FGF controls epithelial-mesenchymal transitions during gastrulation by regulating cell division and apicobasal polarity

During embryonic development, most tissues arise from a series of cell conversions between epithelial and mesenchymal states (Baum et al., 2008). During epithelial-to-mesenchymal transition (EMT), cells decrease contacts and lose polarity, while the opposite occurs during mesenchymal-to-epithelial transition (MET) as cells regain cell polarity and increase cell-cell adhesion (Thiery et al., 2009). Both processes have been studied because they are pivotal for tissue and organ specification, and also because their misregulation can contribute to metastasis (Nieto, 2013).

In Drosophila embryos, the presumptive mesoderm undergoes a series of orchestrated cell movements (Bae et al., 2012; reviewed by Gilmour et al., 2017). Previous studies have focused on the mechanism by which these cells invaginate into the embryos as well as how they initiate migration once internalized. Mesoderm cells express the FGF receptor Heartless (Htl) (Gisselbrecht et al., 1996), whereas the overlying ectoderm expresses the FGF ligands Pyramus (Pyr) and Thisbe (Ths) (Gryzik and Müller, 2004; Stathopoulos et al., 2004). FGF signaling is thought to orient collapse of the invaginated mesodermal tube through chemoattractant-directed movement, whereas signaling is likely not required to support directed migration because a subset of cells moves directionally even in the absence of FGF signaling (McMahon et al., 2008; Muha and Müller, 2013; Murray and Saint, 2007; Wilson et al., 2005). At the end of the spreading process, mesoderm cells form a monolayer; FGF signaling is required to help establish this organization, as the mesoderm remains multilayered in mutants (McMahon et al., 2010). This multilayered phenotype, specifically, is also exhibited by mutants of the myospheroid (mys) gene, which encodes β-PS integrin (Leptin et al., 1989; MacKrell et al., 1988). Mys protein is localized at the interface between mesoderm and ectoderm at the end of the spreading process, as the monolayer is formed (Roote and Zusman, 1995). In FGF mutants, Mys expression is predominantly lost from the mesoderm-ectoderm interface (McMahon et al., 2010).

Here, we have investigated whether FGF signaling has additional roles in supporting Drosophila mesoderm cell migration that relate to EMT and MET. During gastrulation of vertebrate embryos, FGF supports mesodermal EMT by regulating levels of E-Cadherin transcription (Ciruna and Rossant, 2001); in Drosophila, this process is thought to be FGF independent, as collapse of the invaginated tube (considered the fly equivalent EMT) does occur in FGF mutants, albeit misdirected. However, in a recent study, we determined that the htl gene is maternally provided (Irizarry and Stathopoulos, 2015), opening up the possibility that the zygotic mutants characterized in previous studies exhibit only partial loss-of-function phenotypes. Furthermore, as the mutants fail to specify a mesoderm monolayer at the end of the spreading process at gastrulation (McMahon et al., 2010), we also assayed whether the FGF signaling pathway is involved in re-establishing cell polarity within mesoderm cells in order to support their subsequent differentiation into different cell lineages.

To provide insight into the role of FGF signaling in supporting mesoderm cell migration during Drosophila gastrulation, we examined the localization of the FGF receptor Htl. Antibody staining of endogenous protein or Htl-mCherry fusion protein (Irizarry and Stathopoulos, 2015) shows localization at the apical side of cells within the invaginated mesodermal tube (Fig. 1A,D″,D‴,E″,E‴, Fig. S1A,D-F). At collapse, when mesoderm cells begin their migration, Htl becomes enriched at the side of cells in contact with the ectoderm, especially within protrusions that invade the ectoderm (Fig. 1B,E″,E‴,G″,G‴, Fig. S1B,F,H), and remains enriched at the mesoderm-ectoderm boundary even when a monolayer forms (Fig. 1C,H″,H‴, Fig. S1C,I). This localization pattern suggests that FGF signaling may be involved in regulating cell-cell contacts, possibly through modification of adherens junctions (AJs).

We detected adherens junctions present in mesoderm cells over the course of migration using the colocalization of α- and β-Catenin, two proteins associated with AJs (Fig. 1D-H,D′-H′) (Baum and Georgiou, 2011; Oda et al., 1998; Tepass et al., 2001). AJs are concentrated at the apical side of invaginated mesoderm cells early on (Kolsch et al., 2007; Weng and Wieschaus, 2016), and we found that Htl is also concentrated in the same position (Fig. 1D‴). As the tube collapses, AJs become newly concentrated at the positions in which the mesoderm contacts the ectoderm (Fig. 1E‴, arrows) as do Htl and E-Cadherin (Fig. S1D-F). Later, immediately before cells divide, only diffuse staining of α- and β-Catenin is present, suggesting contact of rounded mesoderm cells with the ectoderm is weakened (Fig. 1F-F‴, Fig. S1G). As the mesoderm cells initiate their dorsally directed spreading movement, AJs are detectable again and enriched at the mesoderm-ectoderm interface, together with Htl (Fig. 1G‴, arrows) and E-Cadherin (Fig. S1H). At the end of the mesoderm-spreading process, when cells establish a monolayer organization, AJs are found at both mesoderm-ectoderm and mesoderm-yolk interfaces (Fig. 1H‴, arrow and arrowheads, respectively) as described previously by electron microscopy studies (Tepass and Hartenstein, 1994).

To investigate a role for FGF in regulating AJs, the number of AJs was counted within the mesoderm in control embryos as well as in various FGF mutant backgrounds from stage 7 to 10 using colocalization of α- and β-Catenin as a proxy (Fig. 2D,D′; and see Materials and Methods). To start, AJ number in htl^AB42 zygotic mutant (htl^M+Z−; Gisselbrecht et al., 1996) embryos was compared with control (htl^AB42/+ heterozygotes, which exhibit normal mesoderm spreading) (Fig. 2A,B). No difference between samples was identified at stages 7 and 8 (Fig. 2E). However, htl^M+Z− embryos had significantly fewer AJs at the end of the mesoderm-spreading process (stage 9/10) compared with control (Fig. 2E, orange line-associated asterisk, P<0.05), demonstrating a role for zygotic Htl in supporting increased AJ number during monolayer formation, which occurs at stage 9/10. To analyze a maternal role for Htl, we isolated embryos from females containing germline clones of the htl^AB42 loss-of-function mutation (i.e. htl^M−Z−) (Irizarry and Stathopoulos, 2015). In htl^M−Z− embryos, AJs were more abundant at stage 7 but also failed to increase at stage 9/10, when compared with control (Fig. 2E, gray line-associated asterisks, P<0.05); in embryos from germline clones that were zygotically rescued (htl^M−Z+), AJ numbers were similar to control at all stages (Fig. 2E).

These results show that either maternal or zygotic htl expression is sufficient to support normal number of AJs at stage 7, whereas zygotic, but not maternal, htl expression is required at stage 9/10 to support an FGF-dependent increase in AJ number. This analysis defines a role for Htl in regulating AJ number at gastrulation, the earliest characterized role for maternal Htl in the embryo, and suggests that FGF has distinct roles in the mesoderm at different stages. The observed decrease in AJ number from stage 7 to stage 8 supports the view that mesoderm cells exhibit a prolonged EMT process, as suggested by previous studies (e.g. Schafer et al., 2014), whereas the increase in AJ number measured at later stages 9/10 likely relates to a previously unappreciated process during mesoderm development.

To provide further insight into the role of FGF signaling, gain-of-function mutants were investigated. The FGF ligand Ths or constitutively active FGF receptor Htl (HtlCA) was ectopically expressed within the mesoderm cells at high levels using the twist-Gal4 driver (twi_h-Gal4; Carmena et al., 1998; Greig and Akam, 1993). Previous studies have shown that, in both cases, mesoderm spreading is aberrant, multilayered and uneven (Kadam et al., 2009). We found that ectopic expression of Ths ligand resulted in consistently high AJ number (Fig. 2C,F; green line-associated asterisks, P<0.05), whereas constitutively active receptor had a smaller, though reproducible, effect in which AJ numbers were increased at earlier stages but decreased at later stage 9/10 relative to the control (Fig. 2F; yellow line-associated asterisks, P<0.05). We hypothesized that both levels as well as direction of FGF receptor activation are important, as these variables have been shown to impact FGF function in other systems (e.g. Cota and Davidson, 2015).

To further understand the changes in AJs during mesoderm development, we investigated the localization of the cell polarity protein Bazooka (Baz), which has been found to act upstream of AJs in the establishment of epithelial polarity (Harris and Peifer, 2004; Laprise and Tepass, 2011). Baz protein is apically localized in the epithelialized Drosophila early embryo (Fig. 3A, stages 6-10) as well as within neuroblasts (NB), which are large cells present between the mesoderm and ectoderm (Fig. 3A, stages 9 and stages 10; Fig. 3E) (Roegiers et al., 2001). Recent studies have shown that Baz becomes concentrated at the apical side of cells within the invaginated mesodermal tube (Fig. 3A, stage 7) and is subsequently lost following tube collapse (Weng and Wieschaus, 2017). Nevertheless, we found that some Baz staining remains in puncta at cell-cell contacts throughout the mesoderm-spreading process, which likely correspond to AJs because α- and β-Catenin also colocalize (Fig. 3A, stage 8 and stage 9; data not shown). At the end of mesoderm migration, cells form a monolayer, and, surprisingly, Baz was found to localize to the side of mesoderm cells facing the yolk (Fig. 3A, stage 10).

Baz localization in FGF mutants is aberrant. In htl^M+Z− mutants, Baz protein remains clustered after tube collapse, whereas in control embryos, expression is scattered throughout the mesoderm (Fig. 3G,G′ compare with 3F,F′). At stage 10, Baz exhibits only partial localization to the presumptive apical side of mesoderm cells in the absence of zygotic Htl (Figs 3I,I′ and 4A,B, asterisk, P<0.0001) compared with clear localization to the presumptive apical side of mesoderm cells in the control (Fig. 3H,H′). Baz expression is even more noticeably reduced in the mesoderm of htl^M−Z− mutants at stage 10 (Fig. 3J,J′ compare with H,H′, Fig. 4B), although expression is detectable at earlier stages in scattered positions between cells as in the zygotic mutant (data not shown). Similar results were also obtained when embryos were analyzed by fluorescence-based staining using the same antibody, or using an anti-RFP or anti-GFP antibody to localize Baz-mCherry or Baz-GFP fusion protein (Fig. 3B-D, Fig. S1J,K,M-T). When co-stained with an anti-Htl antibody, we also discovered that Htl and Baz do not colocalize once a monolayer is specified (Fig. 3B).

To investigate whether FGF indirectly supports cell polarity by acting to support monolayer formation, we analyzed the localization of Baz in a htl^M+Z−,stg double mutant background in which monolayer formation is partially restored (arrow, Fig. 3Q) as opposed to the multi-layered, uneven mesoderm observed in htl^M+Z− single mutant embryos (e.g. Fig. 3I). string (stg) encodes a Cdc25-type phosphatase that regulates cell division (Edgar and O'Farrell, 1989). Normally, mesoderm cells divide twice during gastrulation; these divisions fail to occur in stg mutants, resulting in a monolayer with much fewer cells (Fig. 3M) (Leptin and Grunewald, 1990). However, Baz localization is undisrupted (Figs 3L,L′ and 4B). In contrast, Baz localization to the apical side is significantly diminished in htl^M+Z−,stg double mutants (Fig. 3P,P′ compare with L,L′, Fig. 4B), despite the increased propensity of the mesoderm to achieve a monolayer when compared with htl^M+Z− single mutants.

Furthermore, expression of FGF ligands or HtlCA in the mesoderm at high levels results in both aberrant migration (Fig. 3O,S; Michelson et al., 1998; Stathopoulos et al., 2004) and mislocalization of Baz (Figs 3N,N′,R,R′ and 4B, Fig. S2C,C′). Collectively, these results support the idea that FGF signaling is important for correct localization of Baz protein.

More generally, these results also suggest that mesenchymal cells transition to an epithelial-like state by regaining cell polarity at stage 9/10 when they form a monolayer. We examined localization of other cell polarity markers, including: atypical protein kinase C (aPKC), which is often located at the apical domain of cells (Wodarz et al., 2000); Myospheroid (Mys), which is often located on the basal side (Fernandes et al., 2014; Narasimha and Brown, 2013); as well as Neurotactin (Nrt), which is excluded from apical domains but present in basolateral regions (Müller and Wieschaus, 1996). Like Baz, aPKC was found to localize to the presumptive apical side of mesoderm cells at stage 9/10 once the monolayer is achieved (Fig. 4C,C′). This localization of aPKC is lost in htl mutants (Fig. 4E-F′,H,H′), upon ectopic expression of ths ligand in the mesoderm (Fig. 4D-E), or when Baz level is reduced in the mesoderm using tissue-specific RNAi (Fig. 4E,G,G′). Localization of aPKC to the apical domain appears reinforced upon ectopic expression of Baz-GFP in the mesoderm (Fig. 4E,I,I′). Nrt localization was also affected, as it was no longer excluded from the presumptive apical domain of cells in FGF gain or loss of function mutants (Fig. 4J-L). Previous studies had found that FGF signaling is required to support expression of the β-PS1 integrin Mys at the interface between mesoderm and ectoderm cells at the end of mesoderm spreading at stage 9/10 (McMahon et al., 2010). Additionally, through expression of a Mys RNAi line in the mesoderm, we defined that Mys expression in ventral regions is mesodermally derived (Fig. S2G,G′). Even less Mys is present at the mesoderm-ectoderm interface in embryos obtained from females containing htl^AB42 germline clones (htl^M−Z−) (Fig. 5I,I′ compare with 5A,A′ and 5E,E′,Q). This decrease likely relates to an FGF-direct effect on Mys protein, as transcript levels are not substantially changed in the mutant (Fig. S2I,J). Taken together, these data demonstrate that mesoderm cells acquire an apico-basal polarity, in which Baz, aPKC, Nrt and Mys are spatially localized within cells, and that this process is regulated by FGF signaling.

We also investigated the relationship between basally localized Mys and apically localized Baz, which influence the localization of one another in other cell systems (e.g. Fernández-Miñán et al., 2008). In embryos derived from mys^XG43 germline clone females (mys^M−Z−) (Boube et al., 2001; Leptin et al., 1989), which lack Mys expression (Fig. S2K-L,R), Baz expression was either greatly diminished or completely lost (Figs 3T,T′ and 4B). In turn, upon ectopic expression of Mys in the mesoderm, mesoderm cells formed a multilayer organization with only low levels of Baz localized to the apical position in regions where a monolayer is achieved (Fig. S2D-F′). Reducing Baz levels in the mesoderm through RNAi resulted in loss of Mys expression in the ventral region (Fig. S2N-O,R). Therefore, Mys and Baz appear to regulate each other.

To examine whether these mesoderm cells completely switch to an epithelial state and/or retain mesenchymal characteristics, we examined expression of E-Cadherin (E-Cad) and N-Cadherin (N-Cad), cell-adhesion molecules typically associated with either epithelial cells or mesenchymal cells, respectively (Iwai et al., 1997; Oda et al., 1994). Several studies have reported their expression dynamics and reciprocal regulation known as the ‘Cadherin switch’: as E-Cad levels decrease in the mesoderm, N-Cad levels increase (Oda et al., 1998; Schafer et al., 2014). Surprisingly, we found that E-Cad and N-Cad are co-expressed at the monolayer stage and exhibit spatially biased localization within cells. N-Cad is expressed in the apicolateral domain (Fig. 4T,T′), while E-Cad is present at both the mesoderm-ectoderm interface and the mesoderm-yolk interface at low levels (Fig. 4O,O′). Performing tissue-specific E-Cad RNAi in the mesoderm led to loss of E-Cad from both interfaces (Fig. 4Q,S,S′), indicating that E-Cad expression is indeed mesoderm derived (Clark et al., 2011). Whether expression of E-Cad at the monolayer stage is regained or perdures from earlier stages and becomes differentially localized remains unclear; however, the presence of E-Cad at the monolayer stage further supports the view that mesoderm cells undergo an MET, in which they exhibit characteristics of epithelial cell fate. The ‘hybrid Cadherin state’ uncovered, in which both E-Cad and N-Cad are co-expressed but differentially distributed within cells, is suggestive of distinct functions for these cadherins (see Discussion).

We hypothesized that local activation of the FGF receptor supports the definition of basal cell polarity by controlling Mys localization as Htl protein re-localizes to the presumptive basal cell surface after tube collapse (see Fig. 1B) and ligands Pyr and Ths are expressed in the overlying ectoderm (Kadam et al., 2009). Low-level expression of the wild-type htl cDNA using a different twist-Gal4 driver (twi_l-GAL4) (Lee et al., 2003; Marqués et al., 2002) can rescue htl zygotic mutant phenotypes to support monolayer formation (six out of seven embryos, Fig. 5H,P) and localized expression of Mys and Baz (Fig. 5G,G′,O,O′,Q,R). Expression of HtlCA in the same manner can also support monolayer formation in a fraction of embryos (five out of eight embryos, Fig. 5D,L) but, in contrast, little rescue of the Mys and Baz phenotype was observed (Fig. 5C,C′ compare with Fig. 5E,E′; Fig. 5K,K′, compare with Fig. 3I,I′; 5Q,R).

In contrast, ectopic expression of either HtlCA or ligands at higher levels in the mesoderm using twi_h-Gal4 driver resulted in severe multilayered mesoderm phenotypes in which both Mys (Fig. 5M,M′,N,Q, Fig. S2A,A′,B,S,S′,T,R) and Baz, as previously discussed (Figs 3N,N′,R,R′, 4A and Fig. S2C,C′M), are ectopically expressed. Under similar conditions, no ectopic expression was observed for aPKC (Fig. 4D-E), Nrt (Fig. 4N) and E-Cad at the meso-ectoderm interface (Fig. 4P-Q).

Collectively, these results demonstrate that the direction of FGF signaling is important for the acquisition of cell polarity at the end of mesoderm migration at the monolayer stage, whereas levels, but not necessarily direction, of signaling is important for mesoderm EMT (i.e. collapse and spreading). These results also suggest that FGF regulates localization of Baz and Mys to support cell polarity.

We reasoned that monolayer formation and/or polarity may influence differentiation of muscle cell types by ensuring even distribution of mesoderm cells upon the ectoderm. FGF signaling supports spreading of the mesoderm, such that cells may contact the dorsal ectoderm and be induced by signaling pathways active there to specify distinct cell types, including visceral mesoderm and pericardial cell/dorsal somatic lineages marked by bagpipe (bap) and even-skipped (eve) expression, respectively (e.g. Frasch, 1995; Frasch et al., 1987; Staehling-Hampton et al., 1994). For example, eve transcription requires both FGF and Dpp signaling, whereas bap expression is supported directly by Dpp signaling but requires FGF signaling only indirectly to support spreading (see Fig. 6B, compare with 6A; e.g. Stathopoulos et al., 2004). We assayed various mutant backgrounds that affect baz and mys levels, and found that the greater the associated non-monolayer phenotype, the less uniform the expression of bap or Eve observed (Fig. 6). For example, tissue-specific RNAi of baz or ectopic expression of mys supports non-uniform expression of bap transcript or Eve protein (arrows and arrowhead, Fig. 6E,G,J,L, also see Fig. S2E,H,N,P for spreading phenotypes), whereas mys loss-of-function mutants were less severe (Fig. 6D,F,H). Uniform mesoderm migration and monolayer formation likely ensures that the proper number of mesoderm cells contact ectoderm cells in order that the mesoderm can be induced to form all the various muscle lineages. In addition, the polarized localization of signaling regulators such as Mys to the mesoderm-ectoderm interface may support levels/efficiency of signaling pathway activation (Sawala et al., 2015; see Discussion).

We hypothesized that the widespread expression of Mys/Baz upon FGF gain-of-function influences the ability of mesoderm cells to migrate properly. Live in vivo imaging coupled with tracking analysis was used previously to investigate the htl^M+Z− mutant phenotype (McMahon et al., 2008) and the mys mutant phenotype (McMahon et al., 2010). An H2A-GFP transgene (Clarkson and Saint, 1999) is used to image mesoderm cell movements in time over the course of mesoderm cell migration. Tracking of mesoderm cell migration in htl^M+Z− mutant embryos demonstrated that movement is non-collective and a monolayer is not achieved, but did not identify any cell division phenotype in comparison with wild type (McMahon et al., 2008). In contrast, mesoderm migration in mys mutant embryos exhibited normal migration, like wild type, until the last stage, in which they failed to form a monolayer (McMahon et al., 2010). Here, we used this live imaging plus tracking approach to obtain additional insight into phenotypes associated with ectopic expression of the ligand Ths as well as knockdown of Baz, both in a mesoderm-specific manner. These two mutations appear to impact spreading in opposite ways from analysis of fixed samples (e.g. see Fig. 5M-N and Fig. S2M,N).

Initially, in movies of H2A-GFP transgene-containing embryos that also ectopically express Ths ligand through twi_h-Gal4 (Fig. 7A,B; and Movie 2), we found that cells barely move in the angular direction (Fig. 7C,G, Fig. S3C) compared with the normal spreading behavior exhibited by wild-type mesoderm cells (Fig. 7C,F, Fig. S3B; Movie 1; see also McMahon et al., 2008). A metric of collective cell migration (A) can be used to quantify this behavior that relates the azimuthal angular position of a cell at the beginning of the migration compared with its position at the end of the migration, when movements in the embryos are translated into cylindrical coordinates (Fig. 7A,B, Fig. S3A) (McMahon et al., 2008). When these data are plotted for each cell in the migrating collective, the slope of the line is a measure of the degree of spreading and the coefficient R characterizes the collective nature of cell movements. For wild type, A is close to 2, as also demonstrated previously, whereas upon ectopic expression of Ths, we found that A=1 (Fig. S3E,F, respectively). Together, our data show the cells fail to migrate dorsally but remain as a collective (R=0.93, Fig. S3F) upon ectopic expression of Ths in the mesoderm. These results also help explain how loss-of-function (i.e. htl^M−Z−) and gain-of-function (i.e. high level twi_h>ths) FGF mutants support similar phenotypes that result in increased AJs, but for different reasons. FGF mutants fail to take down AJs at the early stages (e.g. stage 7), resulting in aberrant migration at later stages (e.g. Fig. 5F), whereas ectopic expression of ligand leads to precocious expression of adhesion molecules (i.e. Mys), which accounts for similarly increased AJ number and cell migration phenotypes (e.g. Fig. 5N).

The tracking analysis also enabled us to examine more carefully the role for FGF in regulating cell divisions. In wild-type embryos, mesoderm cells divide twice with a period of about 60 min between divisions; furthermore, those cells located closer to the ectoderm divide earlier than those located at a distance (Fig. 7D,I,L). Upon ectopic expression of Ths, mesoderm cell division is delayed for most cells as the period between cell divisions increases to 90 min (Fig. 7D,J,M). Surprisingly, a subset of cells also undergoes a third division (Fig. 7M, black outlines).

We also obtained and analyzed movies of H2A-GFP transgenic embryos driving baz RNAi in the mesoderm through twi_h-Gal4. twi_h>baz RNAi embryos exhibited variations in phenotype, ‘strong’ or ‘weak’, characterized by a delayed invagination or apparently normal invagination, respectively. In addition, the ‘strong’ baz RNAi phenotype was associated with non-collective spreading and aberrant cell division (Fig. 7H,K, showing cells cross the midline; R=0.86, data not shown; Movie 3), whereas the ‘weak’ phenotype was associated with decreased spreading strength (A=1.5, Fig. S3G,H; Movie 4) but normal cell division (Fig. S3I,J). Surprisingly, in the embryo exhibiting strong Baz RNAi phenotype, cell division was non-uniform: some cells divide only once (Fig. 7N, gray circles), whereas some divide three times (Fig. 7N, black circles; as observed for ths ectopic expression, Fig. 7M). Baz expression was detected throughout the mesoderm-spreading process, and mesoderm-specific baz RNAi is associated with a decrease in the number of AJs at both stage 7 and stage 9/10 (Fig. 7E, green line-associated asterisks, P<0.05). Therefore, Baz plays multiple roles in the process of mesoderm migration: (1) it maintains or stabilizes AJs facilitating invagination; (2) it ensures a slow decrease in AJ number to support a prolonged EMT and the collectiveness of mesoderm movement; (3) it supports monolayer formation by increasing apical AJ number and reinforcing the basal Mys localization; and (4) it regulates cell division, which may be indirect and dependent on the above functions.

Previous studies had suggested that cell divisions contribute to EMT (e.g. Clark et al., 2011). Specifically, in the mesoderm, cells divide at stage 8 as the furrow collapses and mesoderm cells touch down to the ectoderm. Cell division is thought to decrease cell-cell adhesion associated with EMT, possibly by exerting a pulling force between cells as they divide. However, a potential link between mitosis and FGF signaling activity had yet to be investigated. As our tracking analysis had demonstrated that FGF regulates cell division rate, we examined more closely whether FGF mutants exhibit a cell division phenotype at gastrulation.

Using an antibody against phospho-Histone H3 (anti-PH3), it can be observed that mitosis is highly synchronized among the mesoderm cells, as indicated by their uniform staining pattern of PH3: condensed histone staining during tube collapse (Fig. 8A) and histone staining in smaller puncta during segregation and cytokinesis (Fig. 8B) (Shibata et al., 1990; Su et al., 1998). However, this synchrony in cell division is disrupted in the htl mutants (Fig. 8H,I,K,L). htl^M−Z− embryos exhibit less PH3 staining, suggesting an overall reduced mitotic activity (nine out of nine embryos, Fig. 8K,L); this is supported by the finding that these mutants sometimes contain visibly fewer mesoderm cells (see the gaps in Fig. 3K). In contrast, all cells divide upon high-level ectopic expression of Ths (seven out of seven embryos, Fig. 8E,F), consistent with the tracking analysis, which revealed an additional division in the subset of mesoderm cells in this mutant (Fig. 7J).

How FGF signaling regulates cell division to support EMT is unclear, but it has been noted that the htl^M+Z−,stg double mutant embryo exhibits a prolonged invaginated tube state (Fig. 8Y; Clark et al., 2011). Furthermore, we found that AJs, identified by α- and β-Catenin colocalization, concentrate at the center of the invaginated tube at stage 7, and likely remain in this position until a later stage when the developing midgut primordium becomes prominent (blue arrows, Fig. 8Y compare with W; X, asterisks, P<0.01; Fig. S4). Because this blocked-EMT phenotype is much more severe than either stg or htl single mutant alone, FGF signaling likely also functions through a distinct pathway rather than regulating mitosis to support EMT (Fig. 8D, see Discussion).

To provide insight into its cell division-independent role, we investigated whether FGF signaling impacts the expression of Snail, a well-known transcriptional repressor of E-Cadherin (Cano et al., 2000; Thiery and Sleeman, 2006). However, Snail protein levels in htl^M+Z−, htl^M−Z− or twi_h>ths mutants were largely comparable with those of the control embryo at stage 7 (Fig. 8C,G,J,M), indicating that the differences in the number of AJs observed in FGF loss of function or gain of function mutants (Fig. 2E,F) are not likely due to the lack of Snail.

Most studies in Drosophila have focused on how Htl FGFR activation acts through Ras/MAPK; however, FGFRs in other systems can also activate PI3K (Ornitz and Itoh, 2015). We assayed whether Ras- and PI3K-mediated intracellular pathways are responsible for carrying out different parts of the FGF functions during mesoderm development. Surprisingly, we found that rasC40B^M−Z− mutant embryos derived from germline clone females (Prober and Edgar, 2000) exhibit spreading defects that may be associated with asynchrony in cell division, in a manner that is similar to what we observed in the htl mutants (eight out of eight, Fig. 8K,L,N,O). However, rasC40B^M−Z− embryos can form a cell monolayer that acquires apicobasal polarity, as indicated by localization of Baz to the mesoderm-yolk interface (Fig. 8P,P′,R). In contrast, tissue-specific RNAi-mediated knockdown of class III PI3K Vps34/PI3K59F (Ribeiro et al., 2011) in the mesoderm has no effect on the timing of cell division (five out of five embryos, Fig. 8S,T), but does result in a dramatic reduction in apical Baz localization (Fig. 8R,U,U′), suggesting a loss of polarity. Together, our data indicate that these two roles of FGF signaling during mesoderm migration are carried out by two different intracellular pathways: a Ras-dependent activity controlling cell division to support spreading/EMT and a PI3K-dependent activity that relates to the reacquisition of cell polarity/MET.

Our study demonstrates that FGF signaling serves multiple roles during gastrulation to support this complex morphogenetic process and to direct mesoderm tissue specification. We found that FGF signaling (1) regulates cell division to support EMT, which we show does not only represent the collapse of the tube but continues throughout the mesoderm-spreading process to allow cell movements; (2) establishes cell polarity of the monolayer through control of localization of polarity proteins, including the β-PS integrin Mys to the presumptive basal side of cells; and (3), in a role possibly related to these activities above, regulates AJ number. Modifying FGF levels only led to small changes in cell number through effects on cell division, and the relatively large changes in AJ number could not relate solely to changes in cell number. Therefore, we favor the view that FGF has multiple roles, including regulation of cell division, cell polarity and AJ number. Furthermore, we propose that monolayer formation and re-epithelialization of mesoderm cells through MET is important, because it ensures that subdivision of the mesoderm occurs properly throughout the trunk, allowing accurate specification of different muscle subtypes (Fig. 6).

The role of FGF in supporting EMT is mediated by intracellular Ras signaling, whereas its role in supporting MET requires intracellular PI3K (Fig. 8N-V). The different requirement of intracellular signaling pathway effectors, known to mediate FGF signaling, supports the view that FGF has at least two functions in supporting mesoderm migration and monolayer formation at gastrulation. Furthermore, although activation of Ras has been shown to be an effector of FGF signaling in Drosophila (Carmena et al., 1998; e.g. Schulz and Gajewski, 1999), a role for PI3K in supporting FGF-regulated MET is the first characterized activity for this pathway in Drosophila, to our knowledge; however, it has been documented in the mammalian system that FGF signaling activates the PI3K pathway (Ornitz and Itoh, 2015). PI3K is also known to directly impact the specification of cell polarity and has been shown to regulate integrin trafficking and cell migration through endocytic pathways (Abe et al., 2009; Ribeiro et al., 2011). Future experiments will aim to define the molecular mechanism by which FGF activation leads to PI3K activation versus Ras; possibly this distinction relates to quality, quantity and direction of input from ligands.

Only a few studies have attempted to link FGF signaling to the establishment of cell polarity, a hallmark component of MET. In C. elegans and zebrafish, FGF indirectly regulates cell polarity by influencing Wnt signaling activity (Minor et al., 2013; Venero Galanternik et al., 2015), whereas in Ciona intestinalis, localized signaling by FGF ligands has been shown to directly control asymmetric cell divisions or convergent extension that relies on planar cell polarity components (Cota and Davidson, 2015; Shi et al., 2009). In vertebrates, ectopic activation of FGF signaling has been linked to aberrant growth, resulting in cranial malformation. This suggests that polarized activation of FGF signaling is necessary to provide directional growth (Li et al., 2013). In Drosophila embryos, FGF signaling through a different receptor, Breathless, has been shown to regulate E-cadherin turnover to support the collective movement of cells in the gut (Parés and Ricardo, 2015), and mutation in polarity proteins such as Cdc42 results in mesoderm-spreading defects (Clark et al., 2011). Our results have demonstrated a previously uncharacterized role for FGF in supporting mesoderm cell polarity, as well as the importance of the directional activation of the FGF receptor in this context, because HtlCA is not able to rescue the polarity phenotype exhibited by htl^M+Z− mutants whereas the wild-type Htl gene can (Fig. 5C,K compared with G,O; Q,R).

We also found that at the end of spreading, E-Cad and N-Cad are both expressed in the mesoderm monolayer, but in a complementary manner: E-Cad is at the basal side, whereas N-Cad is in the apicolateral domain. This complementary expression pattern of cadherins has been observed in other tissues and cell types, such as the dental epithelial cells of fish and mammals (Heymann et al., 2002; Verstraeten et al., 2013) and the ommatidia of the Drosophila retina (Chan et al., 2017; Mirkovic and Mlodzik, 2006). In the ommatidia, Wnt and EGF signaling pathways are integrated through E-Cad and N-Cad, which have opposing effects on their rotation through interactions with the cytoskeleton components. The precise cell arrangement and the coordinated cell movement require the cadherins to be expressed in the correct domain at correct level. The fact that mesoderm cells also place E-Cad and N-Cad in complementary cellular domains suggests that AJs at the apical and basal side of the mesoderm are of distinct organization and may carry out different functions, possibly interacting with different signaling pathways, such as Wnt, Dpp and FGF.

The epithelial state of the mesoderm monolayer, achieved at the end of gastrulation, has been previously underappreciated, but our analysis suggests the monolayer is important in integrating different signaling pathways and ensuring the uniform distribution of different precursor lineages derived from mesoderm. At this stage, the combination and level of Dpp, Wnt and FGF signaling received by mesoderm cells will determine what type of muscle the cells will differentiate into. Therefore, forming a monolayer facilitates effective induction of consistent number of precursors from each hemisegment. In addition, the polarized expression of adhesion molecules such as Mys, E-Cad and N-Cad also serve as signal integrators (Moser et al., 2009; Wheelock and Johnson, 2003), making the signal transduction more efficient and regulated.

Alternatively, FGF signaling is generally linked to the regulation of EMT, but this role often involves regulation of Snail expression levels (Ciruna and Rossant, 2001). We found no clear evidence that Snail expression requires FGF, but did uncover that FGF signaling regulates cell division as well as supports another undefined activity that promotes EMT: htl^M+Z−,stg double mutant phenotypes are more severe than stg mutant alone. Like Htl, mutants for the Rho GTP exchange factor Pebble also impact mesoderm cell division and spreading; however, it is thought that Pebble and FGF work in parallel pathways (Gregory et al., 2010). We show that htl^M−Z− mutants exhibit decreased proliferation, whereas upon ths ligand ectopic expression proliferation is increased (Figs 7M and 8E,F). However, htl^M+Z− ,stg double mutants exhibit stabilized AJs that appear to delay the tube collapse, retaining apical-inside organization at later stage 9/10. In addition, previous studies have shown that cell-cell junctions are mechanosensitive (Weng and Wieschaus, 2016), and therefore future studies should involve deciphering the interplay between FGF signaling, cell division and membrane tension.

We propose that regulation of cell division by FGF may be a conserved mechanism of action in which a slow, prolonged EMT is directed by this pathway in order to support collective cell migration. A gradual decrease in AJs as accomplished by cell division that is controlled by both Stg/Cdc25 and the FGF-Ras pathway may allow the mesoderm cells to maintain just enough cell-cell contacts to ensure that movement as a group is collective (reviewed by Campbell and Casanova, 2016). How migrating cells manage decisions to divide, changes in cell shape (including protrusive activity), acquire cell polarity and initiate differentiation programs remains to be determined, but more-recent studies have shown that sometimes these processes are interdependent (Matus et al., 2015) and this is also likely the case at gastrulation.

All the fly stocks and crosses were maintained at 25°C. htl-mCherry was generated through an in-frame insertion of the mCherry reporter into a htl construct that is able to rescue the mutant phenotype (Irizarry and Stathopoulos, 2015). htlAB42/TM3,ftz-lacZ (#5370), His2AV-GFP (#5961), twi-GAL4 (#914; Carmena et al., 1998; Greig and Akam, 1993; and #58804; Lee et al., 2003; Marqués et al., 2002), string7M53 (#2500; Jürgens et al., 1984), shg^GFP (#60584) UAS-baz-GFP (#29037; Benton and St Johnston, 2003), UASp-baz-mCherry (#65844; Harris and Peifer, 2005), UASp-baz-GFP (#65845; McKinley et al., 2012) and UAS-mys RNAi (#33642; Zhai et al., 2012) were obtained from the Bloomington stock center. UAS-baz RNAi (GD2914; McDonald et al., 2008) and UAS-PI3K59F RNAi (KK100962; Ribeiro et al., 2011) were from the VDRC stock center. UAS-ths [AMS289-22] (Stathopoulos et al., 2004), UAS-pyr [AMS330-3] (Kadam et al., 2009) and UAS-htlCA (Michelson et al., 1998) have been described previously. FRT101,mysXG43/FM7c,ftz-lacZ and UAS-mys,UAs-srcEGFP/CyO,ftz-lacZ were gifts from Dr Hilary L. Ashe (University of Manchester, UK) (Sawala et al., 2015; Schöck and Perrimon, 2003; Wieschaus et al., 1984). htlAB42 (Irizarry and Stathopoulos, 2015), rasC40B (Prober and Edgar, 2000) and mysXG43 (Boube et al., 2001; Leptin et al., 1989) germline clones were made using standard FRT-mediated germline clone methodology (Chou and Perrimon, 1996). htl, stg double mutants were made through homologous recombination of htlAB42 and string7M53. The twi-GAL4 driver on the first chromosome (#914) exhibits stronger expression in the embryonic mesoderm and is referred to as twi_h-GAL4; the one inserted on the third chromosome (#58804) is referred to as twi_l-GAL4. The genotype ‘control’ in this study refers to htlAB42/TM3 heterozygote, unless mentioned otherwise.

To examine the transcription level of bap and mys, antisense RNA probes were labeled with digoxigenin and recognized by anti-DIG antibody conjugated to alkaline phosphatase (1:200, Roche Applied Science). NBT and BCIP were used as substrates for signal detection. Primary antibodies used in this study were: rabbit anti-β-Gal (1:1000, Invitrogen, A-11132), goat anti-GFP (1:5000, Rockland, 600-103-215), rabbit anti-GFP (1:400, Abcam, ab6556), rabbit anti-RFP (1:400, BML PM005), mouse anti-Armadillo (1:40, DSHB, N2 7A1 Armadillo), rat anti-α-Catenin (1:40, DSHB, DCAT-1), rat anti-DE-cadherin (1:40, DSHB, DCAD2), rat anti-Twist (1:200, made in house; Trisnadi and Stathopoulos, 2014), mouse anti-Neurotactin (1:40, DSHB BP 106 anti-Neurotactin), mouse anti-even skipped (1:50, DSHB 2B8), rabbit anti-Htl (1:200, a gift from Dr Alan Michelson, Harvard Medical School, USA; Michelson et al., 1998), rabbit anti-Baz (1:500, a gift from Dr Jennifer Zallen, Sloan Kettering Institute, USA; Blankenship et al., 2006), mouse anti-βPS (1:40, DSHB, CF.6G11), rat anti-DN-Cadherin (1:200, DSHB, DN-Ex #8), goat anti-aPKC (1:200, Santa Cruz Biotechnology, C-20), guinea pig anti-Sna (1:1000, a gift from Dr Eric Wieschaus, Princeton University, USA; Weng and Wieschaus, 2016) and rabbit anti-PH3 (1:400, Millipore, 06-570). For chromogenic staining, VECTASTAIN Elite ABC system was used in combination with DAB substrate for signal development (Vector Lab). For immunofluorescence, Alexa Fluor 488, 555 and 647 secondary antibodies were used (Molecular Probes). DAPI (Molecular Probes) was used for counterstaining. Embryos stained with anti-RFP, rabbit anti-GFP, anti-Arm, anti-α-Catenin, anti-Neurotactin, anti-Baz, anti-Sna or rabbit anti-Htl antibodies were fixed using a heat-methanol method. Standard formaldehyde fixation and staining were used for other antibodies. To quantify Baz immunofluorescent staining, embryos were cleared in 70% glycerol overnight and manually cross-sectioned with a razor blade to remove the anterior and posterior regions along the AP axis. The middle regions were then mounted in 70% glycerol and imaged with LSM 800. Quantification was carried out with the 16-color color lookup table function of ImageJ. Embryos were mounted in Permount (Fisher Scientific) for whole-mount studies or embedded in acetone-araldite (Electron Microscopy Sciences) for cross-sectioning. Plastic sections (8 μm) were obtained using a microtome (LKB Bromna 2218 Historange) and mounted in 1:1 acetone:araldite solution. Fluorescent images were obtained with a LSM 800 confocal microscope (Carl Zeiss).

To compare the adhesion property of the mesoderm cells among the wild type and mutants, we used an anti-α-Catenin and an anti-Armadillo (i.e. β-Catenin) antibody simultaneously to label the AJs, and an anti-Htl antibody to determine the genotype of the embryos. Embryos from stage 7 to 10 were collected from control (htl^M+Z−/TM3), htl^M+Z−, htl^M−Z+, htl^M−Z−, twi_h>ths, twi_h>htlCA, twi_h>mys RNAi and twi_h>baz RNAi mutants, and fixed by heat. Standard immunofluorescence staining procedure was followed and all the stained embryos were manually picked, positioned on the slides that were pre-coated with heptane glue and mounted in 70% glycerol. Confocal scans of 68-100 µm along the z-axis were taken with Zeiss LSM 800. For each embryo, 8-10 z-projections through the middle one-third along the body axis from its 3D reconstructed image were collected and used to count the AJs. To count the number of AJs in the mesoderm on one z-projection, we used Photoshop (Adobe) to mark the boundary between mesoderm and ectoderm, increase the contrast and lower the exposure offset to make the white colocalization signals between α-Catenin (violet) and Armadillo (green) more visible. The microscope settings were kept the same as much as possible, except that more laser power were used in mutant embryos with a taller furrow. All the images were processed the same way. A mean of AJs numbers in the mesoderm was first calculated from all the cross-sections for each embryo. Then, an average (as the mean of the group) and a standard deviation (s.d.) were calculated for the embryos that were pooled together based on their genotypes and developmental stages. Stage 9 and 10 embryos were combined. Error bars represent s.d. Shapiro-Wilk Normality test was used to test for normal distribution of the means within the group. In the cases with fewer than five individuals (n<5) in the group, we performed the test on the AJs number counted from all the images acquired for the embryos in the group. Our data exhibited normal distribution. Student's t-test was used for significance analysis. P<0.05 was considered statistically significant. Number of embryos for each genetic background at stages 7, 8 and 9-10 are as follows, respectively: control, 5, 5 and 3; htl^M+Z−, 3, 7 and 6; htl^M−Z−, 6, 4 and 4; htl^M−Z+, 4, 3 and 3; twi_h>ths, 1, 4 and 3; twi_h>htlCA, 2, 3 and 3; twi_h>mys RNAi, 3, 3 and 4; twi_h>baz RNAi, 4, 6 and 5.

To analyze chromogenic antibody staining of Baz, Mys, aPKC and E-Cad, each genetic background (normally 50-120 embryos at stages 6-10) were embedded for plastic sectioning. Poorly stained embryos, undeveloped embryos or embryos with morphological defects that are clearly not due to the genetic manipulations were excluded from our analysis. Sections mounted unevenly that were difficult to focus on during imaging were excluded. Bright-field images were photographed with a Zeiss Axio Imager under similar settings. A minimum of five embryos or six sections at each stage were imaged so that normality could be tested. In cases where fewer than five embryos were recovered, we were very confident in the consistency of the phenotype. Images were converted to gray scale, reversed and measured in ImageJ. To calculate the relative enrichment, boxes (ROI) of the same size were placed in the mesoderm and the ectoderm, mean gray value in the mesoderm ROI was divided by the mean gray value in the ectoderm ROI, and the quotient was recorded as one data point. First, an average enrichment was calculated for each embryo. Then the averages and the standard deviations (s.d.) were calculated within the same genetic background. Error bars represent s.d. The Shapiro-Wilk Normality test was used to test for normality. Student's t-test was used for significance analysis. Fluorescent anti-Nrt images were analyzed by comparing the percentage of polarized mesoderm cells in the control and mutant embryos. Colocalization of α- and β-Catenin in stg and htl^M+Z−,stg embryos was also quantified by measuring the mean gray value in the ROI. Similarly, the quotients were compared between the two lines.

Two twi_h>ths, twi_h>baz RNAi and yw embryos were imaged as previously described using a Zeiss LSM 710 inverted microscope at 940 nm wavelength (McMahon et al., 2008; Supatto et al., 2009). Nuclear tracking was performed using Imaris (Bitplane). Imaris data were then exported to Matlab (The Mathworks) for analysis using custom scripts (Supatto et al., 2009). Briefly, distributions of the ectoderm cells were fitted into a cylinder in order to register the 3D positional tracking data in cylindrical coordinates (r, θ and L), which allows the analysis of cell movement along the corresponding body axis of the embryo (Fig. 7A,B). Two different color codes were applied to show the organization of the mesoderm cells (Fig. 7C,D). In a spreading plot θ (t) (Fig. 7F-H), color code marks the angular position of cells in the furrow at stage 7, in order to better depict their spatial displacement over time. t=0 (j0) represents the onset of germ band elongation. Start (jstart) is the time when the furrow forms the best cylindrical shape, and end (jend) is defined as 120 min after start. For cell division analysis, cells are color coded for their radial position in the furrow to show the difference in division time according to the distance from the ectoderm (Fig. 7I-N). The spreading profile or the θ_end (θ_start) plot shows how well the mesoderm spreads as a collective, in which the angular position of each cell at the beginning and the end of the spreading process is fitted into a regression line: θ_end=A.θ_start+B. The slope A characterizes the general spreading behavior and should be close to 2 in the wild-type embryos. The coefficient R should be close to 1 when the cells are migrating as a collective.

We are grateful to H. Ashe, A. Michelson, E. Wieschaus and J. Zallen for sharing fly stocks and antibodies, and to Z. Ákos, F. Macabenta, A. McMahon, W. Supatto and N. Trisnadi for help with Drosophila genetics, sharing preliminary results and antibodies, imaging and tracking, comments on the manuscript, and modifications to scripts as well as helpful discussions.