligands regulate specific aspects of migration. Therefore, in this work, we explored the roles of Pyr and Ths during mesoderm migration using in vivo imaging and quantitative analyses; this general approach was used previously to decipher the FGFR mutant phenotype (McMahon et al., 2008; Supatto et al., 2009).

In addition to studying the two FGF ligands, we examined other molecules that could contribute to specific steps during mesoderm migration to test the hypothesis that mesoderm migration has temporally distinct inputs. We chose to examine the small GTPase Rap1 and integrins, as both have been implicated in migration and linked to FGF signaling (Mori et al., 2008; Carmona et al., 2009; Franzdottir et al., 2009). Rap1 has a demonstrated role in cell adhesion and migration in other systems (Huelsmann et al., 2006; Jeon et al., 2007; Boettner and Van Aelst, 2009). Rap1 regulates cell adhesion and migration, in part, through integrin activation (Reedquist et al., 2000; Kooistra et al., 2007; Boettner and Van Aelst, 2009; Carmona et al., 2009). Integrins are required for cell-cell junction formation and provide a physical link from these junctions to the actin cytoskeleton (Delon and Brown, 2007; Vicente-Manzanares et al., 2009). Of the two βPS subunits, only the βPS integrin, Myospheroid (Mys), is expressed in the Drosophila embryo during mesoderm migration (Leptin et al., 1989). Mys is involved in recruiting two alpha integrin subunits, αPS1 (Multiple edematous wings) and αPS2 (Inflated), to the cell membrane to form adhesion complexes that are important for cell migration and muscle attachment throughout Drosophila development (Leptin et al., 1989; Brown, 2000; O'Reilly et al., 2008). This evidence led us to investigate a role for Rap1 and Mys during mesoderm migration.

In this work, we present evidence that mesoderm migration is a multi-step process with temporally distinct migratory events. We show that movements in the radial direction, specifically collapse and monolayer formation, are controlled by FGF signaling. Dorsal movements appear to be FGF-independent. We find that the integrin subunit Mys is required only for monolayer formation. These results indicate that collapse, spreading and monolayer formation are not only temporally distinct, but also probably molecularly distinct.

All crosses and strains were maintained at 25°C. The following lines were used: yw; klar¹; htl^AB42/TM3,ftz-lacZ; His2AV-GFP; twi-gal4; twi-CD2; mys¹,FRT19A/FM7c,ftz-lacZ (Bloomington Stock Center); DfBSC25; pyr^e02915; ths^e02026; Df²³⁸ (Kadam et al., 2009); pyr¹⁸; ths⁷⁵⁹ (Klingseisen et al., 2009); Rap1^CD3 (Asha et al., 1999); klar¹,His2AV-GFP; klar¹,His2AV-GFP,htl^AB42/TM3,P[Dfd-GMR-nvYFP]3, Sb¹ (McMahon et al., 2008). Wild-type refers to yw or His2AV-GFP flies. Germline clones were made for Rap1^CD3 and mys¹ using standard FRT-mediated germline clone methodology (Chou and Perrimon, 1996).

Anti-sense morpholinos were designed using the GeneTools Oligo Design and Ordering System (Gene Tools, LLC). The following sequences were used to make morpholinos: pyr, CATTGGGCATGAACTTGTGGAACAT; ths, GCAGTCTCTCTAACTGATTCGACAT; Gal4, CGATAGAAGACAGTAGCTTCATCTT; mys, TCGAGGATCATGGCTTTGGCGGTTA.

Morpholinos were resuspended in water to a final concentration of 1.5 to 2 mM. Filtered liquid green food coloring was added at 1/10 (vol/vol) to aid in visualization of injection. The injection protocol used was a modified version of Misquitta and Paterson (Misquitta and Paterson, 1999). yw or His2Av-GFP flies were collected in 15-minute intervals, washed with water to remove yeast and debris, lined up on a glass slide in a small volume of water and allowed to dry for 10 minutes before injection. Embryos were then covered with a thin layer of Halocarbon Oil 27 (Sigma-Aldrich). A morpholino or buffer alone was loaded into machine-pulled (Narishige) glass needles (FHC Inc.). To prevent needle clogging, morpholinos were heated to 65°C and allowed to cool at room temperature prior to being loaded into the needle. Morpholinos were injected into the ventral or dorsal side of the pre-cellularized embryo using a Picospritzer (Parker Instrumentation) set to a 40 millisecond 60 PSI ejection, delivering approximately 100-200 pL into each embryo. Embryos were allowed to recover for at least two hours at 18°C in a humidified chamber. When the embryos reached stage 5, the embryos were set up for fixing or live imaging as previously described (Frasch, 1995; McMahon et al., 2008; Supatto et al., 2009).

Double-stranded RNA (dsRNA) targeted to Pyr and Ths transcript was designed as an additional control to confirm the mesoderm migration phenotype following previously described methods (Misquitta and Paterson, 1999). The following primers were used to amplify portions of the pyr and ths cDNAs (Stathopoulos et al., 2004): pyr, 5′-GGATCCTAATACGACTCACTATAGGATTGCGCGGCTACAGATACT-3′ and 5′-GGATCCTAATACGACTCACTATAGGATATTTCGCCTTGATTTGCG-3′; ths, 5′-GGATCCTAATACGACTCACTATAGGAGATCACCTGGACAATTCCG-3′ and 5′-GGATCCTAATACGACTCACTATAGGCCGTATGGGTCTCTTCATGG-3′. dsRNA was made from PCR products using the T7 RNA polymerase.

Embryos were fixed and stained using in situ, antibody or combined antibody and in situ protocols as previously described (Lehmann and Tautz, 1994; Frasch, 1995; Kosman et al., 2004). The following antibodies were used in this study: guinea pig anti-Twist (Mike Levine, UC Berkeley, USA), rabbit anti-Even skipped (Manfred Frasch, University of Erlangen, Nürnberg, Germany), rabbit anti-Beta galactosidase (Molecular Probes), mouse anti-rat CD2 (Serotec) and mouse anti-integrin-βPS (Developmental Studies Hybridoma Bank). Embryos were mounted in Permount (Fischer Scientific) for wholemount studies or embedded in acetone-araldite (Electron Microscopy Sciences) and cut with a microtome (LKB Bromna) to create 10 μm sections. Fluorescent images were obtained with a Pascal confocal microscope (Carl Zeiss).

Embryos were imaged as previously described (McMahon et al., 2008; Supatto et al., 2009) using a Zeiss LSM 510 inverted microscope (Carl Zeiss) at 940 nm wavelength (Chameleon Ultra laser, Coherent). At least three embryos for each of the following backgrounds were imaged and tracked: wild-type, pyr morpholino, ths morpholino, htl^AB42 mutant, pyr and ths double morpholino and mys morpholino. In addition, one null mutant was imaged for pyr (pyr^e02915) and ths (ths^e02026/ths^Df238) to confirm that the morpholino data was consistent with the null alleles. Nuclear tracking was performed on imaging data as previous described (McMahon et al., 2008) using Imaris software (Bitplane). Data from Imaris was exported to Matlab (The Mathworks) using ImarisXT and analyzed as previously described using custom Matlab scripts (Supatto et al., 2009). Briefly, tracking data from the ectoderm was fit to a cylinder in order to convert the coordinate system used during imaging (i.e. Cartesian) into cylindrical coordinates. This allows for analysis of each movement along the corresponding body axis. A color code was applied to show the organization of the mesoderm cells as they collapse and spread along the ectoderm.

To quantify intercalation events, a customized Matlab program was created to examine each row of mesoderm cells over time. Each cell was sequentially highlighted in blue so that it could be followed during monolayer formation. A cell was counted as being stably incorporated into the monolayer if it joined the monolayer and remained through stage 11 (∼130 minutes after tube collapse). Linear fits to the data from the final four time-points in Fig. 5J were performed, with the time data centered and scaled. The intercept parameters for the linear fits were then compared pair-wise by Welch's modified t-test. The largest P-value found was 0.0018.

Protrusive activity was quantified by measuring the number of large protrusions (i.e. greater than one hair-like extension per cell) per image, within a length of ∼60 microns across per image. The numbers were compared using Welch's t-test.

Mesoderm migration involves a series of complex behaviors that transform a tube of cells into a sheet of cells (Leptin and Grunewald, 1990; Wilson et al., 2005). Before migration begins, the mesoderm invaginates into the interior of the embryo via apical constriction of epithelial mesoderm cells, forming a mesoderm tube. Next, the mesoderm cells lose their epithelial characteristics and migrate toward the ectoderm (mesoderm tube collapse, Fig. 1A,B). The cells then change direction and move dorsally along the ectoderm (Fig. 1D,E). Lastly, mesoderm cells that are not in contact with the ectoderm do so, forming a monolayer (Fig. 1G,H).

Using live imaging of wild-type embryos, we explored whether the movements that encompass mesoderm migration are distinct or overlap temporally. Embryos with ubiquitously expressed H2A-GFP were imaged and mesoderm cells were tracked as previously described (Supatto et al., 2009). Tracking data was transformed into cylindrical coordinates using Matlab to fit the body plan of the embryo, so that the radial coordinate, r, represents cell movements from the center to the surface of the embryo, toward the ectoderm layer (e.g. collapse of the mesodermal tube and intercalation; Fig. 1C,I), and the movement along the curvature of the embryo, θ, represents motion in the angular direction, associated with the dorsoventral axis (e.g. dorsal spreading; Fig. 1F). In our previous study, we focused on decomposing the 3D movement of cells in particular directions (McMahon et al., 2008). In this work, we highlight the fact that collapse, spreading and monolayer formation are temporally distinct (Fig. 1C,F,I). We hypothesized that these movements involve different types of migratory behaviors guided by distinct molecular signals. As a result, our aim was to define the role of the genes involved in regulation of mesoderm migration within this temporal and spatial framework.

FGF signaling has been previously shown to be important for mesoderm migration. We showed recently that the preliminary function of the FGFR Heartless (Htl) is to support symmetrical collapse of the mesoderm onto the ectoderm (McMahon et al., 2008). We set out to find whether the ligands for Htl – the FGFs Pyramus (Pyr) and Thisbe (Ths) – are both required for mesoderm migration and, if so, whether they have distinct roles in migration. Pyr and Ths are expressed in dynamic patterns throughout development and have non-overlapping expression domains during mesoderm migration (Fig. 2A,B) (Gryzik and Muller, 2004; Stathopoulos et al., 2004). The pyr and ths mutant phenotypes were previously described using fixed sections. One study found that pyr and ths mutants both have a mesoderm monolayer defect (Kadam et al., 2009), whereas another claimed that only pyr was important for monolayer formation (Klingseisen et al., 2009), demonstrating that analysis of dynamic processes using fixed sections can be inconsistent, especially if the phenotype is variable or subtle.

We confirmed by statistical analysis of fixed sections that pyr and ths mutants do both exhibit a non-monolayer mesoderm phenotype, one weaker than that of the FGFR htl mutant (Fig. 2C-J; see Fig. S1 in the supplementary material). All available mutants produce similar phenotypes, with ths supporting a severe non-monolayer phenotype more frequently than pyr (Table 1; see Fig. S1 in the supplementary material). Placing a pyr allele over a ths allele was able to rescue monolayer formation, dismissing the possibility that a second site mutation contributes to the observed phenotype (see Fig. S1 in the supplementary material). In addition, we generated morpholinos and double-stranded RNAs (dsRNAs) to both pyr and ths, which produced similar phenotypes to the loss-of-function mutants (Fig. 2K-N; see Fig. S1 in the supplementary material). By analyzing several different mutant backgrounds, it is clear that both Pyr and Ths are important for mesoderm migration. We therefore used in vivo imaging to determine their precise role in this dynamic process.

We used two-photon microscopy to image pyr and ths mutants expressing ubiquitous H2A-GFP, which permits simultaneous tracking of mesoderm and ectoderm cells during gastrulation (Fig. 3A-D) (McMahon et al., 2008). This permitted us to decompose the migration into different types of movements and to decipher the subtle non-monolayer phenotypes.

To facilitate more efficient live imaging, we utilized translation blocking morpholinos (MOs) designed against pyr and ths transcripts to reduce the number of imaging sessions required to obtain mutant data; when assaying embryos of zygotic recessive mutant backgrounds, only one of four embryos was a homozygous mutant, whereas each morpholino injected embryo displayed the expected phenotype. Morpholinos injected into pre-cellularized embryos were able to reproduce the pyr and ths phenotypes of loss-of-function alleles (Fig. 2K-M; see Fig. S1G in the supplementary material). In addition, co-injection of pyr and ths morpholinos supported a mutant phenotype that was more severe and comparable to that of htl mutants (compare Fig. 2J with 2N) as well as the mutant background Df(2R)BSC25, which removes both pyr and ths genes (data not shown) (Stathopoulos et al., 2004).

We imaged both morpholino and null mutants for Pyr and Ths (see Materials and methods) and tracked a subset of mesoderm cells over time using Imaris software (see Movies 1 and 2 in the supplementary material). As with wild-type embryos (see Fig. 1), tracking data was converted into cylindrical coordinates to fit the body plan of the embryo (Fig. 4A). When the movement was decomposed into r and θ, it revealed that ths mutants, like previously characterized htl mutants, have a mesoderm tube collapse defect in which cells from the uppermost part of the tube fail to migrate toward the ectoderm (blue lines, Fig. 4B-C,E,G). In htl mutants, tube collapse was asymmetrical, with the tube falling either toward the left or right half of the embryo, resulting in an indirect migratory defect along θ (Fig. 4D,F) (McMahon et al., 2008). Unlike in htl mutants, however, movement in the angular direction was at worst very mildly affected in a few cells in ths mutants, suggesting that Pyr can keep the collapse symmetrical in the absence of Ths (Fig. 4H). pyr mutants displayed little to no defects along r or θ (Fig. 4I,J), which suggests that Ths is able to support mesodermal tube collapse in the absence of Pyr.

To further characterize the non-monolayer phenotype of pyr and ths mutants, we focused on the small cell movements and rearrangements found in intercalation, as the non-monolayer in pyr mutants cannot be accounted for by a collapse defect. In this particular case, we investigated whether intercalation events might support the generation of the mesoderm monolayer during gastrulation (Fig. 5A). We quantified the rate and number of intercalation events in wild-type and mutant backgrounds to see whether mesoderm intercalation is dependent on FGF signaling and whether the timing of intercalation corresponds with monolayer formation. Monolayer formation occurs during stage 9 and 10 and involves the transformation of a multi-layer into a single cell layer (∼80-90 minutes into migration; Fig. 5B,C).

By focusing on the position of mesoderm cells during stage 9 and 10 (gray spots in Fig. 5), it was apparent that a subset of cells is not incorporated into the monolayer in pyr, ths and htl mutants (Fig. 5D-I, arrowheads). We used the tracking data from each mutant to examine the timing and number of intercalation events. We found that pyr has a reduced number of intercalation events compared with wild-type, that ths mutants have even less intercalation events than pyr, and that htl mutants have the fewest events (Fig. 5J; number of cells assayed was 303, 241, 262 and 213 for wild-type, pyr, ths and htl, respectively, with P<0.002 in all cases). Together, these data suggest that the presence of FGFs throughout the ectoderm is important for intercalation of all mesoderm cells to form a monolayer. The defects did not necessarily correspond with the particular expression pattern of each ligand. This is not surprising as ligands, including Fgf8, might have non-autonomous effects due to diffusion (e.g. Yu et al., 2009).

It has been previously shown that Htl, Pyr and Ths can influence cellular projections during collapse and spreading (Schumacher et al., 2004; Wilson et al., 2005; Klingseisen et al., 2009). As migratory defects often coincide with a failure to regulate cell shape changes and protrusive activity (McDonald et al., 2008), we examined whether the ligands control cell shape changes that might also be important for movement during monolayer formation. We visualized the protrusions using twist promoter-supported expression of CD2, a cell-surface protein not native to Drosophila, which permits examination of cell extensions exclusively in the mesoderm during stage 9 and 10 (Twist-CD2) (Dunin-Borkowski and Brown, 1995). We found that, as previously published, the leading edge is affected at stage 9, as cells fail to polarize in embryos lacking both Pyr and Ths or in pyr single mutants (see Fig. S2A-C in the supplementary material, arrow), but not in ths single mutants (see Fig. S2D in the supplementary material) (Klingseisen et al., 2009). Similar effects were observed when the ligands were ectopically expressed in the mesoderm: overexpression of pyr leads to severe loss of cellular extensions, whereas overexpression of ths has a minor effect (see Fig. S2E-F in the supplementary material) (Klingseisen et al., 2009).

Conversely, it had not been examined previously whether mesoderm cells extend protrusions toward the ectoderm during monolayer formation. We found that, in the pyr/ths double mutant, mesoderm cells extend fewer large protrusions into the ectoderm than in wild-type embryos (see Fig. S2A,B in the supplementary material, arrowheads). Mesoderm sections from double-mutant embryos contained 4.0±0.8 protrusions per image (n=11), whereas those from wild-type exhibited 7.7±0.9 protrusions (n=11, P<0.01). pyr and ths single mutants also failed to extend as many protrusions into the ectoderm as wild-type (see Fig. S2C,D in the supplementary material; 4.3±0.9 and 4.8±1.5, respectively, n=9 for each; P<0.01 for each compared with wild-type). These data suggest that protrusive activity might be important for monolayer formation and provide insights into the mechanism by which FGF signaling might control radial intercalation.

After characterization of the FGF mutants, we screened for other genes that produce similar phenotypes to the FGF mutants using fixed section analysis. To this end, we discovered that embryos mutant for the GTPase Rap1 have collapse and monolayer formation defects similar to those mutant for Htl (Fig. 6A-C,E-G). However, Rap1 mutants also exhibit defects in ventral furrow formation and germband elongation, making the interpretation of its primary roles in mesoderm migration difficult (Roote and Zusman, 1995; Asha et al., 1999). Therefore, we sought out targets of Rap1 that displayed more-specific mesoderm migration defects.

Several studies suggest that Rap1 is required for activation of integrins at the cell membrane, which in turn is required for cell adhesion and migration (for a review, see Bos, 2005; Boettner and Van Aelst, 2009; Vicente-Manzanares et al., 2009). This led us to explore the role of integrins during mesoderm migration in Drosophila. There are two beta integrin subunits in Drosophila, but only the βPS subunit, Myospheroid (Mys), is expressed during mesoderm migration, in between the mesoderm and ectoderm at stage 9 and 10 (see Fig. S3A-F in the supplementary material) (Leptin et al., 1989; Gotwals et al., 1994).

We found that mys mutants exhibit a non-monolayer mesoderm defect at stage 9 and 10 in fixed sections (Fig. 6D,H). Surprisingly, we found that Mys localization is affected in both htl and Rap1 mutants, with gaps and reduced expression of Mys in htl mutants and a total absence of Mys in Rap1 mutants (Fig. 6I-K). These results suggest that Mys plays a specific and crucial role during monolayer formation and that Mys expression might be regulated, at least in part, by FGF signaling. This result also suggests the possibility that FGF signaling functions during mesoderm migration and spreading to regulate cell adhesion (see Discussion). To definitively test the role of Mys in intercalation, we dissected the phenotype using quantitative imaging methods.

We performed live imaging on H2A-GFP embryos injected with a translation blocking morpholino designed against the mys transcript (see Fig. S3G-J in the supplementary material). The Mys morpholino was able to reproduce the phenotype of the genetic null mutant and also eliminate Mys protein expression in the embryo (see Fig. S3K,L in the supplementary material). We tracked a subset of mesoderm cells from mys mutant imaging data and analyzed movement in r and θ (see Movie 3 in the supplementary material). We found that neither collapse (r) nor spreading (θ) is affected by loss of Mys (Fig. 7A,B). Like the FGF ligand mutants, we found a reduced number of intercalation events during monolayer formation in mys mutants compared with wild-type (Fig. 7C,F,G; 90 cells were assayed for mys mutants, P<0.05). We also found that mesoderm membrane protrusions into the ectoderm were completely absent in mys mutants during the same time interval as monolayer formation (i.e. stage 9 and 10), which could be contributing to the observed intercalation defects (Fig. 7D,E; see Fig. S2G,H in the supplementary material). Our data indicate that Mys is important for monolayer formation and provide support for the view that this migratory event is molecularly distinct from earlier events, as ventral furrow formation, collapse and spreading are unaffected in mys mutants.

Mesoderm migration in Drosophila is a combination of complex three-dimensional movements involving many molecular components. We have demonstrated here that live imaging, coupled with quantitative analyses, is important for studying complex cell movements, as it allowed us to decompose migration into different movement types and thus to describe subtle phenotypes. First, we extended analysis of the directional movements of mesoderm cells within wild-type embryos, focusing on the temporal sequences of events. We found that cells follow a sequential and distinct set of trajectories: movement in the radial direction (tube collapse: –5 to 15 minutes, 0=onset of germband elongation), followed by movement in the angular direction (dorsal migration: 15 to 75 minutes) and ending with small intercalation movements in the radial direction (monolayer formation: 75 to 110 minutes). These movements appear temporally distinct (i.e. stepwise), and thus we searched for molecular signals controlling each process.

We investigated which mesoderm movements were FGF-dependent and, in particular, either Ths- or Pyr-dependent. The interaction between Htl and its two ligands provides a simpler system relative to vertebrates (which exhibit over 120 receptor-ligand interactions) in which to study how and why multiple FGF ligands interact with the same receptor. Previously, we had found that FGF signaling via the Htl FGFR controls collapse of the mesodermal tube but not dorsal-directed spreading (McMahon et al., 2008). Here we demonstrated that FGF signaling is also required for monolayer formation. In addition, we defined distinct, non-redundant roles for the FGF ligands: Ths (but not Pyr) is required for collapse of the mesodermal tube, whereas both Pyr and Ths are required for proper intercalation of mesoderm cells after dorsal spreading.

This analysis raises questions about ligand choice during collapse and monolayer formation. Within the mesodermal tube, cells at the top require a long-range signal in order to orient towards the ectoderm during tube collapse, whereas the signals controlling intercalation during monolayer formation can be of shorter range. We suggest that the ligands have different activities that are appropriately tuned for these processes. In fact, recent studies of the functional domains of these proteins suggest that Ths has a longer range of action than Pyr (S. Tulin and A.S., unpublished results), in agreement with our analysis that Pyr does not support tube collapse, but does have a hand in monolayer formation.

We have demonstrated here that Rap1 mutants have a similar mesoderm phenotype to the FGFR htl mutant, with defects in collapse and monolayer formation. We were unable to establish whether Rap1 acts downstream of FGF signaling, as the complete loss of Mys in Rap1 mutants is more severe than the patchy expression of Mys seen in htl mutants. Therefore, Rap1 could be working in parallel to or downstream of FGF signaling during mesoderm migration. Rap1 has been implicated in several morphogenetic events during Drosophila gastrulation and probably interacts with many different signaling pathways (Roote and Zusman, 1995; Asha et al., 1999). Further study of Rap1, along with other GTPases, will shed light onto their role during mesoderm migration, how they interact with one another and what signaling pathways control them.

We chose to focus on the more-specific phenotype of mys mutants, as its localization is affected in htl mutants and it exhibits a monolayer defect that is similar to pyr and ths mutants. Integrins are important for cell adhesion, so it is not surprising that cells fail to make stable contact with the ectoderm through intercalation in mys mutants. However, some cells do contribute to monolayer formation in the absence of Mys, implying that other adhesion molecules are involved in maintaining contact between the mesoderm and ectoderm. These other adhesion molecules might be activated downstream of FGF signaling as the htl mutant monolayer phenotype is more severe than the mys mutant. Discovering the downstream targets of Htl, which we suggest might regulate cell adhesion properties, will help to shed light on the mechanisms supporting collapse of the mesodermal tube (which is not dependent on Mys) and monolayer formation (which is Mys-dependent).

Cell protrusions, such as filopodia, are important for sensing chemoattractants and polarizing movement during migration (for a review, see Mattila and Lappalainen, 2008). Previous studies have focused on protrusive activity at the leading edge during mesoderm migration in Drosophila and shown that these protrusions are FGF-dependent (Schumacher et al., 2004; Klingseisen et al., 2009). In this study, we have found that protrusions exist in all mesoderm cells, not just the leading edge, and that these protrusions also extend into the ectoderm.

Our study demonstrates that FGF signaling, as well as integrin activity, is required to support protrusive activity into the ectoderm; this is a potential mechanism by which FGF signaling and Mys could control movement toward the ectoderm during monolayer formation. The function of protrusions at the leading edge remains unclear, as they appear to be reduced in pyr and mys mutants (see Fig. S2 in the supplementary material), but migration in the dorsal direction still occurs in both mutant backgrounds. One interpretation is that FGF and Mys are important for generalized protrusive activity and that extensive protrusions are required for intercalation but not dorsal migration.

Based on our study, we propose that mesoderm migration is a stepwise process, with each event requiring different molecular cues to achieve collective migration (Fig. 8). Invagination of the mesoderm is the first step in this process and is dependent on Snail, Twist, Concertina, Fog and several other genes (Parks and Wieschaus, 1991; Reuter and Leptin, 1994; Morize et al., 1998; Aracena et al., 2006; Seher et al., 2007; Martin et al., 2009). Next, collapse of the mesoderm tube onto the ectoderm requires Htl activation via Ths. Rap1 might be involved in this process as well but the phenotype of Rap1 mutants is complex and it is unclear which phenotypes are primary defects (see Fig. 6C,G) (Roote and Zusman, 1995; Asha et al., 1999).

Following collapse, mesoderm cells spread dorsally by an unknown mechanism. Dorsal migration is unaffected in pyr and ths mutants and occurs in all cells that contact the ectoderm in htl mutants, implying that FGF signaling is, at most, indirectly involved in this step owing to the earlier tube collapse defect (McMahon et al., 2008). Whether dorsal migration requires chemoattractive signals or whether the cells simply move in this direction because it is the area of least resistance remains unclear.

Finally, after dorsal spreading is complete, any remaining cells not contacting the ectoderm intercalate to form a monolayer. This process is controlled by a combination of both Pyr and Ths interacting through Htl and also by Rap1 and Mys. In other systems, intercalation can lead to changes in the properties of the cell collective, for instance, lengthening of a body plan (Keller, 2006). However, we have shown here that dorsal migration and spreading are not a result of intercalation, as intercalation occurs after spreading has finished (Fig. 5).

Coordination of these signals to control collective migration enables the mesoderm to form a symmetrical structure, which is essential for embryo survival. This model begins to address the question of how hundreds of cells move in concerted fashion and is relevant for a generalized understanding of embryogenesis and organogenesis. We find that mesoderm migration is accomplished through sequential movements in different directions, implying that collective migration might be best achieved by distinct phases of movement.

We thank Manfred Frasch, Iswar Hariharan, Maria Leptin, Mike Levine and Arno Müller for providing fly stocks and antibodies. We also thank the Caltech Biological Imaging Center for use of their microscopes. This work was supported by grants to A.S. from the NIH (R01 GM078542), and in part by a grant from the Jane Coffin Childs Memorial Fund for Medical Research as G.T.R. is a fellow of the Jane Coffin Childs Memorial Fund for Medical Research. Deposited in PMC for release after 12 months.