Differential roles of PtdIns(4,5)P₂ and phosphorylation in moesin activation during Drosophila development

In eukaryotic cells, the organisation and dynamic properties of the plasma membrane depend on a finely regulated interplay between the lipid bilayer and the subjacent cortical actin cytoskeleton, which work together as a single functional unit. Proteins of the ezrin, radixin and moesin (ERM) family act as dynamic linkers between the actin cytoskeleton and the plasma membrane and hence are implicated in the regulation of membrane organisation in numerous cellular and developmental processes (Bretscher et al., 2002; Polesello and Payre, 2004). ERM proteins cycle between an active form, which interacts with both membrane proteins and actin filaments, and a dormant form that localises in the cytoplasm. The dormant conformation is stabilised by an intramolecular interaction that occurs between the N-terminal region (FERM domain) and the C-terminal tail, which contains the F-actin-binding domain (Gary and Bretscher, 1995; Turunen et al., 1994). Adoption of an open, active conformation requires disruption of this interaction in ERM proteins, which unmasks the F-actin-binding region and frees the FERM domain for interaction with its membrane partners.

Two regulatory mechanisms are known to participate in ERM protein activation. First, structural data show that phosphorylation of a conserved threonine residue in the C-terminal domain hinders the intramolecular interaction between FERM and the C-terminus, favouring the open conformation (Matsui et al., 1998; Pearson et al., 2000). Accordingly, many studies suggest that phosphorylation is a key factor controlling ERM activity in vivo (Carreno et al., 2008; Chorna-Ornan et al., 2005; Hayashi et al., 1999; Hipfner et al., 2004; Polesello et al., 2002). Second, evidence obtained from biochemical and cellular assays shows that ERM proteins directly bind to phosphatidylinositol-(4,5)bisphosphate [PtdIns(4,5)P₂], which also promotes their activation (Barret et al., 2000; Nakamura et al., 1999; Niggli et al., 1995; Yonemura et al., 2002). Indeed, studies focusing on the mammalian ezrin have shown that PtdIns(4,5)P₂ binding is a prerequisite for its subsequent phosphorylation (Fievet et al., 2004). Thus, PtdIns(4,5)P₂ is generally envisaged as a membrane cofactor, with specific kinases or phosphatases being the chief players controlling the dynamic aspects of ERM protein regulation. However, recent reports in some cellular contexts show that drastic fluctuations of PtdIns(4,5)P₂ levels have a rapid impact on ERM activation, indicating that the level of PtdIns(4,5)P₂ could have an instructive role on ERM protein activity (Hao et al., 2009; Rasmussen et al., 2008).

To bring new insight into the in vivo mechanisms activating ERM proteins, we evaluated the respective contributions of PtdIns(4,5)P₂ and phosphorylation to the dynamic regulation of ERM proteins during morphogenesis. For this, we developed a genetic approach, focusing on the influence of these two regulatory inputs on the activity of the sole Drosophila ERM representative, the moesin protein (Moe) (McCartney and Fehon, 1996; Polesello et al., 2002). We assessed the localisation and biological activity of engineered Moe variants specifically perturbing these two regulatory events, showing that PtdIns(4,5)P₂ binding targets the Moe protein to specific membrane domains in a phosphorylation-independent process. In fact, as long as it interacts with PtdIns(4,5)P₂, a Moe form that cannot be phosphorylated at T559 displays a substantial degree of activity, which varies depending on the cellular context examined. In addition, PtdIns(4,5)P₂-dependent membrane recruitment of Moe appears both a prerequisite and the rate-limiting step for its subsequent phosphorylation. By contrast, expression of a Moe phosphomimetic form provokes profound deleterious effects on development, which are only partially suppressed when PtdIns(4,5)P₂ interaction is prevented. Altogether, our data indicate that phosphoinositide levels have an intrinsic role during ERM activation in vivo, with dynamic phosphorylation further contributing to the fine-tuning of ERM activity in specific morphogenetic contexts.

Biochemical studies have shown that human ezrin binds to PtdIns(4,5)P₂ (Barret et al., 2000; Niggli et al., 1995). To determine whether this feature also applies to Moe, we analysed its ability to cosediment with liposomes containing increasing concentrations of PtdIns(4,5)P₂ under conditions of physiological ionic strength (Niggli et al., 1995). We observed a clear positive correlation between the amount of PtdIns(4,5)P₂ present in the liposomes and the proportion of Moe associated with these vesicles (Fig. 1A). Conversely, Moe did not associate with liposomes containing only phosphatidylcholine or supplemented with either phosphatidylserine or phosphatidylethanolamine, indicating that Moe interacts with PtdIns(4,5)P₂ and not with other common membrane phospholipids (Fig. 1B). Similarly to ezrin (Barret et al., 2000), substitution in the Moe protein of two conserved lysine doublets (K254-K255 and K263-K264) by asparagines (MoeKN form) abolished binding to PtdIns(4,5)P₂ (Fig. 1C), whereas its weak affinity for other phospholipids was not altered (Fig. 1A,B). Thus, although we cannot rule out an interaction of Moe with other phosphoinositides, these results indicate that this ERM protein, like ezrin, binds to PtdIns(4,5)P₂.

To evaluate the influence of PtdIns(4,5)P₂ binding on the subcellular distribution of Moe, we expressed different mutant GFP-tagged versions of these proteins in Drosophila S2 cells. The MoeWT-GFP protein was seen in the cytoplasm and also associated with F-actin-rich structures in the cell cortex (Fig. 2A). By contrast, MoeKN-GFP did not accumulate in the membrane and localised preferentially in the cytoplasm (Fig. 2B,G). Although significant amounts of a Moe form that cannot be phosphorylated (MoeTA-GFP, carrying the T559A mutation) were still seen associated with the cortex (Fig. 2C,G), the double mutant MoeKNTA-GFP behaved in a similar manner to MoeKN-GFP and was cytoplasmic (Fig. 2D,G). As previously observed (Kunda et al., 2008), the phosphomimetic form MoeTD-GFP (T559D) appeared enriched in the cortex and its expression induced formation of ectopic F-actin structures at the membrane (Fig. 2E,G). In addition, MoeTD-GFP elicited clear morphological defects: cells appeared contracted and rounded (Fig. 2E,H). However, when combined with KN mutations, the phosphomimetic version MoeKNTD-GFP localised at the cortical level, similarly to MoeWT-GFP (Fig. 2F,G). Therefore, the presence of the phosphomimetic mutation promotes Moe access to the cell cortex, even when PtdIns(4,5)P₂ binding is impaired. However, MoeKNTD-GFP expression did not induce morphological alterations, unlike MoeTD-GFP (Fig. 2F,H).

To test the activity of these different Moe variants in vivo, we monitored their ability to substitute for the endogenous product during fly development. We generated transgenic lines expressing Moe mutant forms under the control of an UAS promoter (Brand and Perrimon, 1993) and selected those that, upon induction with the ubiquitous tubulin-GAL4^LL7 driver, produced Moe levels comparable with those seen for the endogenous protein (supplementary material Fig. S1A). We then analysed whether the different Moe versions could rescue the phenotypes associated with the two loss of function alleles Moe^PL106 and Moe^PL54, both caused by P-element insertions in different Moe regulatory regions (Polesello et al., 2002). The Moe^PL106 mutant allele leads to lethality during late development, but we observed Moe^PL106 adult escapers appearing at low frequency in our fly cultures. Wings of these escapers were often crumpled and reduced in size and their eyes displayed a rough-eye phenotype, because ommatidia were reduced in number and appeared disorganised (Fig. 3A). We also tested the rescuing properties of Moe variants using Moe^PL54, the strongest available Moe allele, which leads to 100% lethality in embryonic or early larval stages.

Expression of MoeWT-GFP greatly enhanced the viability and restored the visible phenotypes of both Moe^PL106 and Moe^PL54 mutants (Fig. 3A,B). However, MoeKNTA-GFP showed little rescuing activity, confirming that PtdIns(4,5)P₂ binding and T559 phosphorylation are the major events controlling Moe activation in vivo. In contrast to previous reports (Hipfner et al., 2004; Speck et al., 2003), we found that expression of MoeTD-GFP did not rescue the mutant phenotypes and instead provoked lethality in both Moe mutants and wild-type siblings. Thus, MoeTD-GFP interferes with normal development, mirroring the deleterious effects observed in S2 cells. Unexpectedly, we also observed that all lines expressing MoeTA-GFP significantly enhanced the viability of Moe^PL106 flies. In addition, the wings of these rescued animals showed wild-type morphology, indicating that MoeTA-GFP retains some activity (Fig. 3A). However, Moe^PL106 mutants rescued by MoeTA-GFP still exhibited obvious eye phenotypes, confirming that Moe phosphorylation is essential for proper morphogenesis in this tissue (Fig. 3A). By contrast, MoeKN-GFP, which cannot bind to PtdIns(4,5)P₂, retained only a weak activity, because it slightly enhanced Moe^PL106 viability. In addition, these escapers had rough eyes and crumpled wings, suggesting that Moe PtdIns(4,5)P₂-binding activity is necessary for the morphogenesis of these tissues (Fig. 3A,B). Finally, we observed that the toxic effects induced by expression of MoeTD-GFP were suppressed if its interaction with PtdIns(4,5)P₂ was prevented. Indeed, expression of MoeKNTD-GFP also enhanced the viability of Moe^PL106 flies and restored the wing phenotypes (Fig. 3A,B). However, MoeKNTD-GFP did not rescue the eye phenotype, and elicited even stronger eye defects in Moe^PL106 adults (Fig. 3A).

Since our rescuing assays used GFP-tagged proteins, it remained possible that the GFP moiety could influence their activity. We therefore produced transgenic lines driving robust expression of a set of equivalent full-length untagged Moe variants (supplementary material Fig. S1B). Moe mutant flies rescued after expression of these untagged versions showed phenotypes identical to those observed after expression of the corresponding tagged forms (not shown). However, untagged forms better enhanced the viability of the Moe mutants, suggesting that addition of a GFP moiety is slightly detrimental for Moe activity. Indeed, expression of untagged MoeTA and MoeKNTD now also rescued the lethality of the strong Moe^PL54 allele (Fig. 3B and supplementary material Fig. S2), further indicating that these two forms have significant biological activity. We did not observe dominant phenotypes in wild-type flies after expression with tubulin-GAL4 of the different Moe variants, except in two cases: MoeTD, which caused lethality in early larva and MoeKNTD, whose expression did not affect viability, but elicited a mild rough-eye phenotype.

In summary, our results indicate that in vivo regulation of Moe does not rely exclusively on control of its phosphorylation state. In fact, the form that cannot be phosphorylated (MoeTA) is the variant that displays the highest activity and is able to rescue the strong Moe^PL54 mutant. By contrast, inhibition of binding to phospholipids has a deeper impact on Moe activity, which can nevertheless be compensated to a limited extent by a concomitant phosphomimetic mutation.

Our data indicate that Moe regulation by T559 phosphorylation is essential for eye morphogenesis but could have a less-prominent role during wing development. This unexpected result prompted us to characterise at the cellular level how Moe activity is regulated in the developing wing disc, where Moe has been shown to be required for epithelial integrity (Hipfner et al., 2004; Speck et al., 2003). Consistently, we observed wing cells detached from the main epithelium in Moe^PL106 adult escapers (Fig. 3A). To study this phenotype in a reproducible system, we generated mosaic wing discs containing clones of homozygous Moe mutant cells (Fig. 4A-F). All observed Moe^PL106 mutant clones contained extruding cells in the most basal part of the epithelium (Fig. 4A,D). Most of these cells had pyknotic nuclei (Fig. 4A,D) and expressed high levels of activated Caspase3 (Fig. 4E,F), indicating that they undergo apoptosis. We tested the ability of the different Moe variants to rescue this phenotype when driven exclusively in the Moe^PL106 mutant cells. Expression of either MoeWT-GFP or MoeTA-GFP prevented the formation of extruding cells in all observed clones, confirming that regulation of Moe by phosphorylation is dispensable for epithelium integrity (Fig. 4G,I). Conversely, we observed apoptotic extruding cells with all the other Moe variants, including MoeKNTD-GFP (Fig. 4H,J-L), which indicates that this form is also unable to provide full rescue activity in the wing cells; a fact that went unnoticed in our escaper analysis probably because of the regulative properties of the wing disc. In fact, the development of this appendage can compensate for substantial amounts of cell death without showing obvious defects.

Analysis of the distribution of these Moe variants in wing cells was also informative. MoeWT-GFP accumulated in F-actin-rich cortical structures and reached its highest levels in the apical-most part of the cell (Fig. 5A). We also detected a weaker and more diffuse signal in the cytoplasm and nucleus (Fig. 4G). The subcellular distribution of MoeTA-GFP was similar, but accumulation in the most apical part of the cells was less prominent (Fig. 5B). By contrast, MoeKN-GFP and MoeKNTA-GFP did not concentrate in F-actin-rich regions and showed diffuse staining, presenting a graded distribution with higher levels in the apical portion of the epithelium (Fig. 4H,J and Fig. 5D,E). MoeTD-GFP was never seen in the cytoplasm and was exclusively found in association with the cell cortex, reaching abnormally high levels in the lateral membrane and triggering ectopic accumulation of F-actin in these regions (Fig. 4K and Fig. 5C). Finally, the MoeKNTD-GFP distribution was more similar to that of MoeWT-GFP and differs from MoeTD-GFP in several aspects: it was readily detected in the cytoplasm, it accumulated preferentially in apical regions and its expression did not affect the organisation of the actin cytoskeleton (Fig. 4L and Fig. 5F).

We have shown that MoeKNTD has some biological activity, because it could improve the viability of Moe mutant flies. But can an active phosphorylated form such as MoeKNTD occur in vivo if PtdIns(4,5)P₂-binding is prevented? We tested this possibility using a specific antibody against phosphorylated T559 (Moe-P) (Karagiosis and Ready, 2004). As expected, Moe-P levels were clearly diminished in Moe^PL106 mutant clones (Fig. 6A), whereas the Moe-P signal was recovered upon reintroduction of MoeWT-GFP in the mutant cells (Fig. 6B). However, no Moe-P signal was detected in mutant cells expressing the MoeKN-GFP form (Fig. 6C), a result confirmed in immunoblots of protein extracts obtained from Moe^PL106 adult flies expressing either MoeKN-GFP or MoeKN (Fig. 6D).

Thus, our results indicate that PtdIns(4,5)P₂ binding has an essential role in the wing epithelium: it is a prerequisite for Moe phosphorylation and is essential to localise Moe to the cell cortex. By contrast, T559 phosphorylation appears to be dispensable in this tissue both for cortical accumulation of Moe and Moe activity.

PtdIns(4,5)P₂ binding seems to be required for in vivo regulation of Moe activity. One possibility is that PtdIns(4,5)P₂ could regulate the subcellular localisation of Moe. To test this hypothesis, we visualised the distribution of PtdIns(4,5)P₂ in wing cells by studying the localisation of a phospholipase C (PLC) δ1 pleckstrin-homology domain coupled to GFP (PH-GFP), a protein probe that binds specifically to PtdIns(4,5)P₂ (Stauffer et al., 1998; Zelhof and Hardy, 2004). In wild-type wing cells, PH-GFP was seen associated with the cortex throughout the cells, accumulating at high levels in the apical domain (Fig. 7A). This distribution was similar to that observed for MoeTA-GFP (Fig. 7E), which is consistent with the idea that in the absence of T559 phosphorylation, PtdIns(4,5)P₂ controls the subcellular localisation of Moe.

To further evaluate this possibility, we increased the cellular levels of PtdIns(4,5)P₂ and monitored the consequences on Moe localisation. Type I phosphatidylinositol kinases (PIP5KIs) have a prominent role in the synthesis of PtdIns(4,5)P₂ from the PtdIns(4)P precursor (Yin and Janmey, 2003). In Drosophila, the skittles (sktl) locus codes for a PIP5KI (Hassan et al., 1998). Upon ectopic expression of Sktl, PH-GFP accumulated at high levels in the most apical part of the cells (Fig. 7A,B), where we also observed the formation of abnormal membrane protrusions (Fig. 7B,D,F,I,J). This indicates that Sktl overexpression boosts the PtdIns(4,5)P₂ levels in the apical domain of the wing cells, interfering with normal membrane organisation. Raising the PtdIns(4,5)P₂ levels also affected the localisation of both MoeTA-GFP (Fig. 7E,F) and MoeWT-GFP (Fig. 7C,D), which were then over-represented in the apical cortex. By contrast, the distribution of MoeKN-GFP was not altered in the presence of ectopic PtdIns(4,5)P₂ (Fig. 7G,H). We also monitored the phosphorylation state of endogenous Moe in the presence of ectopic Sktl. Our immunostaining revealed abnormal accumulation of Moe-P staining in the apical part of Sktl-overexpressing cells (Fig. 7I,J), and we also found increased levels of Moe-P in protein extracts obtained from the same discs (Fig. 7K). Thus, our results indicate that high PtdIns(4,5)P₂ levels can recruit Moe to the cell cortex, affecting both its subcellular distribution and its availability as a substrate for phosphorylation.

We also tested whether loss of the sktl gene is accompanied by a reduction of the Moe-P levels in the wing epithelium. In the ovaries, germ line clones for null sktl alleles induce a developmental arrest in early stages, but a clear reduction of Moe-P levels was observed in sktl hypomorphic backgrounds, which do not hinder ovary early development (Gervais et al., 2008). Wing epithelial integrity is compromised in homozygous mutant cells for the hypomorphic sktl^Δ5 allele (Hassan et al., 1998), because we observed the presence of extruding cells in the most basal part of the disc when apoptosis was prevented by concomitant expression of the p35 anti-apoptotic factor (Hay et al., 1994) (supplementary material Fig. S3A,B). However, we did not detect a significant decrease of the Moe-P levels in these mutant cells (supplementary material Fig. S3C-F), a negative result that is difficult to interpret because PtdIns(4,5)P₂ levels could only be slightly reduced in these clones.

The aim of this work was to better understand how ERM protein activation is regulated in a developing organism and for this, Drosophila morphogenesis constitutes an attractive system in which to analyse the cellular mechanisms involved. We show that two highly conserved mechanisms control Moe activation: PtdIns(4,5)P₂ binding and phosphorylation of the invariant T559 residue. Thus, the study of the regulation of Moe in flies offers conclusions that could be extrapolated to ERM proteins of other species, including mammals. We show that substituting the endogenous Moe by point mutations affecting T559 phosphorylation or PtdIns(4,5)P₂ binding leads to distinguishable morphogenetic defects, thereby allowing genetic dissection of their respective contributions. Our results indicate that the role of PtdIns(4,5)P₂ is not restricted to a pre-activation step and instead, intrinsically regulates Moe subcellular localisation and activity.

Previous studies in mammalian cells have shown that ezrin directly binds to PtdIns(4,5)P₂, allowing identification of two lysine doublets in the N-terminal FERM domain that are required for this interaction (Barret et al., 2000). These residues are conserved in the Moe protein and their replacement by asparagines (MoeKN) similarly blocks interaction with PtdIns(4,5)P₂, suggesting that PtdIns(4,5)P₂-binding is an ancient feature shared by all extant ERM proteins.

Our data support the view that these lysine residues act as a PtdIns(4,5)P₂ sensor, permitting efficient membrane recruitment of ERM proteins from a cytosolic pool of dormant molecules (Fig. 8). In the polarised wing cells, PtdIns(4,5)P₂ is found in the plasma membrane and accumulates in their most apical part – a region containing high levels of Moe. We show that the KN mutations block Moe recruitment to the cell cortex. Conversely, experimental elevation of the PtdIns(4,5)P₂ levels in the apical portion of the cell is accompanied by accumulation in the same region of MoeTA-GFP and MoeWT-GFP, but not MoeKN-GFP. Thus, these observations suggest that PtdIns(4,5)P₂ availability controls Moe access to the cell cortex and has an instructive role during this recruitment step, which can be achieved independently of Moe phosphorylation. Our results confirm that, as with mammalian ezrin (Fievet et al., 2004), Moe phosphorylation cannot be achieved in wing cells if its interaction with PtdIns(4,5)P₂ is hindered by the KN mutations. Thus, PtdIns(4,5)P₂ binding appears to be a prerequisite for subsequent Moe phosphorylation. Moreover, we also observed increased amounts of phosphorylated Moe as a consequence of raising the PtdIns(4,5)P₂ levels in the wing, suggesting that kinase availability is not the limiting factor determining the degree of Moe phosphorylation in this tissue. Previous observations made in the fly ovary reported a reduction of phosphorylated Moe levels in mutant clones for a PIP5K sktl hypomorphic allele (Gervais et al., 2008). However, we did not observe the same reduction in the wing cells. This discrepancy might reflect the existence of genuine tissue differences affecting Moe regulation, but it is difficult to draw a conclusion at this stage, owing to possible genetic redundancy with other PIP5Ks encoded by Drosophila (for instance, the PIP5K59B gene) and the specific nature of the available sktl hypomorphic alleles, which all affect regulatory sequences that could be tissue specific (Gervais et al., 2008; Hassan et al., 1998).

Our genetic assays revealed that MoeKN still retains a low biological activity. If the KN mutations prevent PtdIns(4,5)P₂ binding, how can this protein still link F-actin to the cell cortex? Interestingly, recent work has shown that membrane association of vertebrate moesinKNTD and ezrinKNTD is sensitive to PtdIns(4,5)P₂ (Hao et al., 2009). Thus, although mutation of the lysine doublets prevents efficient access to the membrane, it does not completely abolish interaction with PtdIns(4,5)P₂, which contributes to maintain ERM KNTD forms in the cortex. Crystallographic analyses of InsP₃-complexed Radixin show that the presumptive PtdIns(4,5)P₂-interacting region is a cleft that is present in a highly positively charged surface of the FERM domain (Hamada et al., 2000). Interestingly, the two conserved lysine doublets are in this charged surface, but not located in the cleft that interacts with InsP₃. It is tempting to speculate that PtdIns(4,5)P₂ could first interact with the charged surface of ERM proteins and favour membrane recruitment before lodging into the cleft and promote ERM activation. Our results suggest that a small fraction of MoeKN could reach the membrane compartment, presumably by stochastic motion, and then interact with PtdIns(4,5)P₂ via its intact FERM cleft. The fly MoeKNTD form is more efficiently recruited to the cortex and displays a higher activity than MoeKN because this open molecule can also interact with F-actin. Indeed, MoeKNTD is found preferentially in regions rich in F-actin. Finally, MoeTD displays the highest affinity for the cortex because of its ability to bind both PtdIns(4,5)P₂ and F-actin. Moreover, its presence in the cortex triggers accumulation of F-actin, which in turn could allow further recruitment of MoeTD, as seen in both S2 cells and lateral regions of wing cells.

A surprising finding of our studies is that a Moe form that cannot be phosphorylated in T559 still displays significant activity in various developmental processes. Although overexpression of such forms partly counteracts endogenous ERM protein activity in both wild-type and pathological contexts (Carreno et al., 2008; Chorna-Ornan et al., 2005; Khanna et al., 2004; Polesello et al., 2002; Yu et al., 2004), our genetic assays show that MoeTA has a rescuing activity in the absence of the endogenous product. MoeTA, which can associate with the membrane, significantly improves the viability of Moe mutant flies, including the strong Moe^PL54 allele. Detailed examination of wing mosaics containing Moe mutant cells further illustrate that phosphorylation of Moe T559 is dispensable for the integrity of this epithelium. These observations indicate that, at least for certain aspects of tissue homeostasis, regulation via PtdIns(4,5)P₂ binding could be sufficient to control Moe distribution and activity, although we cannot formally exclude the possibility that Moe phosphorylation in other unknown residues also contributes to activation of the MoeTA form. Although our results appear in conflict with previous reports (Hipfner et al., 2004; Speck et al., 2003), it is worth noticing that these latter studies used Moe forms carrying a N-terminal Myc tag, which could significantly affect Moe activity. Our results show that Moe proteins tagged with GFP at the C-terminus have a reduced activity when compared with their full-length equivalents. If addition of an N-terminal tag further reduces the activity of tested proteins, it would explain why previous studies failed to detect a rescue with MoeTA (Hipfner et al., 2004; Speck et al., 2003). However, this tag would reduce the toxic effect of MoeTD sufficiently to obtain a weak rescue (Hipfner et al., 2004; Speck et al., 2003), similarly to what we observe expressing an untagged MoeKNTD form.

Our results therefore allow us to postulate the existence of a pool of open, non-phosphorylated Moe molecules that nonetheless can fulfil in vivo their role as linkers between the actin cytoskeleton and the membrane (Fig. 8). The idea that different active ERM forms, phosphorylated and not phosphorylated, could coexist at the cell cortex is also consistent with results from kinetic analyses showing the existence of three different pools of ezrin in the membrane of cultured cells (Coscoy et al., 2002). Although stable and highly mobile molecules could respectively correspond to phosphorylated and PtdIns(4,5)P₂-bound ezrin_, the pool of molecules displaying intermediate mobility in the cell cortex could reflect the presence of unphosphorylated ezrin interacting with actin filaments and membrane proteins. In any case, our data provide unambiguous evidence that active ERM protein distribution cannot be faithfully deduced from the ERM phosphorylation pattern.

Previous work has shown that Moe has an essential role during eye differentiation, where it is required for the morphogenesis of the highly specialised microvilli that constitute the photoreceptor rhabdomeres (Karagiosis and Ready, 2004). These large membrane protrusions accumulate high levels of PtdIns(4,5)P₂ and phosphorylated Moe (Karagiosis and Ready, 2004; Stauffer et al., 1998; Zelhof and Hardy, 2004). Consistently, our rescue experiments show that eye morphogenesis requires proper Moe regulation by both phosphorylation and PtdIns(4,5)P₂ binding. Although MoeTA is sufficient to restore the viability of Moe mutants, the eye phenotype is not suppressed in rescued animals, showing that T559 phosphorylation is crucial in this tissue. Phosphorylation might be essential to provide a stable association between Moe and its partners in these specialised giant villi of the apical domain. The situation in the undifferentiated wing epithelium appears different, because expression of MoeTA in this tissue is enough to ensure epithelial integrity. These results indicate that, depending of the cellular context, transient or stable association of ERM proteins with the cortex can be the preferred choice. Modulation of PtdIns(4,5)P₂ levels would allow a dose-sensitive activation of ERM molecules at the membrane, with phosphorylation being a secondary step necessary for the formation and maintenance of specialised cellular structures.

The aberrant morphology of S2 and wing cells and the developmental lethality induced by MoeTD show that an open form, mimicking a constitutively phosphorylated molecule has strong effects on cell organisation and morphogenesis. Thus, phosphorylation-dephosphorylation cycles are also an important aspect in the regulation of Moe function during development. The ability to cycle between phosphorylated and dephosphorylated Moe has been shown to be necessary for photoreceptor physiology, suggesting that Moe dephosphorylation mediates remodelling of photoreceptor cytoarchitecture in response to light (Chorna-Ornan et al., 2005). Recent work further suggests that local hydrolysis of PtdIns(4,5)P₂ after phospholipase-C (PLC) activation, as it is the case in phototransduction, can contribute to the fast removal of ERM proteins from the plasma membrane (Hao et al., 2009).

A system of double regulation based on activation by PtdIns(4,5)P₂ levels and stabilisation by phosphorylation has evolved to finely tune the activity of ERM proteins in a whole set of different cellular structures and developmental situations. Clearly, this feature contributes to the versatility of ERM proteins and helps to explain why they have a central role in such a varied number of morphogenetic processes.

Moe cDNA was subcloned in the pGEX-4T1 plasmid, allowing production of GST-Moe fusion protein. MoeKN was generated by mutation of lysine 254, 255, 262, 263 to asparagine, as previously done for human ezrin (Barret et al., 2000). The KN mutations were generated by replacement of the NcoI-BspeI wild-type fragment in pGEX-Moe with a set of overlapping mutant oligonucleotides (Eurogentec). To produce MoeKN GFP-tagged versions, the Pfl23II-BspEI fragment carrying the KN mutation was inserted in the corresponding MoeWT-GFP, MoeTA-GFP, MoeTD-GFP plasmids (Polesello et al., 2002). These constructs were then subcloned into pAc5.1 (Invitrogen) for S2 cell transfection and into pUASp for generation of transgenic lines. For transgenic lines encoding untagged proteins, the full-length Moe coding region was amplified by PCR and cloned into pUASp.

Plasmids encoding GST-Moe were used to transform TG1 bacteria. After overnight growth, IPTG (0.5 mM) was added for 1 hour and bacteria centrifuged at 4000 r.p.m. for 15 minutes. All further details refer to 1 litre culture. Bacteria pellets were sonicated in 15 ml PBS containing 5 mM EDTA and protease inhibitors (Complete from Roche). Samples were then centrifuged at 40,000 g for 30 minutes. Supernatants were diluted to 50 ml and 3 ml glutathione agarose beads slurry (v/v) was added. After incubation for 2 hours at 4°C, beads were washed with 50 mM Tris-HCl, pH 7.4 and 100 mM NaCl. When the OD₂₈₀ was 0, 100 U benzonase was added for 45 minutes at room temperature. After washing the beads, gels were equilibrated in the same buffer supplemented with 2.5 mM CaCl₂. Thrombin (50 U) was then added and samples incubated for 1 hour at 37°C. Cleaved proteins were eluted and their concentration calculated, taking into account their extinction coefficient. To avoid interference of calcium with the different assays involving phospholipids, EGTA (5 mM) was added to the recombinant proteins, which were stored at 4°C and used within 48 hours.

Phospholipids (PC, PS and PE) and PtdIns(4,5)P₂ were purchased from Sigma and Lipid Products (Redhill, UK). Lipids, resuspended in water, were mixed in the appropriate proportions and dried using a Speedvac. Dry pellets were then allowed to swell in 10 mM Tris-HCl, pH 7.4 containing 1 mM EGTA at 42°C for 2 hours. After liposome incubation with the Moe proteins overnight, samples were centrifuged for 30 minutes in a 42.2 Ti rotor (Beckman) at 100,000 g. Supernatant and pellet samples were then analysed by western blotting. Antibodies against Moe and secondary anti-rabbit antibodies coupled to HRP were used at a 1:10,000. Primary antibodies were first incubated with GST-agarose beads to minimize crossreactivity with GST. Blots were developed using the ECL kit (Amersham) and quantification was performed using the ImageJ software.

S2 cells were maintained in Schneider's Drosophila medium (GIBCO) supplemented with 10% heat-inactivated FCS (GIBCO) and penicillin-streptomycin. Cells were transfected using Fugene-6 according to the manufacturer's instructions (Roche Diagnostics) and observed 24 hours later. Cells were allowed to spread for 1 hour on No. 1.5 coverslips (Corning) previously treated with 20 μl of 0.5 mg/ml concanavalin A (Sigma) in water and allowed to air dry (Rogers et al., 2002). Cells were then fixed for 20 minutes in BRB80 buffer (80 mM PIPES, pH 6.9, 1 mM MgCl₂, 1 mM EGTA) containing 3% formaldehyde (EM Sciences) and 1 mg/ml saponin. Blocking and permeabilisation was performed in TBS (20 mM Tris-HCl, pH 7.5, 154 mM NaCl, 2 mM EGTA, 2 mM MgCl₂) containing 2% BSA and 0.02% saponin for 1.5 hours. Coverslips were then incubated for 30 minutes on 20 μl droplets of Texas-Red-phalloidin 1:100 (Molecular Probes) in TBS-BSA-saponin. After washes in TBS-BSA-saponin (15 minutes) and TBS (twice for 15 minutes), coverslips were mounted onto glass slides using Hardset VectaShield (Vector). Images were acquired using a Leica DMIRE2 inverted microscope with a MICROMAX-1300Y-HS cooled camera (Princeton Instrument). Fluorescence levels were quantified by analysing image stacks (recorded using the same settings) with the ImageJ software (http://rsbweb.nih.gov/ij). Morphometry analyses were performed using calculation of the circularity index of ImageJ.

Wing discs were dissected in PBS, fixed for 20 minutes in paraformaldehyde 4% in PBS and incubated overnight in Rhodamine phalloidin (Sigma) and 10 mM Topro3 (Molecular Probes) in PBS-Tween 0.1%. After three 30 minutes washes in PBS 0.1% Tween20, discs were mounted in Vectashield (Vector). Primary antibodies were rabbit anti-Moe-P, 1:500 (Karagiosis and Ready, 2004), rabbit anti-activated Caspase3, 1:100 (Cell Signalling), mouse anti-arm, 1:100 (DSHB). We used as secondary antibodies Rhodamine anti-rabbit, FITC anti-mouse and Cy5 anti-mouse, all diluted 1:300 (Jackson Laboratories). For immunostaining, discs were fixed as above and incubated overnight with the corresponding primary antibody dilutions in PBS 0.3% Triton X-100, 1% BSA. Samples were visualised with a Leica TCS SP2 confocal microscope. False colours images were generated applying the Rainbow LUT of the ImageJ software to disc pictures obtained using the same confocal settings.

The following strains were used: ywMoe^PL54, ywMoe^PL106 (Polesello et al., 2002), ywFRT19A, tubulin-GAL4^LL7 (Lee and Luo, 1999), apterousGAL4^md544 (Calleja et al., 1996), UASMoeWT-GFP, UASMoeTA-GFP, UASMoeTD-GFP (Polesello et al., 2002), UASPH-GFP (Zelhof and Hardy, 2004), sktl^Δ5 and UASSktl (Hassan et al., 1998) and UASp35^BH2 (Hay et al., 1994). The transgenic lines UASMoeWT, UASMoeKN, UASMoeTA, UASMoeTD, UASMoeKNTA, UASMoeKNTD, UASMoeKN-GFP, UASMoeKNTA-GFP and UASMoeKNTD-GFP were generated in this work. For rescue experiments, ywMoe^PL54/FM6; tubulin-GAL4^LL7/TM2 and ywMoe^PL106/FM6; tubulin-GAL4^LL7/TM2 females were crossed to males carrying the different UASMoe^Mut insertions in an autosome. We calculated for each cross the ratio between the observed number of ywMoe⁻; tubulin-GAL4LL7/UASMoe^Mut surviving males and the ywMoe⁻/+; TM2/UASMoe^Mut female siblings. This ratio should be 1 for a total rescue. Wing disc clones were generated using the MARCM system (Lee and Luo, 1999), giving 1 hour heatshock at 37°C to first instar larvae. Moe^PL106/PL106 positively labelled cells were generated in ywMoe^PL106 FRT19A/hsFLP¹ tubulin-GAL80^LL1 FRT19A; UASmCD8GFP^LL5/+; tubulin-GAL4^LL7/+ females. The Moe^PL106/PL106 mutant clones rescued by the different Moe^Mut-GFP versions were obtained using the stocks ywMoe^PL106 FRT19A/hsFLP¹ tubulin-GAL80^LL1 FRT19A; UASMoe^Mut-GFP/+; tubulin-GAL4^LL7/+. The sktl^Δ5 mutant clones were observed in larvae of the genotype elav^C155 UASmCD8GFP^LL4 hsFLP¹; FRTG13 tubulin-GAL80^LL2/FRTG13 sktl^D5; tubulin-GAL4^LL7/UASp35 ^BH2. All crosses were done at 25°C.

Adult flies were dehydrated in absolute ethanol, coated with colloidal gold and visualised with a Hitachi scanning electron microscope. Wings were mounted in 6:5 ethanol:lactic acid and visualised with a Zeiss Axiophot microscope.

30 wing imaginal discs per genotype were dissected in PBS, pooled together, brought to 20 μl with PBS and homogenised by boiling in 10 μl of 3× Laemmli buffer. Then, 20 μl homogenate was loaded on an SDS-PAGE gel for immunodetection with anti-Moe-P antibody. 3 μl of the same sample was loaded on a different gel for immunodetection with anti-Moe used at 1:50,000. Four adult flies for each genotype were homogenised in 100 μl of boiling Laemmli 3× buffer and 5 μl homogenate was loaded on an SDS-PAGE gel and blotted with anti-Moe or anti-Moe-P in the same conditions. Signal was detected using the ECL kit from Amersham.

We are grateful to H. Bellen, D. Ready, A. Zelhof, the Bloomington Stock Center and the Iowa DSHB for stocks and antibodies. We also wish to thank three anonymous referees for their constructive comments on the manuscript. This work was supported by grants from the ACI Biologie du Développement et Physiologie Intégrative (030023), ARC (3832), FRM (Equipe 2005) and the Canadian Institutes for Health Research (MOP-89877). C.P. was supported by the Ministère de la Recherche and ARC (1114), C.R. by the Ministère de la Recherche and a Lavoisier fellowship. S.C. is a FRSQ ‘Junior I’ fellow. F.R. benefited from a Long Term EMBO fellowship.