Dynamic analysis of filopodial interactions during the zippering phase of Drosophila dorsal closure

The fusion of sheets of epithelial cells is a common event during embryonic development and also occurs during the process of wound healing(Martin and Parkhurst, 2004). Failure of epithelial fusions during human embryonic development gives rise to a spectrum of birth defects including spina bifida and cleft palate. A widely used model of epithelial fusion is the process of dorsal closure (DC), which occurs during Drosophila embryogenesis(Harden, 2002; Kiehart, 1999). During DC, two epithelial sheets sweep towards one another over the surface of the embryo and fuse at the dorsal midline to form a continuous epidermis. Live imaging studies have revealed that dynamic needle-like protrusions called filopodia project beyond the leading edges of the epithelial sheets during DC(Jacinto et al., 2000). When filopodia from the two epithelial sheets meet one another, they interdigitate in a process known as `zippering'. Suppressing filopodia formation during DC by expressing dominant-negative Cdc42 or disassembling microtubules leads to a failure of fusion, suggesting that zippering is an essential part of the fusion process (Jacinto et al.,2000; Jankovics and Brunner,2006). Filopodial zippering has also been observed in other systems, including cultured keratinocytes and in the embryonic mouse eyelid,suggesting that it is a universal mechanism for epithelial fusion(Vasioukhin et al., 2000; Zenz et al., 2003).

Embryonic epithelial fusions must occur in a precise fashion when the fusing sheets are patterned, as is the case for neural tube closure in vertebrates and DC in flies. The epithelium of the Drosophila embryo is finely patterned prior to DC and imprecise fusion would disrupt this patterning. Early in development, the embryo is divided into a series of repeating units called parasegments, with the boundary between parasegments forming anterior to stripes of cells expressing the transcription factor Engrailed (Lawrence and Struhl,1996). Later in embryogenesis, visible segment boundaries form posterior to the engrailed stripes(Larsen et al., 2003). Thus,by DC, the embryo is patterned into segments, with each segment being divided into an anterior (A) and a posterior (P) compartment by the parasegment boundary. Engrailed and Hedgehog are expressed exclusively in P compartments(Dahmann and Basler, 2000),whereas the Hedgehog receptor Patched is expressed exclusively in A compartments (Nakano et al.,1989).

DC occurs with remarkable accuracy, such that patterning is perfectly maintained across the fusion seam at single-cell resolution. In order to achieve this level of accuracy, each cell in the leading edge must be able to identify, and specifically fuse with, its matching cell in the opposing epithelial sheet. Interestingly, cell-cell matching is perturbed by genetic interventions that abolish filopodia formation (dominant-negative Cdc42, Ena sequestration), suggesting that, in addition to mechanical zippering,filopodia might also play a role in the cell-cell matching that occurs during epithelial fusion (Gates et al.,2007; Jacinto et al.,2000).

Filopodia are widely observed in biology and they often appear to be sensory structures, allowing a cell to explore its environment, searching for guidance cues, other cells or suitable sites for attachment(Gerhardt et al., 2003; Ribeiro et al., 2002; Ritzenthaler et al., 2000; Zheng et al., 1996). However,although there is much circumstantial evidence that filopodia perform a sensory function, in most cases it is not possible to directly observe this occurring in the living embryo.

We have performed experiments to gain a clearer understanding of how cell-cell matching is achieved during DC. We demonstrate how filopodia contribute to matching and provide clear evidence of sensory and motile functions of filopodia in a live organism. This work also provides new insight into the organisation of the dorsal epithelium in the fly embryo.

DNA encoding mCherry and the actin-binding domain of DrosophilaMoesin (C-terminal 137 residues) were cloned by PCR and inserted into pUASp as KpnI-NotI and NotI-BamHI fragments,respectively (Edwards et al.,1997; Rorth, 1998; Shaner et al., 2004). Expression of this construct is non-toxic and co-localisation with GFP-actin confirmed that it effectively labels the actin cytoskeleton (data not shown).

DNA encoding the actin-binding domain of Drosophila Moesin and the SV40 terminator sequence were cloned by PCR and inserted into NotI-BglII and BglII-EcoRI sites of pCASPER2, respectively. The 1160 bp of Drosophila genomic DNA immediately upstream of the patched coding region(Forbes et al., 1993) was fused to DNA encoding eGFP by PCR and the resulting fragment was cloned into pCR4TOPO (Invitrogen) according to the manufacturer's instructions. Genomic DNA from 11,200 to 1160 bp upstream of patched was then inserted upstream of this as a BamHI-NdeI fragment. Finally, the complete patched-GFP fragment was inserted into pCASPER2 upstream of Moesin as a BamHI-NotI fragment.

UAS-RFP-Moesin and patched-GFP-Moesin transgenic lines were generated in a w background by P-element-mediated germline transformation. UAS-RFP-Moesin and patched-GFP-Moesin insertions on chromosome 2 were recombined with engrailed-Gal4 (Bloomington Stock Center) (Brand and Perrimon,1993). sGMCA (constitutively expressing GFP-Moesin) (obtained from Daniel Kiehart, Duke University, Durham, NC) on chromosome 2 was recombined with engrailed-Gal4 and UAS-RFP-Moesin(Kiehart et al., 2000).

Embryos were dechorionated in bleach then mounted in Voltalef oil under a coverslip. Images were collected on a Leica SP2 or SP5 confocal microscope and processed using ImageJ, Photoshop (Adobe) and Volocity (Improvision). Wounding was performed using a Spectra-Physics nitrogen laser.

In order to directly observe cell matching occurring during DC we generated a fly line in which two distinct populations of epithelial cells were labelled with different fluorescent proteins, enabling us to compare interactions between like and unlike cells during zippering. The strategy we used for this is illustrated in Fig. 1A. P compartments were labelled red by expressing the F-actin-binding domain of Moesin fused to the red fluorescent protein mCherry (henceforth RFP-Moesin)under the control of the engrailed (en) promoter using the UAS-Gal4 system (Brand and Perrimon,1993; Edwards et al.,1997). Alongside this, A compartments were labelled green by expressing GFP-Moesin directly under the control of 11,200 bp of sequence upstream of the patched (ptc) coding region. Consistent with the known expression pattern of ptc, the ptc-GFP-Moesin transgene was expressed in a 1- to 2-cell-wide stripe at either end of A compartments during DC, thus flanking each en-RFP-Moesin stripe(Fig. 1B)(Nakano et al., 1989). Expression of RFP-Moesin and GFP-Moesin were strictly segregated by parasegment boundaries; however, expression of both fluorophores was sporadically observed in individual cells abutting segment boundaries. The expression patterns of RFP- and GFP-Moesin were maintained through DC and perfect matching of red and green stripes along the fusion seam was observed as zippering proceeded (Fig. 1Bi-iii).

Examination of the leading edge cells during and after DC in en-RFP-Moesin/ptc-GFP-Moesin embryos revealed an unexpected but reproducible aberration in patterning. Isolated cells expressing ptc-GFP-Moesin rather than en-RFP-Moesin were frequently present within the en domain (Fig. 1C). These misplaced ptc cells were found in a conserved location, towards the posterior end of the en domain and exclusively at the leading edge. On zippering, the misplaced ptc cells fused with their matching counterpart in the opposing sheet, thus forming an island of two ptc-GFP-Moesin-expressing cells within the en domain(Fig. 1Ciii). The segmental distribution of the misplaced ptc cells was also well conserved; they were reproducibly present in segments A1-A5, but rarely in other segments(Fig. 1D). In order to identify the origin of the misplaced ptc cells, we used live imaging,commencing at the start of DC. As DC proceeded, a single ptc cell became isolated from the anterior edge of the A compartment by dorsalward migration of en cells to the leading edge(Fig. 1F; see Movie 1 in the supplementary material). Following DC, the pairs of misplaced ptccells remained within the en stripe. Thus, the misplaced ptccells derive from the A compartment, but ultimately reside in the P compartment owing to an epithelial rearrangement. This rearrangement is surprising because differential adhesion characteristics should prevent mixing of A and P cells.

The arrangement of trichomes on late embryos has been widely used as a morphological readout of epithelial patterning and we were interested to establish the trichome characteristics of the misplaced ptc cells(Payre, 2004). In order to visualise the trichome pattern, GFP-Moesin was expressed constitutively,alongside RFP-Moesin in en stripes as a positional marker. The trichome pattern of each segment of the dorsal epithelium consists of a broad band of cells possessing trichomes alongside a narrow band with naked cuticle,corresponding to the anterior end of the A compartment (illustrated in Fig. 1A)(Bokor and DiNardo, 1996). Imaging revealed that, in common with the anterior-most cells of the A compartment, the misplaced ptc cells were naked(Fig. 1E). By contrast, the en cells that surround the misplaced ptc cells all possessed trichomes. The misplaced ptc cells therefore share the morphological characteristics of the A compartment, despite residing within the P compartment.

Having characterised the dorsal epithelium in en-RFP-Moesin/ptc-GFP-Moesin embryos, we then used this fly line to investigate cell matching during DC in live embryos. In these embryos,we can clearly see red and green filopodia and observe their behaviour during zippering. Filopodia appeared to scan the opposing epithelium and interactions were only observed between filopodia of the same colour(Fig. 2A; see Movie 2 in the supplementary material). When filopodia of differing colours came into close proximity with one another, no observable interaction took place. Notably,both red-to-red and green-to-green interactions were observed, indicating that both A and P compartments are actively involved in the matching process. Thus,our data suggest that at least two distinct recognition mechanisms mediate cell matching during DC. The misplaced ptc cells within the en domain described above are consistently able to recognise and specifically fuse with one another, indicating that they have recognition properties distinct from their neighbouring en cells.

Often, the two epithelial sheets are poorly aligned immediately prior to zippering and mismatched fusion appears likely. Under these conditions, we observed that filopodia could identify and bind to matching cells several cell-diameters distant in the opposing epithelial sheet(Fig. 2B; see Movie 3 in the supplementary material). Having made contact with the appropriate partner, the filopodia form tethers, linking the matching cells together. Contraction of these filopodial tethers then draws matching cells towards one another,correcting the misalignment. These filopodial tethers thus appear to be able to exert sufficient contractile force to drag the entire epithelial sheet. Our data therefore suggest that, in addition to acting as sensory devices,filopodia also play an active role in motility, pulling the cell towards its point of attachment. This is consistent with recent in vitro studies demonstrating that filopodia can exert significant pulling forces(Kress et al., 2007). Realignments such as that shown in Fig. 2B are common, with 44% of stripes observed (n=63)exhibiting an adjustment of one cell width or more during zippering. Notably,the filopodial interactions occurring during DC do not necessarily become permanent adhesions: we noted at least one tether break during zippering of 42% of the stripes observed (n=74) (see Fig. 2Bi; for an example, see Movie 3 in the supplementary material). These breakages occur when a tether forms in isolation, unsupported by other tethers pulling in the same direction.

Our data show that filopodia are able to identify matching cells and pull misaligned epithelia into the correct alignment; however, this is under normal conditions in which there is a relatively high level of symmetry in the patterning of the two opposing epithelia. We were curious to know what would happen if the patterning of the two epithelia was asymmetric. We observed that asymmetries occasionally occurred spontaneously, and we imaged DC in several such asymmetric embryos. The embryo shown in Fig. 3A (see also Movie 4 in the supplementary material) has an extra misplaced ptc cell within an en domain. This extra ptc cell fused with the neighbouring A compartment and this subsequently pushed the remainder of the segment out of alignment. This resulted in en cells being brought into close proximity to the opposing en stripe of the neighbouring segment and,interestingly, fusion then occurred between these en cells, despite the fact that they reside in different segments. DC subsequently continued as normal, but one segment out of register. We observed a similar sequence of events in three other asymmetric embryos. These observations demonstrate that although matching is specific to the compartment, it is not specific to the segment. Therefore, asymmetries that are sufficiently great that filopodia can reach the equivalent compartment of a neighbouring segment have the potential to result in mismatch.

We next used laser ablation of leading edge cells to assess the effect of inducing asymmetry in symmetrically patterned embryos. We were able to specifically ablate a single en stripe of leading edge cells from one of the epithelial sheets. The corresponding en stripe of the opposing epithelial sheet no longer had matching cells with which to fuse, and we focused on the behaviour of this unpartnered en stripe upon confrontation with the opposing epithelial sheet. The most common outcome (11 of 13 embryos) was that this en stripe became constricted and was effectively extruded from the leading edge, such that the en cells did not participate in zippering (Fig. 2B). In the remaining cases, the unpartnered en stripe partially constricted, then fused with the opposing en stripe of the neighbouring segment, as observed in Fig. 2A.

These data demonstrate that when faced with asymmetries that cannot be resolved by filopodial searching and tethering, zippering occurs in such a way that contact between the mismatched cells is minimised.

What is the molecular basis of cell matching? The data above are consistent with matching being based on just two sets of molecular interactions, one allowing A compartment cells to recognise one another and the other performing the same function for P compartment cells. An obvious possibility is that the molecules that mediate cell matching during DC are the same as those that maintain the integrity of these compartments throughout the epithelium. Alternatively, there could be a different set of recognition molecules present exclusively at the leading edge to mediate cell-cell matching. Filopodia are also observed during the healing of wounds in the ventral epithelium and we reasoned that these wound filopodia should exhibit matching behaviour if the molecules that mediate matching are present throughout the epithelium(Wood et al., 2002). Laser wounds were made to the ventral epithelium across en stripes such that the wound edge consisted of both en-RFP-Moesin and ptc-GFP-Moesin cells. On healing of these wounds, we observed repeated interactions between the filopodia of matching cells, but not between mismatching cells. In the example shown in Fig. 4, a number of filopodial tethers formed between ptc cells on opposite sides of the wound as closure proceeded. Near the end of closure, a filopodial tether formed between en cells on opposite sides of the wound, leading to fusion of these cells and regeneration of the en stripe(Fig. 4; see Movie 5 in the supplementary material). This sequence of events was observed in six out of ten similar wounds. In the remaining wounds, the tethers between ptccells became permanent adhesions before the en cells were close enough to form tethers and hence the en stripe was not regenerated. These data suggest that both A and P compartment cells away from the leading edge can carry out filopodial matching analogous to that occurring during DC,and hence the adhesion molecules that mediate the process are not leading edge specific.

The data presented here demonstrate that specific recognition events ensure the accuracy of fusion during DC. Filopodia facilitate matching by allowing a cell to search for its match and also to pull misaligned sheets into alignment. This explains why genetic interventions that abolish filopodia lead to an increase in mismatching (Gates et al., 2007; Jacinto et al.,2000). It appears that at least two recognition processes act during DC, one for P compartments and one for A compartment cells, but these recognition events are not segment-specific, as fusions can occur between matching compartments from different segments. Filopodial matching is also observed during healing of wounds in the ventral epithelium, suggesting that the molecules mediating recognition are found throughout the epithelium. These data are consistent with the notion that the adhesion molecules that mediate filopodial matching during DC are the same as those that ensure compartment integrity throughout the epithelium; however, the identity of these molecules is currently unknown. Experimental and modelling studies have shown that cells can sort based on differential levels of just one adhesion molecule, and it has been hypothesised that a single adhesion molecule might be responsible for compartmental segregation (Dahmann and Basler, 2000). Our data suggest that, at least during filopodial matching, this is not the case, as we observe specific recognition events for both A and P compartments and neither compartment is obviously dominant in the matching process. It is of course possible that multiple mechanisms contribute to cell matching and segregation, perhaps with different adhesion molecules governing the rapid, transient associations between filopodia and the long-lived adhesions that hold cells together permanently. Whereas segregation between leading edge A and P compartment cells is absolute at the parasegment boundary, as discussed above we reproducibly see a single A compartment ptc cell move into the P compartment at the segment boundary. This might suggest that differences in adhesive properties between cells either side of the segment boundary are small, permitting a degree of mixing. However, during DC, the misplaced ptc cells are consistently able to recognise and specifically fuse with matching cells in the opposing epithelial sheet, indicating adhesive properties distinct from their neighbours. When the arrangement of the misplaced ptc cells is disrupted, it can result in severe mismatches; therefore, correct positioning of these cells is clearly important in epithelial sheet alignment. These cells occupy a unique and defined position in each segment and might assist the matching process by acting as a `keystone' that helps to ensure precise alignment within the segment.

We thank Phil Ingham for providing ptc DNA and Brian Stramer for helpful discussions and technical assistance. T.H.M. is funded by a Wellcome Trust advanced training fellowship.