Differential adhesion and actomyosin cable collaborate to drive Echinoid-mediated cell sorting

Cell sorting is a process in which two cell populations are smoothly partitioned into distinct adjacent tissues, similar to the phase separation between immiscible fluids. The establishment of discrete cellular compartments is key to a variety of developmental processes in different organisms, including gastrulation, neural tube formation and various types of organogenesis (Lecuit and Lenne, 2007; Perez-Pomares and Foty, 2006; Tepass et al., 2002). Cell sorting was first demonstrated when, upon mixing, dissociated presumptive neural and epidermal cells of amphibian gastrulas moved and sorted into separate tissues (Townes and Holtfreter, 1955). Cell sorting also refers to the process whereby cells in different compartments are segregated and form a smooth boundary at the compartment border (Tepass et al., 2002); for example, cells of the anterior and posterior as well as dorsal and ventral compartments of the Drosophila wing imaginal disc cannot intermix.

Several mechanisms have been proposed to explain the behavior of cell sorting, including differential adhesion, differential contractility and interfacial actomyosin cable formation (Martin and Wieschaus, 2010). The first hypothesis proposes that tissues with quantitatively different cell-cell adhesions have different resultant tissue surface tensions and, like dissimilar liquids, are immiscible (Steinberg, 2007). In support of this position, it has been shown that the differential expression of classic cadherins regulates the positioning of cells in the mouse telencephalon (Inoue et al., 2001) and the segregation of neurons into different motoneuron pools in the chick spinal cord (Price et al., 2002). In the second hypothesis, different levels of cell contractility are suggested to lead to differences in cell tissue surface tension and, thereby, immiscible cell populations (Harris, 1976); for example, differential contractility has been implicated in the segregation of germ layers in zebrafish (Krieg et al., 2008). The third proposal suggests that tension generated by interfacial actomyosin cables acts as a mechanical fence to prevent cell mixing, such as along the anteroposterior and dorsoventral boundaries in Drosophila (Landsberg et al., 2009; Major and Irvine, 2005; Major and Irvine, 2006; Monier et al., 2010).

Cadherin is the major cell adhesion molecule of adherens junctions (AJs), which mediate cell-cell adhesion/recognition. The intracellular domain of cadherin, via association with β-catenin, binds α-catenin, which in turn links to actomyosin networks. shotgun (shg) encodes Drosophila E-cadherin (DE-cad). Echinoid (Ed), a nectin homolog, also localizes to AJs and, via association with Canoe (Afadin), links to actomyosin (Wei et al., 2005). A common feature of these two homophilic cell adhesion molecules (CAMs) is that both ed and shg mutant clones exhibit rounded, smooth contours and sort out from surrounding wild-type cells (Laplante and Nilson, 2006; Le Borgne et al., 2002; Wei et al., 2005). This is in contrast to clones of wild-type cells, which show jagged borders. Ed-mediated cell sorting is implicated in driving various epithelial morphogenetic processes, including epithelial tube formation and dorsal closure (Laplante and Nilson, 2006; Lin et al., 2007). Ed-mediated cell sorting exhibits four characteristics (Wei et al., 2005). First, ed is differentially expressed in ed mutants and surrounding wild-type cells. Second, ed mutant cells accumulate a higher density of DE-cad/Armadillo (Arm; a β-catenin homolog) and develop an apical constriction (Fig. 1B). Third, ed mutant cells fail to form AJs with wild-type cells (Fig. 1B, arrowheads). Finally, surrounding wild-type cells confine ed mutant cells by assembling a ring of actomyosin cable (Fig. 1B′, arrow). However, it is currently unknown which factors are required to facilitate Ed-mediated cell sorting.

In this report, we dissect the sequence of these characteristic events during Ed-mediated cell sorting by generating ed-RNAi clones. We show that, following the reduction of Ed levels at the interface, the wild-type interface cells gradually assemble an actomyosin cable and form a smooth boundary. Although there was differential Ed expression, the recruitment of Ed (via homophilic and heterophilic binding) to the wild-type cell side of the clonal interface was sufficient to prevent actomyosin cable formation and thereby rescue cell sorting. Significantly, when actomyosin cable was largely absent, differential adhesion mediated by differential expression of the extracellular domain of Ed was sufficient to cause cell segregation but with jagged borders. Thus, both differential adhesion and actomyosin cable are required and collaborate to drive Ed-mediated cell sorting.

The following stocks were used: C765-Gal4, C381-Gal4, en-Gal4, ap-Gal4, tubulin-Gal80^ts (Bloomington Stock Center), ed^1X5, UAS-ed (Bai et al., 2001), UAS-Ed^Δintra-Flag (this study), UAS-ed^extra-RNAi (VDRC), UAS-ed^intra-RNAi (this study), UAS-Ed^TM-cad^intra (this study), UAS-Fred-Flag (this study), UAS-Nrg-Ed-Flag (this study), tubulin-DE-cad (Pacquelet et al., 2003), and P[act5C>y+>Gal4] P[UAS-GFP.S65T]/CyO (Ito et al., 1997).

To generate ed^1X5 clones when endogenous DE-cad was largely downregulated, we generated hsFLP;FRT^40A ed^1X5/FRT^40A ubi-nls-GFP; C765-Gal4/UAS-Ed^TM-cad^intra larvae. To generate UAS-Ed^Δintra-Flag twinspot clones in the dorsal compartment where Ed was depleted, we generated hsFLP; ap-Gal4,UAS-ed^intra-RNAi; FRT^82B UAS-Ed^Δintra-Flag/FRT^82B ubi-nls-GFP larvae.

UAS-ed^intra-RNAi was generated by subcloning exon 7 of ed as an inverted repeat, followed by ligation into the pMF3 vector (Dietzl et al., 2007). UAS-Ed^TM-cad^intra was generated by ligating two overlapping PCR fragments, with the first PCR fragment containing the transmembrane domain of Ed and a second PCR fragment containing the intracellular domain of DE-cad, followed by ligation into pUAST. UAS-Nrg-Ed-Flag was generated by ligating two PCR fragments, with the first PCR fragment containing the fibronectin type III and transmembrane domain of Ed (with a 3′ FLAG tag) and a second PCR fragment containing the five Ig domains of Nrg, followed by ligation into pUAST. UAS-Fred-Flag was generated by inserting the fred PCR fragment together with a 3′ FLAG tag into pUAST. UAS-Ed^Δintra-Flag was generated by inserting a FLAG tag into the 3′-end of UAS-Ed^Δintra (Bai et al., 2001). pMT-Ed, pMT-Fred-Flag and pMT-Ed^Δintra-Flag plasmids were generated by respectively subcloning the ed, Fred-Flag and Ed^Δintra-Flag PCR fragments into the pMT vector (Invitrogen).

For the endocytosis assay, third instar larvae (en-Gal4; Ed^TM-cad^intra/tub-Gal80^ts) were incubated at 29°C for 7 hours to induce Ed^TM-cad^intra expression. After induction, wing discs were dissected in M3 medium supplemented with 10% fetal calf serum. Discs were pulse labeled with rat anti-DE-cad for 30 minutes in ice-cold medium. After washing three times in cold M3 medium, the discs were transferred to M3 medium at 29°C for 30 minutes, followed by fixation in 4% paraformaldehyde.

For the S2 aggregation assay, S2 cells were transfected in 6-cm diameter dishes using TurboFect Reagent (Fermentas) with 4 μg pMT-Ed, pMT-Fred-Flag or pMT-Ed^Δintra-Flag vector. Transfected cells were induced overnight with 0.7 mM CuSO₄ and mixed, followed by incubation on a rotating shaker (20 rpm) for 4 hours at room temperature.

Immunostaining was performed as described (Wei et al., 2005). Antibodies used were: guinea pig anti-Ed [1:200 (Wei et al., 2005)], rat anti-Ed [1:200 (Ho et al., 2010)], rat anti-DE-cad (DCAD2, 1:50; DSHB), guinea pig anti-Fred [1:200 (Fetting et al., 2009)], mouse anti-Arm (N2-7A1, 1:40; DSHB), rabbit anti-pSer19-MLC (1:10; Cell Signaling Technology), mouse anti-FLAG (1:500; Sigma), Alexa 594-phalloidin (1:200; Invitrogen) and Cy3- and Cy5-conjugated secondary IgGs (1:500; Jackson ImmunoResearch Laboratories). After staining, samples were mounted in 20% glycerol and nail polish was used to seal the sections. Images were acquired using 63× NA1.4 oil-immersion Plan-Apochroma objectives on a confocal microscope (LM510, Carl Zeiss).

ImageJ software was used for quantification of apical area and of Ed and F-actin (pixel intensity). Cell bonds, connected at vertices along clone interfaces, were used to measure circularity and bond angles (determined manually) with ImageJ.

ed^1x5 null allele clones sorted out from the surrounding wild-type cells and exhibited both rounded and smooth contours (Fig. 1B). This differs from cells of the anterior and posterior compartments, which segregate and form a smooth boundary, but resembles phase separation between immiscible fluids that tend to minimize their contact surface (with the shortest interface) and have rounded and smooth interfaces (Fig. 1C). We therefore used circularity, C=4π × area/(perimeter²), where perimeter is the length of clonal interface, to measure the relative roundness and smoothness of a clone (Fig. 1C) (Lawrence et al., 1999). For clones of the same apical area, that with the shortest interface (i.e. cells are immiscible) becomes a circle (C=1), whereas that with the longer interface (i.e. cells are more miscible) will deviate more from a circle and adopt a more jagged interface (Fig. 1C). Moreover, we also used the average of bond angle variation (⁠$Δθ‒$⁠, the average of bond angle differences between two adjacent bond angles) as an alternative measurement of the smoothness (but not roundness) of an interface. For a straight interface, the $Δθ‒$=0.

For ed^1x5 mutant clones, C=0.94±0.03 and $Δθ‒$ =29.18±9.04° (n=20 clones), in contrast to control clones (Fig. 1A) where C=0.32±0.10 and $Δθ‒$ =88.80±16.21° (n=21 clones) (Fig. 1D,E). ed mutant cells also developed an apical constriction, with the apical area positively correlated with clone size (Fig. 1F). For example, the apical area was reduced to only 62±11% (n=16 clones) of that of normal cells when the clone size was less than 50 cells, whereas the apical area was 104±10% (n=8 clones) of that of normal cells when the clone size exceeded 150 cells (Fig. 1F). Moreover, the surrounding wild-type cells assembled a ring of actomyosin cable at the interface, with an F-actin intensity 242±28% (n=6 clones) of that of normal interfaces (Fig. 1G).

We used two different ed-RNAi constructs to silence ed: ed^extra-RNAi and ed^intra-RNAi against the extracellular domain and intracellular domain sequences of ed, respectively. Ed antibody staining confirmed that both constructs specifically inhibited ed expression (Fig. 1H,I). Flip-out clones overexpressing the ed^extra-RNAi or ed^intra-RNAi transgene (referred to as ed-RNAi clones) in the larval wing imaginal disc exhibited similar levels of circularity (Fig. 1D), average bond angle variation (Fig. 1E), apical constriction (Fig. 1F) and actomyosin cable formation (Fig. 1G) at the clonal boundary as ed^1x5 mutant clones. Moreover, ed-RNAi clones also exhibited a lack of DE-cad at some of the clonal interfaces (Fig. 1H). We confirmed that, in wing discs, the actomyosin cable was assembled by the wild-type interface cells (Fig. 1I′, arrowhead) and was enriched in phosphorylated myosin light chain (p-MLC) (Fig. 1J), indicating that wild-type cells actively prevent mixing with the ed mutant cells (see below).

To distinguish the respective contributions of the four characteristics mentioned above in Ed-mediated cell sorting, we dissected the sequence of these events and determined which event led to the onset of smooth boundary formation and therefore effective cell sorting. We combined the flip-out clone with the temperature-sensitive tub-Gal80^ts driver to generate ectopic ed^extra-RNAi clones at permissive temperature (25°C) and then shifted them to restrictive temperature (29°C) for 10-22 hours to drive ed^extra-RNAi expression to differentially reduce the levels of Ed within the ed^extra-RNAi clones. Notably, ed^1x5 mutant clones (starting from one ed^1x5 mutant cell) tended to be round or oval in shape, whereas ed-RNAi clones generated by this protocol (using Gal80^ts), although with a smooth boundary, were not necessarily oval (Fig. 2B,C). It has been shown elsewhere that cell movement is insignificant in the wing disc (Gibson et al., 2006). Thus, the discrepancy in the clone shapes could be caused by the late induction of ed-RNAi expression after the formation of a jagged-bordered clone and the limited mobility of the ed-depleted cells restricted their ability to rearrange themselves into oval-shaped clones. We refer to cell sorting in this section (with reference to Fig. 2 only) as the process whereby ed-RNAi cells segregate from, and form a smooth border with, wild-type cells (by measuring $Δθ‒$⁠), without considering the effect on clone shape, whereas the formation of rounded clones with a smooth border is a criterion of cell sorting in the remainder of the manuscript (by measuring C and $Δθ‒$ and comparing them with those of control and ed^1x5 clones).

By shifting ed^extra-RNAi clones to the restrictive temperature for 10 hours, we observed that Ed was partially depleted in ed^extra-RNAi clones, thus establishing differential Ed expression (Fig. 2A). However, similar to wild-type control clones, the clonal boundary was jagged (⁠$Δθ‒$=80.10±5.37°; n=8 clones) and DE-cad remained at the interface (Fig. 2A′, white lines) in these clones. Moreover, we failed to observe apical constriction of ed^extra-RNAi cells (Fig. 2A; apical area was 103±22% of that of normal cells; n=10 clones) and actomyosin cable formation (F-actin intensity was 102±13% of that of normal cells; n=6 clones). These data suggest that cell sorting had not occurred despite the obvious differential Ed expression.

Next, by shifting to restrictive temperature for 12 hours, we observed clearer differential Ed expression, but the depletion of ed in ed^extra-RNAi clones was not uniform (Fig. 2B). We found that the levels of Ed left at the interface negatively correlated with the levels of actomyosin cable observed (Fig. 2B,D), and that the levels of actomyosin cable at the interface were also negatively correlated with the average bond angle variation (Fig. 2E). Thus, following the reduction of Ed levels at the interface, the wild-type interface cells gradually assembled an actomyosin cable and formed a relatively smooth boundary. Of note, a weak apical constriction was also detected (Fig. 2B′,C′; apical area was 87±21% of that of normal cells; n=9 clones). However, DE-cad still remained on the interface (Fig. 2C) and the levels of DE-cad did not significantly change (see Fig. S1 in the supplementary material) even when Ed levels were reduced. Thus, the appearance of actomyosin cable and an apical constriction, but not the disappearance of DE-cad at the interface, were closely associated with the formation of a smooth interface.

Shifting to restrictive temperature for 22 hours completely depleted Ed in ed^extra-RNAi clones, and Ed-mediated cell sorting became very prominent (⁠$Δθ‒$=44.95±17.30°; n=10 clones). Similar to ed^1x5 mutant clones, we observed: (1) thick actin cable surrounding the entire interface (Fig. 2F′; F-actin intensity was 203±19% of that of normal cells; n=10 clones); (2) more obvious apical constriction of ed^extra-RNAi cells, especially in small rounded clones (Fig. 2G; apical area was 67±6% of that of normal cells; n=10 clones); and (3) a disappearance of DE-cad from some interfaces (Fig. 2G′, arrowhead). Thus, when the levels of Ed at the interface were further reduced, the ed clonal phenotype became more prominent and, importantly, the disappearance of DE-cad at the interface occurred late during this process (Fig. 2H).

Although apical constriction was detected in ed-depleted cells when a large number of wild-type cells surrounded a few ed-depleted cells (in a small ed-RNAi clone, Fig. 1H,I), apical constriction was also detected in wild-type cells when a large number of ed-RNAi cells surrounded a few wild-type cells (in a very large ed-RNAi clone, Fig. 2I). Moreover, apical constriction became less severe when ed^1X5 clones expanded (Fig. 1F) (Wei et al., 2005) or when more wild-type cells were surrounded by a large number of ed-RNAi cells (compare the left and right groups of wild-type cells in Fig. 2I). Importantly, p-MLC was enriched at the clonal interface; however, we did not detect significant p-MLC accumulation in the apically constricted ed-RNAi cells even when the clones were small and exhibited strong apical constriction (Fig. 1J). Together, our data indicate that apical constriction is primarily, if not exclusively, a consequence, but not the cause, of cell sorting.

We showed that the disappearance of DE-cad at the interface was a late event and was, therefore, not crucial in Ed-mediated cell sorting. However, DE-cad accumulated in the apically constricted ed-RNAi cells, which, in turn, generated differential DE-cad expression across the ed clonal boundary (with higher levels in the ed-RNAi clone and lower levels in the surrounding cells). To determine whether differential DE-cad expression across the ed clonal boundary contributes to Ed-mediated cell sorting, we analyzed ed^1x5 clonal phenotypes when DE-cad was removed from the epithelial cells to abolish the differential DE-cad expression. As Ed and the cell polarity protein Bazooka (Baz) mislocalize in shg^IG29 null mutant cells, this indicates that DE-cad is required for the establishment/maintenance of apical-basal polarity (Wei et al., 2005). To devise a system with strongly reduced DE-cad (to abolish the differential DE-cad expression) but without complete disruption of the epithelium, we took advantage of an Ed^TM-cad^intra chimeric construct (with the transmembrane domain of Ed and the intracellular domain of DE-cad). We found that overexpression of Ed^TM-cad^intra, via en-Gal4, enhanced DE-cad endocytosis in the posterior compartment (Fig. 3A). Moreover, flip-out clones overexpressing Ed^TM-cad^intra downregulated endogenous DE-cad and caused cell sorting (Fig. 3B), similar to shg^IG29 mutant clones (Wei et al., 2005). However, unlike shg^IG29 mutant cells, which show no Arm association, Ed^TM-cad^intra-expressing cells accumulated high levels of Arm, which was probably caused by the presence of an Arm binding site in Ed^TM-cad^intra (Fig. 3C).

In addition to F-actin, several polarity markers, including Ed and aPKC, were localized normally on the Ed^TM-cad^intra-expressing cells (Fig. 3D,E), indicating that apical-basal polarity was largely maintained. We therefore used C765-Gal4 to ubiquitously overexpress UAS-Ed^TM-cad^intra to downregulate DE-cad (but maintain Ed, Arm and F-actin distribution) in the wing disc cells and then generated ed^1X5 clones in this background (for genotype, see Materials and methods). As shown in Fig. 3F, DE-cad was undetectable in cells of the ed^1X5 clone (although low levels accumulated at the tricellular junctions of the surrounding wild-type cells); however, under this condition, ed^1X5 clones still formed a smooth boundary (Fig. 3F and quantification in Fig. 1D,E) with actomyosin cable formation, similar to ed^1X5 clones generated in the presence of DE-cad (arrowhead in Fig. 3G′ and quantification in Fig. 1G).

Together, our data indicate that when cells possess normal cell polarity with proper localization of Ed, Arm and F-actin, differential adhesion by the DE-cad extracellular domain or differential expression of DE-cad across the ed clonal boundary was not required for Ed-mediated smooth boundary and actomyosin cable formation. Significantly, in this case ed^1x5 cells also exhibited obvious apical constriction (Fig. 3G′ and quantification in Fig. 1F), indicating that the large reduction of DE-cad-mediated adhesion or the large reduction of DE-cad (to undetectable levels) did not affect Ed-mediated apical constriction (Fig. 3G). However, we cannot exclude the possibility that a residual, very low level of DE-cad might be sufficient to contribute to Ed-mediated apical constriction.

Although we showed above that differential expression of full-length Ed triggered the assembly of actomyosin cable, which in turn led to the formation of a smooth boundary (Fig. 2D,E), differential Ed expression also causes differential Ed adhesion. It is unclear whether differential Ed expression acts solely via inducing actomyosin cable formation to drive cell sorting or, alternatively, if differential Ed adhesion also contributes to cell sorting (see below).

Thus far, we have shown that actomyosin cable assembles in the wild-type interface cells when differential expression of Ed occurs. To confirm this inverse correlation between Ed levels and actomyosin cable assembly, we performed rescue experiments. We found that the ed clonal phenotype could be fully rescued by the ubiquitous expression of UAS-ed using the driver C765-Gal4. The clones became jagged-edged and cells within the clones possessed normal levels of DE-cad (Fig. 4B). Although both Ed and DE-cad localize to AJs and might have overlapping functions (Wei et al., 2005), ubiquitous expression of DE-cad (using tubulin-DE-cad) failed to rescue the ed clonal phenotype, indicating that Ed and DE-cad are not redundant in cell recognition (Fig. 4C; data not shown).

Neuroglian (Nrg) possesses five Ig domains and heterophilically trans-interacts with Ed (Islam et al., 2003). However, unlike Ed, Nrg localizes to septate junctions of the wing disc cells (Genova and Fehon, 2003). We therefore generated Nrg-Ed-Flag by replacing the seven immunoglobulin (Ig) domains of Ed with the five Ig domains of Nrg, with the expectation that Nrg-Ed-FLAG might localize to the AJs and trans-interact with Ed. When ectopic clones overexpressing both ed^intra-RNAi and Nrg-Ed-Flag were generated, we found that Nrg-Ed-FLAG did localize to AJs in the same manner as Ed (Fig. 4D′). However, we could not detect any Ed recruited to the interface of wild-type cells (Fig. 4D), although some Nrg-Ed localized to the clonal interface (Fig. 4D′, arrowhead). This indicates that any interaction between Ed and Nrg-Ed was insufficient to recruit Ed to the interface. As expected, actomyosin cable formed and smoothed the clone boundary (Fig. 4E and quantification in Fig. 1D,E,G).

Next, we determined whether the overexpression of Ed^Δintra-Flag (comprising the transmembrane domain and the entire extracellular domain of Ed) could rescue the ed clonal phenotype. We generated ectopic clones overexpressing both ed^intra-RNAi and Ed^Δintra-Flag (Fig. 4F,G). We found that Ed was recruited to the clonal interface (Fig. 4F,G, arrowhead; Ed recruitment was 106±10% of that of normal cells; n=10 clones). Moreover, some Ed^Δintra-FLAG also accumulated at the clonal interface (Fig. 4F′, arrowhead), although most Ed^Δintra-FLAG remained localized to the interfaces between cells within the Ed^Δintra-Flag-overexpressing clones (Fig. 4F′, arrow). Similarly, Ed and Ed^Δintra colocalized at the interface between S2 cells overexpressing Ed^Δintra and those overexpressing ed (Fig. 4H). Our results together suggest that Ed can form trans dimers with Ed^Δintra (although Ed^Δintra prefers to form trans homodimers). Importantly, no actomyosin cable was present at interfaces (Fig. 4G′ and quantification in Fig. 1G) and the clonal boundary was jagged (Fig. 4G and quantification in Fig. 1D,E). Thus, our data suggest that the adhesion between Ed^Δintra on the ed mutant cells and full-length Ed on the wild-type cells was sufficient for the rescue of cell sorting. This is in sharp contrast with DE-cad, as only full-length DE-cad, but not DE-cad^Δintra (the transmembrane and the entire extracellular domain of DE-cad), can rescue the smooth contour of shg clones (Pacquelet et al., 2003). Thus, Ed-mediated and DE-cad-mediated cell sorting probably involve different molecular mechanisms.

Friend of Echinoid (Fred) possesses seven Ig domains and shares 69% overall sequence identity with the extracellular domain of Ed (Chandra et al., 2003). When we generated ectopic clones overexpressing both ed^intra-RNAi and Fred-Flag, we found that both Ed and Fred-FLAG were recruited to the clonal interface (Fig. 4I, arrowhead; Ed recruitment was 104±9% of that of normal cells; n=10 clones), indicating that Ed can heterophilically form trans dimers with Fred. Consistent with this, Ed and Fred colocalized to the interface between S2 cells overexpressing ed and those overexpressing fred (Fig. 4K, arrowheads). Similar to Ed^Δintra-FLAG, most Fred-FLAG localized to the interfaces between cells within the Fred-Flag-overexpressing clones (Fig. 4I′, arrow); thus, Fred-FLAG also prefers to form trans homodimers rather than Ed-Fred-FLAG heterodimers. Importantly, no actomyosin cable was present at the interface (Fig. 4J and quantification in Fig. 1G) and the clonal boundary was jagged (Fig. 4J and quantification in Fig. 1D,E). Thus, Ed-mediated cell sorting can be rescued by recruitment of Ed, via heterophilic interactions, to the wild-type cell side of the clonal interface, even when differential Ed expression persists. As fred mRNA is detected in the wing disc (Chandra et al., 2003), the failure of endogenous fred to rescue ed mutant clones indicates that the levels of endogenous Fred might be low and only upon ectopic overexpression can Fred heterophilically interact with sufficient amounts of Ed to recruit it to the interface. Consistent with this, using Fred antibody, we only detected Fred signals in fred-overexpressing but not wild-type cells (see Fig. S2 in the supplementary material). Altogether, our data indicate the importance of the presence of Ed at the clonal interface of wild-type cells to prevent actomyosin cable formation.

We established that the differential expression of full-length Ed led to the formation of actomyosin cable and cell sorting. We were also interested in whether differential expression of Ed^Δintra (the transmembrane and extracellular domains of Ed) alone would be sufficient to cause cell sorting. We generated UAS-Ed^Δintra-Flag twinspot clones in the dorsal compartment (where ap-Gal4 is expressed) to drive differential Ed^Δintra expression. We then examined the interface between cells without Ed^Δintra-Flag (with 2× ubi-nls-GFP) and cells with 2× ap>Ed^Δintra-Flag (lack of ubi-nls-GFP) or 1× ap>Ed^Δintra-Flag (with 1× ubi-nls-GFP). However, as full-length Ed binds Ed^Δintra (Fig. 4F,G), the endogenous Ed on Ed^Δintra-deficient cells could still interact with Ed^Δintra on the Ed^Δintra-containing cells (1× or 2×). This would have complicated the results, so we generated instead UAS-Ed^Δintra-Flag twinspot clones in the dorsal compartment where Ed was depleted by ap-Gal4-driven UAS-ed^intra-RNAi (Fig. 5A-D; for genotype see Materials and methods). Unexpectedly, we observed that the interface cells (with 2× or 1× Ed^Δintra-Flag) largely failed to assemble prominent actomyosin cable (Fig. 5B,C and quantification in Fig. 1G; F-actin intensity was 117±7% of that of normal cells; n=10 clones), as compared with the interface cells of ed^1X5 clones (Fig. 1B and quantification in Fig. 1G; F-actin intensity was 242% of that of normal cells). Moreover, in contrast to the ed-RNAi clone, p-MLC was not detected at this interface (compare Fig. 5D with Fig. 1J). These data are similar to those of a recent report showing that actomyosin cable and p-MLC were not detected at the interface between ed (without Ed) and ed; Ed-ΔC (without Ed but with Ed^Δintra) follicle cells (Laplante and Nilson, 2011). Thus, our results indicate that this actomyosin cable, if present, must be very weak. Moreover, the presence of the intracellular domain of Ed is important for prominent actomyosin cable formation in the interface cells.

As mentioned above, it is not yet clear whether differential Ed expression acts solely by inducing actomyosin cable assembly or whether differential Ed adhesion is involved in driving cell sorting. The inability to form a prominent actomyosin cable in the Ed^Δintra-containing interface cells thus provided an opportunity to further evaluate the contribution of differential Ed adhesion. When actomyosin cable was largely absent, Ed^Δintra-deficient cells formed a coherent group, with C=0.73±0.11 (n=20 clones; lines in Fig. 5B′′′,C′). Thus, the circularity of Ed^Δintra clones was closer to that of the ed mutant clones (C=0.94±0.03, Fig. 1D) but deviated more from the control clones (C=0.32±0.10), indicating that Ed^Δintra clones sort out, to some degree, from the Ed^Δintra-containing cells. However, the average bond angle variation of Ed^Δintra clones (⁠$Δθ‒$=68.95±17.22°; n=20 clones) was closer to that of control clones (⁠$Δθ‒$=88.80±16.21°; Fig. 1E) than to that of ed mutant clones (⁠$Δθ‒$=29.18±9.04°), indicating the presence of jagged borders. Together, these results show that Ed^Δintra clones exhibit effects intermediate between ed^1x5 and control clones (with some cell segregation but a jagged tissue border), and we regarded this as imperfect sorting.

As we cannot completely exclude the possibility that F-actin intensity at this low level (117±7% of that of normal cells) might affect, to a small extent, the circularity of Ed^Δintra clones, we suggest that the segregation of cells into two immiscible populations, as detected here, is mainly, if not exclusively, caused by differential adhesion of Ed^Δintra; however, differential adhesion of Ed^Δintra is not sufficient to generate a smooth border. Of note, Ed^Δintra clones also exhibited strong apical constriction (Fig. 5B,C and quantification in Fig. 1F). Thus, apical constriction occurred even when actomyosin cable was largely absent. Together, our results suggest that differential expression of Ed in a group of cells, via differential Ed adhesion, triggers the segregation of cells into separate populations with a jagged border and also induces the formation of actomyosin cable at the interface, which in turn exerts tension to form a rounded clone with a smooth boundary. Thus, both differential adhesion and the induction of actomyosin cable formation are required and act cooperatively to mediate proper cell sorting.

Here, we dissect the sequence of events in Ed-mediated cell sorting and conclude that both differential adhesion and the induction of actomyosin cable formation are required and act cooperatively to mediate cell sorting. We also demonstrated that the relocalization of Ed by Ed, Fred and Ed^Δintra, but not Nrg-Ed, to the clonal interface of the wild-type cells is sufficient to prevent actomyosin cable formation in the wild-type cells. How differential expression of Ed induces actomyosin cable formation only at the Ed⁺ interface cells (but not the Ed^– cells) to generate a polarized response remains unknown. It has been suggested that interfacial tension is the result of cortical tension decreased by adhesion energy at this interface (Kafer et al., 2007; Krieg et al., 2008; Manning et al., 2010). Moreover, cortical myosin II recruitment is regulated by tension in a positive-feedback loop that could promote actomyosin cable formation (Fernandez-Gonzalez et al., 2009). Therefore, we postulated that the reduction of adhesion energy caused by the loss of Ed would increase the interfacial tension so as to induce actomyosin cable formation at that interface. However, although interfacial tension also increases in ed mutant cells we did not detect prominent actomyosin cable formation in these cells. Thus, interfacial tension alone is insufficient to explain this polarized effect.

Laplante and Nilson (Laplante and Nilson, 2011) proposed that, during dorsal closure, asymmetric distribution of Ed is required in the dorsal-most epidermal (DME) cells for the polarized accumulation of actin regulators (such as Enabled, Diaphanous and RhoGEF2) in the actin-nucleating centers (ANCs) of DME cells, and that this in turn promotes actomyosin cable assembly at the leading edge. Ed-mediated cell sorting resembles embryonic dorsal closure, where the DME cell is equivalent to the Ed⁺ interface cell, the leading edge is equivalent to the interface of ed mutant clones, and the ANC is equivalent to the interfacial tricellular junction of Ed⁺ interface cells. We also found a similar polarized accumulation of actin regulators, such as Enabled, at the tricellular junction of Ed⁺ interface cells of ed-RNAi clones (data not shown). As cells within ed mutant clones cannot have a polarized distribution of Ed and actin regulators to form actomyosin cable, this provides an alternative mechanism for generating a polarized effect. Moreover, as both Nilson's group and our group demonstrated that the intracellular domain of Ed is required for actomyosin cable formation, it is possible that the asymmetric distribution of Ed might, via its intracellular domain, regulate the polarized accumulation of actin regulators at the interfacial tricellular junctions of Ed⁺ cells that in turn promotes actomyosin cable assembly.

The induction of actomyosin cable formation only at the Ed⁺ interface cells was observed not only when a large number of wild-type cells surrounded a few ed-depleted cells (in a small ed-RNAi clone, Fig. 1I,J) but also when a large number of ed-RNAi cells surrounded a few wild-type cells (in a very large ed-RNAi clone, see Fig. S3A,B in the supplementary material). The actomyosin cable at the interface supplies the tension needed to form a smooth border, tension that can be supplied either by the Ed⁺ interface cells surrounding a small ed-RNAi clone or by the Ed⁺ interface cells within a large ed-RNAi clone. However, apical constriction was present in ed-depleted cells surrounded by a large number of wild-type cells (in small ed-RNAi clones). Similarly, apical constriction was also detected when a few wild-type cells were surrounded by a large number of ed-RNAi cells (in large ed-RNAi clones, Fig. 2I). As we did not detect significant p-MLC accumulation in the apically constricted ed-RNAi (Fig. 1J) or wild-type (see Fig. S3 in the supplementary material) cells, we suggest that myosin-mediated contraction is not important in the generation of apical constriction.

Ed-mediated cell sorting is similar to the process of dorsal closure. However, during dorsal closure, amnioserosa cells actively undergo pulsed contraction that leads to a reduction in their apical surfaces. This, together with the actomyosin cable acting as a ratchet, pulls the surrounding epidermal cells towards the midline (Solon et al., 2009). By contrast, the apical surface of Ed-deficient cells gradually increases when the ed-RNAi clones expand. Moreover, the actomyosin cable of the interface cells acts not as a ratchet but instead as a mechanical fence to smoothly separate wild-type and Ed-deficient cells. Finally, Ed-mediated cell sorting involves the polarized assembly of actomyosin cable only in the wild-type interface cells. This is in contrast to the formation of the anteroposterior boundary in the embryo, where the formation of actomyosin cable by cells on both sides of the boundary is postulated to be the primary mechanism of cell sorting (Monier et al., 2010). Here, we suggest that differential adhesion of Ed alone is sufficient to trigger the segregation of cells into separate populations with jagged borders, but it remains unknown whether differential adhesion mediated by differential expression of as yet unidentified compartment-specific CAMs plays a role in establishing the initial anteroposterior boundary, where actomyosin cable ensures that this boundary remains straight.

We thank P. Rorth, S. A. Spencer, the Developmental Studies Hybridoma Bank (DSHB), Bloomington Stock Center and Vienna Drosophila RNAi Center (VDRC) for providing reagents and stocks and C. Dahmann and H. Wilson for comments on the manuscript. Grants from the National Science Council, Taiwan, to J.-C.H. are acknowledged.