Potential dual molecular interaction of the Drosophila 7-pass transmembrane cadherin Flamingo in dendritic morphogenesis.

Seven-pass transmembrane cadherins (7-TM cadherins) play pleiotropic roles in epithelial planar cell polarity, shaping dendritic arbors and in axonal outgrowth. In contrast to their role in planar polarity, how 7-TM cadherins control dendritic and axonal outgrowth at the molecular level is largely unknown. Therefore, we performed extensive structure-function analysis of the Drosophila 7-TM cadherin Flamingo (Fmi) and investigated the activities of individual mutant forms mostly in dendritogenesis of dendritic arborization (da) neurons. One of the fmi-mutant phenotypes was overgrowth of branches in the early stage of dendrite development. In da neurons but not in their adjacent non-neuronal cells, expression of a truncated form (ΔN) that lacks the entire cadherin repeat sequence, rescues flies - at least partially - from this phenotype. Another phenotype is observed at a later stage, when dendritic terminals outgrowing from the contralateral sides meet and then avoid each other. In the fmi mutant, by contrast, those branches overlapped. Overexpression of the ΔN form on the wild-type background phenocopied the overlap phenotype in the mutant, and analysis in heterologous systems supported the possibility that this effect might be because the Fmi-Fmi homophilic interaction is inhibited by ΔN. We propose that a dual molecular function of Fmi play pivotal roles in dendrite morphogenesis. In the initial growing phase, Fmi might function as a receptor for a sofar-unidentified ligand and this hypothetical heterophilic interaction would be responsible for limiting branch elongation. At a later stage, homophilic Fmi-binding at dendro-dendritic interfaces would elicit avoidance between dendritic terminals.


Introduction
Seven-pass transmembrane cadherins constitute an evolutionally conserved subfamily of the cadherin superfamily (Usui et al., 1999;Tepass et al., 2000;Yagi and Takeichi, 2000;Hirano et al., 2003). In contrast to the studies indicating that classic cadherin is primarily responsible for intercellular adhesion, genetic studies of the Drosophila 7-pass transmembrane cadherin Flamingo [Fmi; also known as Starry night (Stan)] have shown its pleiotropic role in intercellular communications by controlling epithelial and neuronal cell morphogenesis (Usui et al., 1999;Chae et al., 1999).
Whereas Fmi operates in the PCP pathway in epithelia, Fmimediated control of dendritic and axonal outgrowth appears to occur separately from the PCP pathway (Gao et al., 2000;Lee et al., 2003;Senti et al., 2003), and the molecular function of Fmi in neurons is largely unknown. In an approach to understand the molecular mechanism, we focused on evolutionally conserved, but complicated, structural features in the family of 7-pass transmembrane cadherins (hereafter referred to as the Flamingo family), and performed extensive in vivo structure-function analysis in Drosophila. Extracellular Seven-pass transmembrane cadherins (7-TM cadherins) play pleiotropic roles in epithelial planar cell polarity, shaping dendritic arbors and in axonal outgrowth. In contrast to their role in planar polarity, how 7-TM cadherins control dendritic and axonal outgrowth at the molecular level is largely unknown. Therefore, we performed extensive structure-function analysis of the Drosophila 7-TM cadherin Flamingo (Fmi) and investigated the activities of individual mutant forms mostly in dendritogenesis of dendritic arborization (da) neurons. One of the fmi-mutant phenotypes was overgrowth of branches in the early stage of dendrite development. In da neurons but not in their adjacent nonneuronal cells, expression of a truncated form (⌬ ⌬N) that lacks the entire cadherin repeat sequence, rescues flies -at least partially -from this phenotype. Another phenotype is observed at a later stage, when dendritic terminals outgrowing from the contralateral sides meet and then avoid each other. In the fmi mutant, by contrast, those branches overlapped. Overexpression of the ⌬ ⌬N form on the wild-type background phenocopied the overlap phenotype in the mutant, and analysis in heterologous systems supported the possibility that this effect might be because the Fmi-Fmi homophilic interaction is inhibited by ⌬ ⌬N. We propose that a dual molecular function of Fmi play pivotal roles in dendrite morphogenesis. In the initial growing phase, Fmi might function as a receptor for a sofar-unidentified ligand and this hypothetical heterophilic interaction would be responsible for limiting branch elongation. At a later stage, homophilic Fmi-binding at dendro-dendritic interfaces would elicit avoidance between dendritic terminals.
regions of the Flamingo family consist of cadherin repeats, which function as homophilic binding modules, and other motifs that are also suggestive of protein-protein interaction (Fig. 1). The sequences of these transmembrane domains show similarity to those of the secretin-receptor family of G-proteincoupled receptors (GPCRs). We made a series of transgenic strains to express various mutant forms of Fmi and addressed how the expression of individual forms affects dendrite morphogenesis and PCP.
We studied the role of Fmi in dendritic morphogenesis in a subset of sensory neurons, i.e. the dendritic arborization (da) neurons (Bodmer and Jan, 1987;Jan and Jan, 1993;Campos-Ortega and Hartenstein, 1997). Growth and elaboration of dendritic arbors of da neurons are easily imaged by wholemount time-lapse recordings with the help of various green fluorescent protein (GFP) markers (Gao et al., 1999;Grueber et al., 2003;Sugimura et al., 2003). At late embryonic stages, da neurons start growing two-dimensional dendrites underneath the epidermis, and the growth pauses at the end of embryogenesis (explained later; see Fig. 2A, Fig. 3A). Dendritic growth restarts in larvae, and branches keep growing by expanding the body wall in a coordinated fashion. Dendritic terminals of a subclass of da neurons meet mid-way between homologous cells in ispilateral or contralateral adjacent hemisegments at early larval stages [approximately 30-35 hours after egg laying (AEL)], and tend to turn away before crossing each other (see Fig. 5A, Fig. 6A). This phenomenon is called hetero-neuronal avoidance or tiling, and is mediated by inhibitory interaction at individual dendro-dendritic interfaces (Grueber et al., 2002;Grueber et al., 2003;Sugimura et al., 2003;Jan and Jan, 2003).
Dendrites of dorsal da neurons in fmi mutants show two types of phenotype (Gao et al., 2000;Sweeney et al., 2002). In the first, dorsal branches emerge precociously and overextend towards the dorsal midline in the embryos (the 'overgrowth phenotype ';Fig. 2C,E and Fig. 3B). In the second, along the dorsal midline at a later larval stage, dendritic terminals that outgrow from the contralateral sides do not avoid each other, but overlap (the 'overlap phenotype'; Fig. 5B). Although previous mosaic analysis suggests a cellautonomous role of Fmi in da neurons, we sought to address the question further how Fmi restricts dendritic growth in the embryos and to control heteroneuronal avoidance in the larvae. On the basis of previous studies and our structurefunction analysis, we discuss the possibility that Fmi controls the early outgrowth and the mutual avoidance by way of two distinct mechanisms.

Results
Quantification of rescue of the dendritic overgrowth phenotype by neuron-specific Gal4 strains Branches in the dorsal region of fmi-mutant embryos overextended towards the dorsal midline and, in addition, lateral branches either showed poor growth or were misrouted dorsally ( Fig. 2C) (Gao et al., 1999;Gao et al., 2000). Flies were rescued from the overshooting phenotype by expressing the fmi transgene in the entire nervous system or in all da neurons (Gao et al., 2000). By using our Gal4 drivers that label a small subset of da neurons (Sugimura et al., 2003;Sugimura et al., 2004), we visualized phenotypes with higher resolution and quantitatively assessed the rescue activity of individual Fmi forms. One of those drivers, IH1 GAL4, highlighted two  (Oda and Tsukita, 1999), and we discussed this result and other constructs in Materials and Methods. In the tracing, one of the es neuron accessory cells (yellow) is also drawn. In contrast to the wild type (A,B), the fmi mutant extended the da dendrites more dorsally than es dendrites (C). Yellow bracket in C indicates lateral branches that showed undergrowth and/or misrouting. Neuronal expression of (F) Fmi and (G) Fmi::EYFP but not that of (H) ⌬Ctail, prevented both the dorsal overgrowth and malformation of lateral branches (bracket in F), whereas the rescue effect of ⌬N::EYFP was partial (I,J). Overextended dendrites of the contralateral counterpart are indicated by a magenta arrow (H). Detailed genotypes of individual panels are described in K. The Gal4 driver used in this rescue experiment was IH1. Although IH1-driven Fmi expression prevented the phenotype, it did not cure embryonic lethality of the fmi E59 /fmi E59  dorsal da neurons (ddaD and ddaE) as well as external sensory (es) neurons in the embryo ( Fig. 2A,B).
To judge the dendritic overgrowth of ddaD and ddaE, we took advantage of the relative position of dendritic tips of the ddaD and ddaE neurons and the es neurons in the embryos just before hatching (20-22 hours AEL). In the wild type, tips of the es neurons were located more dorsally than those of ddaD and ddaE neurons (in 32 out of 32 abdominal hemi-segments examined; A2-A6) ( Fig. 2A,B arrows and arrowheads). By contrast, dendrites of ddaD and ddaE neurons extended more dorsally than those of es neurons in 91% of the hemi-segments of fmi null mutant embryos (Fig. 2C). This high penetrance was reduced to 19% when two copies of the fmi transgene were expressed in the mutant (Fig. 2F,J,K). We also expressed in the mutant the full-length Fmi, which had been tagged with enhanced yellow fluorescent protein (EYFP) (Fmi::EYFP), and showed that two copies of Fmi::EYFP exhibited a rescue activity with respect to both the overgrowth phenotype and lethality that was similar to the presence of two copies of the untagged Fmi transgene (Fig. 2G,J,K). Not only the dorsal overgrowth phenotype, but also poor growth in lateral directions was prevented by expression of Fmi or Fmi::EYFP (see brackets in Fig. 2C,F). It should be noticed that neither the distance between cell bodies of es neurons and da neurons nor the length of the es dendrite was significantly altered in wildtype, mutant and rescued animals (data not shown). The overgrowth phenotype is not specific to ddaD and ddaE neurons, shown by the fact that dendrites of a third da neuron, ddaC, also displayed dorsal overextension (Fig. 2, compare D with E). Since none of the Gal4 lines in this or previous studies appeared to drive transgenic expression in embyronic epidermal cells that might make contact with dendrites, the significant rescue by these drivers suggests that fmi is necessary for normal dendritic growth in da neurons but not, as previously discussed, in epidermis (Gao et al., 2000;Ye and Jan, 2004).
⌬N::EYFP, a form without cadherin repeats, partially rescues the mutant embryos from the overgrowth phenotype Individual forms of the designed Fmi-mutants (illustrated in Fig. 1) were examined to see whether their expression can prevent the overgrowth phenotype and whether their overexpression can exert a dominant phenotype on the wildtype background. Among the deletions and truncations examined, ⌬N::EYFP gave the most interesting results that are described below. Although ⌬N::EYFP lacks almost all of the conserved extracellular motifs except for a hormone-receptor domain (HRM), its expression in da neurons partially rescued the mutant embryos from the overgrowth phenotype ( Fig. 2I,J). The HRM is about 60 amino acids (aa) long and conserved in the subfamily of G-protein coupled receptors (GPCRs) (supplementary material Fig. S1); conserved cysteine residues in HRM have been implicated in transmitting signals of ligand binding to intracellular components (Asmann et al., 2000). We collected images of da neurons in live animals that expressed either Fmi::EYFP or ⌬N::EYFP, quantified EYFP fluorescence of the neurons expressing either form, and showed that the expression level of ⌬N::EYFP was comparable to that of Fmi::EYFP (supplementary material Fig. S2).
The rescue effect of ⌬N::EYFP expression on the overgrowth phenotype was also investigated in a different way of quantification, in which the distance between tips of dendritic terminals and the dorsal midline was measured in embryos 21-22 hours AEL (Fig. 3). In the wild-type embryo, dendritic terminals do not reach the dorsal midline, leaving dendrite-free zones on both sides of the dorsal midline ( Fig.  3A). By contrast, about 80% of the terminals in the mutant reached the midline before hatching (Fig. 3B). The distribution of the distance was compared between the wild type, the fmi mutant, and the fmi mutants expressing Fmi::EYFP or ⌬N::EYFP, and the partial rescue activity of ⌬N::EYFP was again indicated (Fig. 3C,D). As predicted from the absence of cadherin repeats, expression of ⌬N::EYFP in S2 cells did not lead to the formation of cell aggregates, whereas expression of Fmi or Fmi::EYFP did (Fig. 8).
In the wild-type embryo, dendrites of da neurons extend on the basal surface of the epidermis, and both da neurons and epidermis express Fmi (Usui et al., 1999;Gao et al., 2000). As described above, expression of fmi in mutant da neurons was sufficient to rescue the fmi-mutant embryos from the dendritic phenotype (Gao et al., 2000;this study). Under the rescue condition described in Fig. 2, Fmi::EYFP or ⌬N::EYFP were distributed in dendrites and cell bodies in a splotchy fashion (Fig. 2L,M), reminiscent of the distribution of endogenous Fmi molecules in the wild-type neurons. However, obvious signals of Fmi::EYFP or ⌬N::EYFP were not detected in the overlying epidermis. Together with our data of the partial rescue by ⌬N::EYFP, all these results are difficult to explain by the hypothesis that the homophilic Fmi-Fmi interaction between dendrites and epidermis plays a crucial role in controlling dendritic extension. One possible explanation of the rescue is that Fmi interacts in a heterophilic manner with an unknown molecule that restricts dendritic growth. The extracellular region of ⌬N::EYFP, which consists of a HRM and a segment more proximal to the membrane, together with extracellular loops of the 7-pass transmembrane domain, might be involved in such a hypothetical heterophilic interaction (Fig. 10A). Alternatively, the role of Fmi in dendritic growth restriction might not require any extracellular signals.
Overexpression of Fmi or Fmi::EYFP results in undergrowth of dendrites To pursue whether overexpression of any forms of Fmi on the wild-type background exerts a dominant effect on dendritic growth, we explored phenotypes when robustly expressing individual Fmi mutant forms in da neurons, by using a total of four copies of two postmitotic, pan-da Gal4 insertions, 109(2)80 and IG1-2 (Fig. 4).
Overexpression of Fmi or Fmi::EYFP gave rise to strong and highly penetrant phenotypes. In all of the observed 40 dorsal clusters, both the total length of dendrites and the number of terminals greatly decreased, although the number of cells per cluster did not change (Fig. 4, compare A with B). However, no other form, including ⌬N::EYFP, affected dendrite formation (Fig. 4C). These results strengthen the proposed role of Fmi in limiting dendritic growth. Moreover, they support the idea that Fmi::EYFP is functionally equivalent to Fmi and that the activity of ⌬N::EYFP to restrict growth is substantially weaker than that of the full-length molecules. It has previously been reported that Fmi overexpression [by using two copies of 109 (2)80] caused a dendritic overgrowth that is similar to the mutant phenotype, although overexpression-induced overextension was much less penetrant than the phenotype seen in the mutant (10%) (see Gao et al., 2000). Because both the previous and our study used the same UAS transgenic strain (Usui et al., 1999), the difference in results appeared to be due to distinct levels of overexpression, because of different copy numbers of Gal4 drivers and/or protocols to assess dendritic overgrowth.
Fmi::EYFP prevents the dorsal-dendrite-overlap phenotype at larval stages, whereas ⌬N::EYFP does not As described above, ⌬N::EYFP appeared to retain the partial activity to control dendritic growth in the embryo. Can the same Journal of Cell Science 119 (6) form rescue the mutant animals form the dorsal-dendrite-overlap phenotype at later larval stages? To address this question, we expressed either Fmi::EYFP or ⌬N::EYFP in transheterozygotes fmi 72 /fmi E59 that survived until early-to mid-larval stages (Fig. 5, details of genotypes are described in the legend). In the trans-heterozygous larvae, terminals of dorsal dendrites that had extended from contralateral sides crossed or overlapped each other in 25 out of 26 abdominal segments examined at 32-35 hours AEL (Fig. 5, compare A with B). To selectively visualize dendritic terminals of class IV da neurons that display significant interneuronal avoidance in normal development (Grueber et al., 2002;Grueber et al., 2003;Sugimura et al., 2003), we ablated da neurons of all the other classes in dorsal clusters of one hemi-segment and its contralateral counterpart, but left one pair of class IV ddaC alone. This experiment showed that terminals of the adjacent ddaC neurons did not avoid each other in five out of five segments examined in the fmi mutants (Fig. 5, compare D,G with E,H).
This penetrant overlap phenotype was substantially prevented by Fmi::EYFP expression (Fig. 5C). By contrast, it was hardly cured by ⌬N::EYFP expression in any of the six segments, where only class IV dendrites were visualized (Fig.  5F,I), which suggests the inability of ⌬N::EYFP to act for heteroneuronal avoidance between dorsal dendrites. These data suggest that extracellular motifs, which were missing in ⌬N::EYFP, are necessary for the inhibitory dendro-dendritic interaction.
⌬N::EYFP expression on the wild-type background causes dendritic terminals to cross each other To investigate the molecular mechanism how Fmi operates in the avoidance between dorsal dendritic terminals, we explored whether expression of ⌬N::EYFP (which failed to prevent the overlap phenotype) exerts a dominant-negative effect on the wild-type background or not. To assess the effect of ⌬N::EYFP expression at high resolution, we employed the driver NP1161 and marker ppk-EGFP for class IV da neurons (including ddaC), which extend dendritic terminals far enough to encounter terminals of counterpart cells in adjacent hemisegments (Grueber et al., 2003;Sugimura et al., 2003;Sugimura et al., 2004).
In all of the more than 50 dorsal segments examined in the control larvae, sharp boundaries were generated between contralateral dendritic territories (Fig. 6A,A'). By contrast, ⌬N::EYFP expression caused incomplete segregation of dendritic terminals in six out of the 50 segments observed (Fig.  6B,B'). ⌬N::EYFP-expressing dendritic terminals overlapped with each other along the midline, and it was difficult to judge, to which cell individual terminals belonged (Fig. 6A'-C', white terminals). Furthermore, terminals occasionally invaded the territory of contralateral ddaC (Fig. 6B'and C, arrowheads). The effect of ⌬N::EYFP-expression was reproduced by using another Gal4 driver, ppk-Gal4, allowing us to express ⌬N::EYFP and membrane-bound GFP in ddaC as well as Dendrites that belonged to one of the adjacent hemisegments are green or magenta, whereas intertwining branches that were difficult to track down are white. (A,D) Control larvae in which dendritic branches covered the whole body walls with minimum overlap. (B and E) In fmi mutant larvae, dendritic terminals that had extended from contralateral sides did not turn away, but rather overlapped or crossed with each other (bracket). Fmi::EYFP expression in da neurons in the fmi mutants rescued the larvae from the overlap phenotype (C), whereas ⌬N::EYFP expression did not confer an inhibitory interaction between terminals (F). Genotypes were Gal109 (2)  another UAS-⌬N::EYFP stock in which the transgene was inserted into a different chromosome (Fig. 6D-F; see legend for details of genotypes). In contrast to those of the control (Fig. 6D), dendritic terminals of ⌬N::EYFP-expressing ddaCs overlapped with each other and/or invaded the contralateral hemi-segment in about 20% of the segments examined (Fig.  6E,F). Such overlap and invasion phenotypes were not observed when Fmi::EYFP was expressed on the wild-type background using the same drivers (data not shown). The Fmi::EYFP expression in Fig. 6 did not cause growth restriction, either, probably because the expression levels shown in Fig. 6 were lower than that shown in Fig. 4C (compare genotypes described in both figure legends).

⌬N::EYFP expression decreases the level of endogenous Fmi at intercellular boundaries in epithelial cells
How does ⌬N::EYFP overexpression interfere with the inhibitory communication at dendro-dendritic interfaces? ⌬N::EYFP might affect the intracellular localization of endogenous Fmi molecules. This hypothesis was technically challenging to verify in dendrites of da neurons, because da dendrites are too thin to examine whether labeled proteins are distributed on the plasma membrane or not. Therefore, we tested our hypothesis by using imaginal epithelia, where Fmi is localized at apically positioned adherens junction and binds in a homophilic manner at intercellular boundaries ( Fig. 7) (Usui et al., 1999).
The Fmi level at cell boundaries was dramatically decreased inside the ⌬N::EYFP-expression domain (Fig. 7A-F). ⌬N::EYFP expression did not lead to disruption of either cellto-cell adhesion or adherens junction, as shown by the fact that the cell adhesion molecule DE-cadherin was still located at the apical intercellular boundaries (data not shown). This result suggests that ⌬N::EYFP expression either downregulates the level of Fmi within each cell and/or relocates Fmi from adherens junction to either apical free cell surfaces, basolateral membrane domains or intracellular compartments. By Journal of Cell Science 119 (6) extrapolation, we speculate that endogenous Fmi can hardly be present at dendro-dendritic interfaces when ⌬N::EYFPexpressing branch terminals meet.
In contrast to the substantial decrease in the Fmi level between cells with high ⌬N::EYFP expression, Fmi accumulated preferentially at those interfaces, where cells with distinct levels of ⌬N::EYFP juxtaposed each other (Fig. 7D-I). This was prominent along the borders of the expression domain ( Fig. 7D-F, arrowheads) and also within the expression domain that was mosaic in terms of the expression level ( Fig. 7G-I,  arrowheads). These interface accumulations provided a contrast to what happens along borders of fmi mutant clones, namely that Fmi is totally missing at any cell boundaries between fmi+ and fmi-cells (Usui et al., 1999). Therefore, at least in the ⌬N::EYFP-expressing cells that made contact with non-or hardly expressing cells, Fmi molecules were not degraded or prevented to exit the ER or Golgi network but redistributed to those cell contact sites.

Coexpression of ⌬N::EYFP with Fmi inhibites Fmimediated homophilic cell adhesion
The above results of ⌬N::EYFP expression in imaginal discs motivated us to study whether its high-level expression would be able to inhibit Fmi-Fmi homophilic interaction in the cell culture system (Usui et al., 1999) or not. We first studied whether expression of the wild-type or each of the mutants results in the formation of cell aggregates or not (Fig. 8A-8F; see also Materials and Methods). In contrast to Fmi-expressing S2 cells, which formed large aggregates, cells that coexpressed Fmi and ⌬N::EYFP did not show such an adhesive property. Moreover, this effect of the coexpression depended on the relative dose of a transfected expression plasmid of Fmi to that of ⌬N::EYFP (Fig. 8G,H). Under the condition in which cells coexpressing Fmi and ⌬N::EYFP did not form aggregates, our western analysis showed that the level of Fmi was not reduced compared with that in cells expressing Fmi alone (data not shown), favoring the idea that ⌬N::EYFP expression does not downregulate the overall level of Fmi.
How does ⌬N::EYFP inhibit Fmi-Fmi homophilic interaction at the molecular level? The structure of ⌬N::EYFP and the effects of its expression are reminiscent of those of a dominant-negative form of classic cadherin, which lacks almost all of its extracellular region but retains its transmembrane domain and the cateninbinding intracellular tail (Fujimori and Takeichi, 1993). When this form was expressed in a keratinocyte cell line, endogenous cadherins localizing at cell-cell boundaries was largely diminished, although the amount of E-cadherin was not significantly affected. Classic cadherins have been shown to dimerize in cis or laterally (between molecules on the same cell) as well as in trans (molecules from adjacent cells), and it has been suggested that the lateral dimeric structure is necessary for intercellular adhesive activity (Ozawa, 2002;Patel et al., 2003). It has been proposed that this lateral molecular interaction is inhibited by the dominant-negative form (Fujimori and Takeichi, 1993).
We hypothesized that ⌬N::EYFP exerts its effect on Fmimediated homophilic interaction in an analogous fashion, and so examined whether ⌬N::EYFP physically interacts with Fmi or not (Fig. 8I). Coexpression of Fmi and HA-tagged ⌬N::EYFP in HEK293T and subsequent immunoprecipitation showed that Fmi molecules coimmunoprecipitated with ⌬N::EYFP. This result implies the possibility that Fmi and ⌬N::EYFP interact with each other in the same cells and that this molecular interaction exerts its dominant effect over Fmi-Fmi binding at intercellular contact sites. This binding of ⌬N::EYFP to Fmi might either lead to the internalization of the ⌬N::EYFP-Fmi complex into the cytoplasm or the distribution of Fmi on the cell surface as nonfunctional complexes. To examine these possibilities, we studied subcellular localization of Fmi in the presence or absence of ⌬N::EYFP in S2 cells (Fig. 8J,K). A subpopulation of Fmi molecules were present on the plasma membrane when the Fmi-expressing plasmid was transfected to S2 cells (Fig. 8J). This localization was not dramatically altered in the steady (J,K) S2 cells were transfected with a plasmid expressing (J) membrane-bound Venus (Venus-pm) or with (K) ⌬N::EYFP plasmid, together with the Fmi expression plasmid. Cells were spread on ConA-coated dishes and stained for GFP (left panels; green in merged right panels) and Fmi (middle panels; magenta in merged right panels). We used 0.05% saponin to permeabilize plasma membranes in order to preserve intracellular vesicular structures. Over 20 cells were observed for each transfection experiment and all cells showed similar protein distributions to cells shown in J and K. state when ⌬N::EYFP was coexpressed (Fig. 8K). Therefore, this result appears to be consistent with the latter hypothesis of the presence of ⌬N::EYFP-Fmi on the cell surface, although it was difficult to rule out the possibility that ⌬N::EYFP affected kinetics of both transport and endocytosis of Fmi.
Our above results, obtained by using imaginal epithelia and cultured cells, are consistent with the hypothesis that overexpressed ⌬N::EYFP molecules bind endogenous Fmi and form nonfunctional complexes and that, although the complex stays on plasma membrane domains, Fmi is redistributed out of dendro-dendritic interfaces. This relocation of Fmi would consequently hinder Fmi-mediated intercellular communication that should elicit avoidance between dorsal dendritic terminals (Fig. 10B).
Misexpression of ⌬N::EYFP leads to effects opposite to that of Fmi on planar polarity Considering multiple in vivo roles of Fmi, we examined how individual mutant forms would behave in epithelial planar cell polarity (PCP), in which Fmi acts through one of the Fz signaling pathways (PCP pathway). Activity of polarity regulators can be easily assayed by overexpression. One of such gain-of-function phenotypes is generated when protein levels are gradient along the anterior-posterior axis in the wing (Fig. 9A-D) (Adler et al., 1997;Usui et al., 1999). Wing hairs point up the gradient of increasing Fmi levels, whereas they point towards the lower concentration of the Fz gradient (Fig.  9B,D). Misexpression of Fmi::EYFP or ⌬C tail caused the pointing-up phenotype, just like when Fmi is overexpressd. By contrast, misexpression of ⌬N::EYFP caused the pointingdown phenotype (Fig. 9C), which is opposite to the effect of Journal of Cell Science 119 (6) Fmi overexpression, and strikingly similar to that of Fz overexpression.
Fmi misexpression not only reorients polarity of wing hairs within the expression domain, but also generates non-cell autonomous effects (Tree et al., 2000). These effects can be detected by visualizing the distribution of endogenous Fmi at the cell-boundary and of other components of the PCP pathway, such as Dishevelled (Dsh), that are normally localized at proximal/distal (P/D) boundaries (Fig. 9H, arrow). Under our conditions of ⌬N::EYFP overexpression, endogenous Fmi was relocated from P/D boundaries to anterior/posterior (A/P) boundaries, not only within the expression domain but also in a 2-3 cell-wide stripe that juxtaposed overexpressing cells (Fig. 9H, arrowhead and  bracket). The non-cell autonomous effect of ⌬N::EYFP was also seen as the reorientation of apical structures of adult cuticle cells (Fig. 9I-M). ⌬N::EYFP-overexpressing clones (⌬N::EYFP+ clones, stained blue in Fig. 9I-M) caused wing hairs and trichomes of neighboring cells to point away from the clones (Fig. 9J-M). This non-autonomous effect was exactly the opposite of that of Fmi+ clones, but was identical to the phenotype of Fz+ clones (Strutt, 2001;Tree et al., 2002). These opposite effects of polarity between Fmi and ⌬N::EYFP expression cannot be explained simply by the fact that ⌬N::EYFP decreased endogenous Fmi levels at cell boundaries, because fmi loss-of-function clones did not produce an obvious non-cell autonomous effect (Usui et al., 1999). The above effects of ⌬N::EYFP overexpression depended on Fz, as shown by the fact that the overexpression phenotype did not appear in fz mutants (data not shown). This result supports the possibility that ⌬N::EYFP cooperates with Fz to transduce the polarizing signal.

Discussion
On the basis of our extensive structure-function analysis, we propose a hypothesis for the molecular functions of Fmi. To control dendritic morphogenesis in the embryo, one likely molecular function of Fmi is that of a receptor for a yet-to-beidentified ligand and that this hypothetical Fmi-ligand interaction is responsible for appropriate pausing of branch elongation (Fig. 10A). This hypothesis also explains an axonretraction effect by Fmi overexpression in the mushroom body (Reuter et al., 2003). The partial rescue activity of ⌬N::EYFP could be due to weak binding to such a hypothetical ligand. ⌬N::EYFP retains its HRM domain and, consistently, our structure-function analysis of the mammalian 7-pass transmembrane cadherin Celsr2 also implied a functional role for an extracellular subregion that includes the HRM domain (Shima et al., 2004). These results suggest that the role of this domain to control dendritic growth is conserved among species.
In contrast to rescue activity of ⌬N::EYFP towards the embryonic overextension phenotype, ⌬N::EYFP is a loss-offunction and dominant-negative form in the inhibitory interaction at dendro-dendritic interfaces in the larval stage. The molecular nature of ⌬N::EYFP was investigated in imaginal discs and in cultured cells. Fmi molecules in ⌬N::EYFP-expressing cells in the disc were not held back in the ER or Golgi on their way to cell membranes. In addition, coexpression experiments in cultured cells showed that, ⌬N::EYFP bound to Fmi and ⌬N::EYFP expression did not dramatically alter the distribution of Fmi at the plasma-membrane, which suggests the possibility that ⌬N::EYFP-Fmi complexes stay on the cell surface but out of contact sites, where abutting cells express ⌬N::EYFP. The simplest explanation of the effect of ⌬N::EYFP expression at dendro-dendritic interfaces is that, Fmi-Fmi interaction plays a role in the mutual avoidance during dynamic cycles of terminal extension and retraction, and this interaction is supported by homophilic interaction of cadherin domains (Fig. 10B). In addition to the likely role of this trans Fmi-Fmi homophilic interaction, other possibilities are not excluded. For example, cis or lateral interaction of Fmi might recruit other cell surface receptors and ligands responsible for the bi-directional signaling for avoidance, such as Eph and ephrin.
We interpreted different results of ⌬N::EYFP expression in the two distinct rescue experiments such that Fmi exerts two types of molecular interactions. Although we showed that both the full-length form (Fmi::EYFP) and the short form (⌬N::EYFP) were produced at similar levels in da neurons of our transgenic flies, it is difficult to totally rule out the possibility that different processes of dendritogenesis (elongation vs interneuronal avoidance) require different threshold levels of protein activity. Our hypothesis needs to be further tested by investigating functional interactions between Fmi and other molecules that operate in dendritogenesis, and by pursuing other approaches to identify binding partners of Fmi.

Molecular cloning
Enhanced yellow fluorescent protein (EYFP, Clontech) was fused to the C-terminal of Fmi by using a spacer, GRVGGGGSGGGGSGGGGSSVD (Huston et al., 1988;Chaudhary et al., 1989). Detailed structures of mutant forms of Fmi and additional information about molecular cloning is available upon request.
Two other extracellular deletions (⌬HR::EYFP and ⌬CR:EYFP), which retained much larger extracellular regions than ⌬N::EYFP, did not allow the mutant embryos to recover from the phenotype to an extent that could be detected by following our protocol ( Fig. 1 and Fig. 2J). Although we found that ⌬HR::EYFP and ⌬CR:EYFP -when expressed in S2 cells or in salivary glands -were present on plasma membranes (data not shown), both forms appeared to be nonfunctional because they were unable to prevent the overgrowth phenotype. Two other assays are summarized in Fig. 1, the cell aggregation assay (Fig. 8) and an assay to examine an activity in planar cell polarity (PCP). We tested whether or not misexpression of each Fmi form in the wing was able to reorient wing hairs (Fig. 9). In contrast to Fmi, Fmi::EYFP, and ⌬C tail, ⌬HR::EYFP and ⌬CR:EYFP did not exhibit detectable activity in any of these assays. Dendritic distributions and expression levels of ⌬HR::EYFP and ⌬CR:EYFP were comparable with that of Fmi::EYFP as were those of ⌬N::EYFP (see Fig. 2L,M).
Inactivity of ⌬HR::EYFP and ⌬CR:EYFP was reminiscent of results of in vivo structure-function analysis of DE-cadherin (Oda and Tsukita, 1999). In the DEcadherin molecule, the last cadherin repeat is followed by EGF-like domains and a laminin G domain. Internal deletions of these motifs cause a reduction in, or loss of activities of, cell adhesion and rescuing mutants. In Fmi and DE-cadherin, such deletions might affect folding of the entire ectodomains. Consequently, it might be that cadherin repeats are misoriented and the mutant molecules no longer retain the homophilic binding property (Oda and Tsukita, 1999).
Within the intracellular C-terminal tail (C tail) of Fmi, a part in the middle includes residues that are conserved among the Fmi family (Shima et al., 2004, see supplementary fig. S1 within). We thus asked how important is the C tail for Fmi function? Expression of the ⌬C tail in S2 cells form caused formation of cell aggregate (Fig. 8D) and its expression in wing epidermis was capable of reorienting planar polarity (data not shown). Yet, the ⌬C tail did not rescue the dendritic overgrowth (Fig. 2H). This result suggests that the C tail plays an indispensable role in dendrite morphogenesis but that it is not essential to control PCP when misexpressed.

Image collection of dendritic trees
For most experiments, larvae and dechorionized embryos were washed in 0.7% NaCl and 0.3% Triton X-100, placed on glass slides and mounted in PBS. Prior to observation the slides were kept at 4°C for a few hours to arrest movement of the specimens. Because dendrites extend on 2D planes almost underneath the epidermis, Z-series of dorsal front images were projected into 2D images, which were then used to measure the distance between dendritic tips and the midline. Time-lapse analysis was done basically as described previously (Sugimura et al., 2003), except that embryos were mounted on dishes with a glass base, and larvae (12 hours after hatching) were mounted in 45% glycerol. Images were collected with a laser scanning confocal microscope LSM510 (Carl Zeiss) and processed with Adobe Photoshop (Adobe Systems). To track entire dendritic branches, we removed signals of gut, epidermis and unknown non-neuronal cells located just above or underneath the branches from individual Z-sections before projection ( Fig. 5D-F).

Cell ablation
In each experiment, an fmi 72 /fmi E59 embryo 20-22 hours after egg laying (AEL) was manually dechorionated, placed on a coverslip and mounted on a slide between spacers made of tape. Individual da neurons in the dorsal cluster were identified under a fluorescence microscope (Olympus) on the basis of their stereotypic arrangement and the shape of cell body. All neurons, except for the class IV da neurons ddaC, of one hemi-segment were ablated by using Micropoint (Photonics Instruments) as described previously (Sugimura et al., 2004). The coverslip was immediately turned over, the embryo was rotated, and then target cells in the contralateral hemi-segment were ablated. The animal was subsequently imaged at about 40 hours AEL.

Histochemistry
Wing imaginal discs and pupal wings were fixed in 3.7% formaldehyde and 0.05% Triton X-100 in PBS, at room temperature for 1 hour or at 4°C overnight, and stained with the following primary antibodies: mouse or rat anti-Fmi (Usui et al., 1999), rat anti-Dsh (Shimada et al., 2001) or rabbit anti-GFP (Molecular Probes). X-Gal staining was carried out according to Hama et al. (Hama et al., 1990) with a slight modification. Wings were dissected from freshly eclosed flies, submerged in 25% EM-grade glutaraldehyde for 2 minutes, immediately washed with PBS, and incubated in X-Gal solution overnight at 37°C. For the staining of adult notums, muscles and other tissues were removed from the thorax, fixed with 1% EM-grade glutaraldehyde in PBS for a few minutes at room temperature, and then processed as described above.

Cell culture and aggregation assay
Fmi or Fmi mutant forms were transiently expressed in S2 cells by co-transfecting individual pUAST plamsids with actin5C-Gal4 (a gift from Yash Hiromi) and pUAST-EGFP. Cell aggregation assays were carried out basically as described (Oda et al., 1994). To examine whether ⌬N::EYFP was able to inhibit Fmi-dependent Journal of Cell Science 119 (6) aggregation or not, we co-transfected S2 cells with actin5C-Gal4, pUAS-EGFP, pUAS-Fmi, and pUAS-⌬N::EYFP or the control pUAST vector at the weight ratio of 1:2:2:20. Spreading of S2 cells on concanavalin A (ConA)-coated dishes was done essentially as described (Rogers et al., 2002).