A central multifunctional role of integrin-linked kinase at muscle attachment sites

In multicellular organisms, tissue integrity depends upon the stable association of cells with the extracellular matrix (ECM). This requires a link between matrix ligands and the cytoskeleton, which is primarily mediated by the integrin family of surface receptors. Each integrin is composed of one α and one β subunit, and because these subunits have very short cytoplasmic tails it has been postulated that other scaffolding proteins assist integrins in the recruitment of downstream components that mediate the link to actin (Legate and Fassler, 2009). The cytoskeletal molecules involved in this link, and the hierarchy of their interactions, are currently a subject of intense investigation, and genetic approaches in model organisms have provided novel insights into the function of these proteins (Legate and Fassler, 2009). One such example is integrin-linked kinase (ILK), a protein initially identified by its ability to bind directly to the cytoplasmic tail of the β1 integrin subunit (Hannigan et al., 1996). ILK is a modular protein containing five tandem ankyrin repeats (ANKRs), a putative phosphoinositide-binding site and a kinase-like domain (Chiswell et al., 2008). ILK is required to maintain the molecular link between integrins and the sarcomeric actin filaments in Drosophila embryonic muscles (Zervas et al., 2001) and to assemble the link between integrins and the contractile apparatus of developing Caenorhabditis elegans muscles (Mackinnon et al., 2002). Knockout of the single mouse gene showed that ILK is required for diverse developmental processes, with its earliest function being in epithelial polarization and normal actin distribution of the developing epiblast (Lange et al., 2009; Sakai et al., 2003). In all three organisms, certain site-directed mutations within the kinase domain of ILK that should eliminate kinase activity did not show any defects, suggesting that the essential role of ILK is not the phosphorylation of target proteins (Lange et al., 2009; Mackinnon et al., 2002; Zervas and Brown, 2002). Instead, ILK binds to multiple proteins, suggesting that its primary function is as an adaptor. The recently published structure of the kinase domain of ILK supports the view that it is a binding adaptor rather than a kinase (Fukuda et al., 2009). Thus, genetic and structural data indicate that ILK is a pseudokinase (Boudeau et al., 2006; Wickstrom et al., 2010). However, we have yet to develop a clear picture of the essential adaptor function of ILK in assisting integrin function.

The large number of proteins that bind ILK suggests it is a central scaffolding molecule for the assembly of a protein complex (Zervas and Brown, 2002). Interacting proteins have been identified by yeast two-hybrid screening and biochemical methods (Boulter and Van Obberghen-Schilling, 2006). Among these proteins are: PINCH, which contains five LIM domains and binds to the ANKRs of ILK through its first LIM domain (Tu et al., 1999; Velyvis et al., 2001); paxillin, which has four LIM domains and five leucine-rich motifs (LD) and binds to the ILK kinase domain (Nikolopoulos and Turner, 2001); and parvin, which has two calponin homology domains and binds to the ILK kinase domain (Nikolopoulos and Turner, 2000; Olski et al., 2001; Tu et al., 2001). Both the ANKRs and kinase domains are essential for recruitment of ILK at focal adhesions in vertebrate cells in culture (Boulter and Van Obberghen-Schilling, 2006; Li et al., 1999). Even before its association with integrins, ILK is in a cytoplasmic complex with PINCH and parvin (Wu, 2005) and a fourth protein, Ras suppressor protein 1 (Rsu1), which binds the fifth PINCH LIM domain (Dougherty et al., 2005; Kadrmas et al., 2004). So far the model of functional coordination between ILK, PINCH and parvin has not been fully supported by genetic studies. ILK is required for parvin (Pat-6) recruitment to sites of integrin adhesion in C. elegans but not the reverse (Lin et al., 2003), and PINCH is not required for ILK recruitment in either Drosophila or C. elegans (Clark et al., 2003; Norman et al., 2007).

Here, we combined genetic analysis and structure–function approaches to examine the mutual interactions of ILK and PINCH, as well as their interactions with talin and paxillin. We discovered that low levels of integrins were sufficient to recruit substantial levels of ILK to the major sites of integrin adhesion in the embryo and larva, namely the muscle attachment sites (MASs), suggesting an amplification mechanism. Complete removal of talin resulted in the loss of ILK, PINCH and paxillin from MASs, in agreement with its crucial contribution to integrin function. Neither ILK nor PINCH was required for paxillin recruitment. Unexpectedly, PINCH stability and recruitment required both domains of ILK (i.e. PINCH-binding ANKRs and kinase domains). In embryos each ILK domain was recruited to the MASs, but recruitment of the isolated ANKRs was lost during development, indicating a change in mechanism. Finally, we tested the function of a number of conserved ILK residues that have been shown to be important for interactions with other proteins, but found them dispensable. Collectively, our results indicate that ILK functions upstream of PINCH in the muscle and that the function of ILK is executed by simultaneous interactions of both the kinase and ANKR domains.

As Drosophila embryos develop, integrins and their associated proteins become most highly concentrated where muscle ends attach to epidermal tendon cells, through intervening ECM. The muscles are large multinucleate cells, so concentration at muscle ends reflects subcellular localization of these proteins into adhesive junctions capable of resisting muscle contraction. These muscle attachment sites (MASs) therefore provide a useful system for examining protein interactions necessary for recruitment of components of the complex that links integrins to the actin cytoskeleton, which we exploit here. We utilized two views of MASs, optical sections showing muscles and tendon cells, and a surface view of the embryo showing the muscle pattern (see Materials and Methods).

We previously reported that localization of ILK to MASs did not require integrin function, as it was still observed in embryos lacking the βPS integrin subunit (Fig. 1A,A′) (Zervas et al., 2001). Similarly, genetic ablation of β1 in mice also did not affect localization of ILK to the myotendinous junctions (Schwander et al., 2003). Furthermore, talin, a core component of the integrin–actin-linker complex, remains strongly localized at Drosophila MASs even when there is a dramatic reduction in integrin levels, caused by a hypomorphic mutation (if^sef) in the inflated gene, which encodes αPS2, the major, if not only, α-integrin subunit expressed in Drosophila muscles (Devenport et al., 2007). Similarly, ILK–GFP was strongly localized to the MASs in the same mutant (Fig. 1B,B′). We were therefore surprised to find that complete removal of αPS2 resulted in a failure in the localization of ILK to MASs (Fig. 1C–D′). An explanation for this difference is that although the embryos examined previously almost completely lacked the βPS integrin subunit (as they were homozygous for a null mutation in the gene encoding βPS and therefore lacked zygotic βPS expressed from the embryonic genome) they did contain a small amount of βPS deposited in the egg by the heterozygous mother, which could have recruited ILK. This maternally provided βPS is not easily detected with antibodies, but its presence has been revealed genetically, as removing it enhances the mutant phenotype caused by the absence of zygotically expressed βPS (Roote and Zusman, 1995). When both the maternal and zygotic βPS integrin subunit was removed, we no longer detected localization of ILK–GFP at MASs (Fig. 1E,E′). Thus, ILK recruitment to MAS does require integrins, albeit in very small amounts.

To delineate further the hierarchy of genetic interactions in the assembly of the ILK-containing complex, we examined the requirement for talin, as this protein is necessary for all integrin adhesive functions (Brown et al., 2002; Calderwood, 2004). The recent identification of Wech, an adaptor protein that binds ILK and talin, provides a mechanism for ILK recruitment through talin (Löer et al., 2008). We tested whether talin was required for not only ILK but also PINCH and paxillin localization at MASs. Talin, as for integrins, was essential for the recruitment of the integrin-associated proteins ILK (Fig. 2A,A′,B,B′,C,C′,D,E), PINCH (Fig. 2A,A″,B,B″,C,C‴), paxillin (Fig. 2D,D′,E,E′) and tensin (Brown et al., 2002; Torgler et al., 2004). In the same manner as for βPS, the zygotic removal of talin did not eliminate ILK from the MAS, nor did it cause the complete loss of PINCH and paxillin (Fig. 2B′,B″,D′). This result was not unexpected, as zygotic talin mutant embryos contain clearly detectable maternal talin (Brown et al., 2002). However, all three proteins were no longer present at MASs when we completely removed talin by generating germline clones (Fig. 2C-C′″,E,E′). Thus, integrins need talin to recruit ILK, PINCH and paxillin, and a small amount of talin can recruit substantial amounts of these proteins.

Wech functions as an adaptor between talin and ILK, and contributes to ILK localization at MASs (Löer et al., 2008). However, this cannot be a universal mechanism, as other sites of integrin adhesion lack Wech (Delon and Brown, 2009). Studies in vertebrate cells have identified additional ILK interactions that provide alternative mechanisms to link ILK to talin. Thus, the ILK–paxillin interaction is required for localization of ILK to focal adhesions (Nikolopoulos and Turner, 2001), and we have shown above that recruitment of both ILK and paxillin requires talin. Talin recruits an isoform of phosphatidylinositol 4-phosphate 5-kinase (Di Paolo et al., 2002), potentially elevating phosphatidylinositol (4,5)-bisphosphate [PtdIns(4,5)P₂] levels, which could recruit ILK through a region that has sequence similarity to phosphoinositide-binding pleckstrin homology (PH) domains (Delcommenne et al., 1998). Finally, undiscovered interactions with the ANKRs or kinase domain could contribute to ILK recruitment.

To address the mechanism of ILK recruitment to MASs, we examined how mutations in ILK interaction domains affected its localization. Wild-type and truncated ILK proteins were fused to GFP, so that their localization could be quantified by confocal microscopy (Figs 3, 4). ILK–GFP was expressed in two ways: first, using its endogenous regulatory elements to express the proteins at normal spatial and quantitative levels (Fig. 3), and, second, by using mef2Gal4 to drive expression of UAS::ILK–GFP, which results in the overexpression of ILK–GFP only within the muscle (Fig. 4), permitting assessment of its recruitment even if the altered protein has reduced stability. We examined ILK recruitment at two stages: in the newly hatched first-instar larvae (1stIL; at 22–28 hours) and in second-instar larvae (2ndIL; at 48–60 hours).

When wild-type ILK–GFP was expressed under its own promoter nearly 95% of the total fluorescent intensity was concentrated at MASs in 1stIL (Fig. 3A,H), whereas in 2ndIL it reached nearly 99% (Fig. 3I,P). When ILK–GFP was overexpressed only ~50% was recruited to MASs in 1stIL, with the rest diffuse in the cytoplasm or associated with sarcomeric Z-lines (Fig. 4A,I). However, in 2ndIL the fraction of overexpressed ILK–GFP at MASs was close to 95% (Fig. 4J,R), suggesting either that the number of ILK recruitment sites had increased or that unlocalized ILK had been degraded.

The paxillin-binding region in human ILK has been mapped to a motif encompassing residues V386 and T387, and mutagenesis in this region reduces the level of ILK recruitment to focal adhesions (Nikolopoulos and Turner, 2001). In Drosophila, only the corresponding T384 is conserved, so we replaced this residue with alanine (T384A), or with aspartic acid (T384D) to mimic its phosphorylation. Both mutant forms of ILK were localized equivalently to the wild-type protein, whether expressed at endogenous levels or overexpressed (Fig. 3F–H,N–P; data not shown), demonstrating that T384 is not involved in protein interactions that mediate localization of ILK. This suggests that paxillin does not contribute to the localization of ILK in MASs, in agreement with the recent finding that functional MASs are formed in the absence of paxillin (Bataillé et al., 2010).

We next examined the role of the ILK region with similarity to PH domains (180–215), which could recruit ILK by binding phosphoinositides in the plasma membrane (Persad et al., 2001). We replaced the conserved residues R208 and R210 with oppositely charged aspartic acid residues (R208D, R210D), but this did not alter ILK recruitment (Fig. 3B,H,J,P and Fig. 4B,I,K,R). This is consistent with the finding that mammalian ILK with part of the PH-like motif deleted did localize to focal adhesions (Zhang et al., 2002).

The localization of ILK to focal adhesions in mammalian cells requires an interaction between PINCH and the ANKRs of ILK (Li et al., 1999); however, PINCH is not required for recruitment of ILK to MASs (Clark et al., 2003) (see also Fig. 6). Consistent with this, expression of ILK with a deletion of ANKRs 2–5, which bind PINCH (Chiswell et al., 2008), leaving the kinase-like domain intact, reduced but did not eliminate the localization at MASs. In 1stIL, 30% of the ILK kinase domain was recruited to MASs, (Fig. 3C,H), whereas in 2ndIL only 20% was recruited (Fig. 3K,P). Similarly, the recruitment of the overexpressed ILK kinase domain to MASs was reduced (Fig. 4C,L). In 1stIL, 25% was recruited but this increased to 60% at 2ndIL (Fig. 4I,R). Therefore, the ANKRs are important for recruiting normal levels of ILK, and more so in 1stIL than 2ndIL, but are not absolutely required for recruitment. Combining the point mutations in the PH domain with the deletion of the ANKRs did not further impair recruitment (Fig. 3D,H,L,P and Fig. 4D,I,M,R).

Another residue required for ILK localization to focal adhesions is F436, which is located in the last α-helix of the kinase C-lobe (Fukuda et al., 2009; Zhang et al., 2002). An F436A mutation reduced recruitment of overexpressed ILK; 25% was recruited to 1stIL MASs (Fig. 4F,I), increasing to 70% in 2ndIL (Fig. 4O,R). Thus, as with the ANKRs, this residue contributes to MAS recruitment but is not essential.

We next tested whether the isolated N-terminal region (1–177), including the ANKRs (1–128), contained a localization signal. Some of the truncated protein (ILK^ANKRs–GFP) was localized to the MASs, with ~18% recruited in 1stIL (Fig. 3E,H), and 10% recruited when overexpressed (Fig. 4E,I). The reduced recruitment of the ILK truncated protein containing the ANKRs became more evident in 2ndIL (Fig. 3M,P; Fig. 4N,R). This suggested that, in contrast with full-length ILK or the ILK kinase domain, where more of these forms of ILK became localized as the larva grows, there is a reduction in the ability to recruit ILK containing ANKRs. We tested whether this was caused by a limited amount of PINCH, by cooverexpressing untagged PINCH, but this failed to increase ILK^ANKRs–GFP recruitment (data not shown). Therefore, the improved recruitment of overexpressed ILK during larval development is mediated by interactions with the kinase domain.

Finally, we investigated whether phosphorylation regulates ILK localization and function. Mammalian ILK contains four putative p21-activated kinase 1 (PAK1) phosphorylation sites, containing the consensus K/RxxT/S motif (Acconcia et al., 2007); PAK1-dependent phosphorylation of ILK enhances its subcellular distribution within the cytoplasm, whereas depletion of PAK1 promotes translocation of ILK to the nucleus (Acconcia et al., 2007). The T180 PAK1 site is well-conserved in ILK, whereas the conservation of other sites is more variable (supplementary material Fig. S1). T180 and S176, which also form a putative PAK1 phosphorylation site, are phosphorylated in Drosophila cells (Bodenmiller et al., 2007). T180 is also within a match to the protein kinase B (PKB) phosphorylation consensus RxRxxTL (Alessi et al., 1996). These results suggest that ILK S176 and T180 are phosphorylated by PAK, PKB or another kinase. Drosophila PAK1 is concentrated at MASs in an integrin-independent manner (Harden et al., 1996). We mutated both potential phosphorylation sites, generating phosphorylation-null (S176G, T180G) and phosphomimetic (S176D, T180D) ILK mutants. When overexpressed, both of these mutant proteins showed only mildly reduced MAS recruitment [in 1stIL 40% was recruited, increasing to 85–90% in 2ndIL (Fig. 4G–I,P–R)] and neither mutant protein showed nuclear translocation. Therefore, in contrast with the findings in mammalian cells, phosphorylation of these residues does not regulate the distribution of ILK in muscles.

In summary, our results show that cutting ILK into two halves results in proteins that are recruited to MASs, albeit less efficiently than the full-length protein, demonstrating that there are at least two protein interactions that contribute to ILK recruitment. By examining recruitment at two stages of larval life, we discovered that the mechanism that recruits the isolated ANKRs of ILK is lost in older larvae.

We next tested how the ILK mutations described above affect the function of the proteins, analysing multiple transgenic lines to control for variability in expression (supplementary material Fig. S2). The C-terminal GFP tag does not interfere with the function of ILK, as ILK–GFP expressed from its own promoter fully rescues the null ilk¹ embryonic lethal mutation (Zervas et al., 2001). Overexpression of untagged UAS::ILK, through the 24B Gal4 line (which drives expression primarily in muscles and tendon cells) rescues ilk lethal alleles to give viable adults, although such flies have wing blisters, presumably due to insufficient expression in the wing (Zervas et al., 2001). UAS::ILK–GFP gave equivalent rescue (Fig. 5A,F).

Deletion of ANKRs 2–5 did not completely eliminate ILK activity, but its function was severely reduced; it only partially rescued the actin-detachment phenotype of ilk mutant embryos (Fig. 5B); ~30% of the mutant embryos hatched into larvae, regardless of whether the construct was expressed at normal or high levels, compared with 0% in the absence of any rescue construct (Fig. 5F). The rescued 1stIL were unable to move normally and died during larvae development. This demonstrates that ANKRs are important for ILK function, and act to maintain the stable association of actin filaments with the plasma membrane, permitting development to the adult. However, these results also show that the ILK kinase domain maintains some ILK function on its own, sufficient to delay muscle defects until larval stages in the 30% of the mutant embryos that escaped lethality.

Deletion of the kinase domain completely abolished ILK function (Fig. 5D,F). This result was expected given that the molecular lesion in ilk¹ is a stop codon at residue W211, which causes a deletion of most of the kinase domain (Δ211–448), and this allele behaves as an amorph (Zervas et al., 2001). We next examined the effect of mutating known interaction sites within the ILK kinase domain. Mutation of the phosphoinositide-binding site (ILK^R208D,R210D–GFP) did not reduce the rescue of lethality when expressed from the endogenous ilk promoter (Fig. 5F), but ~35% (n=32) of the rescued adults (n=97) had blisters in one or both wings. Overexpression of this mutant form of ILK rescued embryonic lethality, but unexpectedly only 20% of the resulting 1stIL survived to adulthood (Fig. 5F). A few rescued 1stIL had a detached-actin phenotype, but the majority died at later stages (data not shown). The expression level of ILK^R208D,R210D–GFP appeared the same as wild-type ILK–GFP (Fig. 3A,B,I,J), and overexpression in wild-type embryos had no effect, so this mutant is not acting as a dominant-negative. Therefore, we suspect that 24B-Gal4 drives suboptimal expression in certain tissues, so that, in combination with the reduced activity caused by the point mutations, it fails to rescue the lethality of the ilk null mutation. Thus, mutation of the phosphoinositide-binding site appears to have the most significant effect in non-muscle tissues, such as the wing. Adding the R208D, R210D mutations did not further impair the partial function of the isolated kinase domain (Fig. 5C,F).

Mutation of the putative paxillin-binding site (T384A or T384D) had no effect on the ability of ILK to rescue activity (Fig. 5F). Similarly, we found no alterations in the functions of ILK in the mutants S176G, T180D and F436A. All the site-directed mutants we engineered and overexpressed with the 24B-Gal4 line rescued the ilk mutants in a manner equivalent to that with wild-type ILK–GFP (Fig. 5F).

We have found that both the ANKRs and the kinase domain are important for ILK function. However, we next tested whether it is necessary for them to be present in the same molecule. Coexpressing the two domains as individual GFP-tagged proteins did not improve the ability of each of them to rescue the embryonic lethality (Fig. 5F). Indeed, unexpectedly, it had the opposite effect, as the rescue activity of ILK lacking the ANKRs was reduced by adding the ANKRs in trans. Thus, the two domains of ILK need to be present in the same molecule, supporting the view that ILK functions as an adaptor, linking at least two other molecules together in Drosophila MASs.

It has been previously shown that removing PINCH does not affect the localization of ILK to Drosophila MASs (Clark et al., 2003). This contrasts with results from mammalian cells showing that ILK recruitment requires PINCH (Tu et al., 2001). One possible explanation for this difference is that existing mutant alleles in the gene encoding PINCH in Drosophila, steamer duck (stck), produce short proteins containing the ILK-binding LIM1 domain; these could recruit ILK because the alleles cause premature stops only after LIM1. To rule this out, we generated new alleles by excision of a P-element close to the main coding exons (see Materials and Methods). This resulted in a new allele, stck^T2, that deletes the region encoding LIM domains 1–3 and part of LIM4. The molecular nature of this mutation means it is unlikely that any PINCH protein is produced, but even if protein were produced, it would lack LIM1. We used this allele to generate embryos lacking both maternal and zygotic PINCH, and found that ILK–GFP was still recruited to MASs, at levels that were not significantly altered from those in embryos with wild-type PINCH (Fig. 6A–B′). This result enables us to conclude definitively that PINCH is not required for ILK stability and localization at MASs. We then tested whether the localization of ILK^ANKRs–GFP at MASs is mediated by its interaction with PINCH, and found that this mutated ILK form was still recruited in embryos lacking PINCH (Fig. 6C–D′). Similarly, the kinase domain of ILK localized at MASs in the absence of PINCH (Fig. 6E–F′). These results demonstrate that even when ILK is split into two halves, neither is dependent on PINCH for MAS recruitment. This implies that the essential function of PINCH in muscles is not to stabilize or recruit ILK, in contrast with recent studies in mammalian cells (Stanchi et al., 2009). Thus, even though ILK and PINCH have shared functions in diverse cells, the precise way they function together differs.

We next examined whether ILK has a role in PINCH localization. As discussed above for PINCH, it is possible that the characterized ilk¹ allele produces enough truncated ILK to bind and stabilize PINCH, because the ilk¹ has converted W211 into a stop, which is after the PINCH-binding ANKRs (Zervas et al., 2001). We tested this possibility by using a new allele that converts W7 into a stop (ilk⁵⁴), removing both the maternal and the zygotic contribution. This resulted in the loss of PINCH at MASs (Fig. 7A–C), and the levels of PINCH protein in the cytoplasm were reduced, even when the kinase domain of ILK was provided (Fig. 7C).

Because ILK is required to maintain PINCH protein levels, this could account for the loss of PINCH localization at MASs in the absence of ILK. We therefore examined whether overexpression of PINCH would restore its localization. In wild-type embryos, overexpressed PINCH–GFP was distributed in a manner similar to that of endogenous PINCH detected by antibody staining (Fig. 7D–E′). In ilk⁵⁴ zygotic mutant embryos, overexpressed PINCH–GFP was detected, demonstrating that it had evaded the dependence on ILK for stability, but was not strongly localized at MASs (Fig. 7F,F′). This demonstrates that ILK is required for both the stabilization and localization of PINCH.

We next examined whether the interaction between the ILK ANKRs and the LIM1 domain of PINCH is important for the ILK-dependent recruitment of PINCH. We showed above that the ANKR-containing N-terminus of ILK localized independently of PINCH, so we tested whether this ILK construct could restore PINCH localization in the absence of ILK, and we were surprised to discover that it could not (Fig. 7G–H″). However, PINCH with LIM1 deleted (PINCH^ΔLIM1) was poorly localized (Fig. 7I,I′), having a distribution similar to full-length PINCH in the absence of ILK (Fig. 7J,J′). Thus, LIM1 is crucial for PINCH MAS localization, but this does not occur through an interaction with the ILK ANKRs. Instead, these results suggest that ILK is indirectly required to recruit PINCH. We were intrigued by the small amount of PINCH^ΔLIM1 localized to MASs, suggesting a second recruitment mechanism. This small amount of localization was retained when the zygotic contribution of ILK or PINCH was removed (Fig. 7K,K′). The protein Rsu-1 binds tightly to the LIM5 domain (Kadrmas et al., 2004), but mutant animals lacking Rsu-1 still recruited PINCH^ΔLIM1 (data not shown). Furthermore, we noted that PINCH^ΔLIM1 was localized to the same extent as full-length ILK in third-instar larvae (Fig. 7L,M), suggesting that the level of the protein that recruits PINCH^ΔLIM1 increases during development.

Thus, our results show that, in muscles, the LIM1 domain is essential for efficient PINCH localization as MASs form during embryogenesis. We have also shown that PINCH localization is dependent on ILK, but unexpectedly ILK recruits PINCH indirectly, rather than through the known binding between PINCH LIM1 domain and the ILK ANKRs. This suggests that ILK recruits an unknown factor, which in turn recruits PINCH through the LIM1 domain. We also determined that there is a LIM1-independent pathway for PINCH recruitment, which arises as development progresses.

Finally, we used our new null alleles of ilk and stck to test whether the interaction between paxillin and ILK plays a role in paxillin localization. We found that embryos lacking both maternal and zygotic ILK or PINCH still recruited paxillin, at levels similar to that in wild-type embryos (Fig. 8A–E″). This is consistent with the robust paxillin localization at myotendinous junctions in zebrafish embryos lacking ILK (Postel et al., 2008).

In cell culture models, focal adhesions are the prototype form of sites of integrin adhesions. Several lines of evidence suggest that the molecular assembly of focal adhesions is a multistep process, where different cytoplasmic proteins enter the adhesion sites in a defined sequential order (Zamir and Geiger, 2001). Here, we provide genetic evidence that supports a hierarchical model for formation of integrin adhesions, using the Drosophila embryonic MASs as a model system. A summary diagram comparing the factors that contribute to ILK recruitment in vertebrate cell focal adhesions and Drosophila MASs is shown in Fig. 9.

First, we re-examined whether integrins are essential for the localization of ILK to MASs. In a previous study, we found that ILK was recruited independently of integrins (Zervas et al., 2001). Here, we corrected this conclusion, and demonstrated that complete elimination of integrins resulted in loss of ILK recruitment. This result, therefore, reveals that the small amount of maternally provided integrin recruits unexpectedly high levels of ILK. This surprising finding has two important implications: that the recruitment of ILK and maintenance of cell–matrix adhesion have different requirements for the amount of integrin (low levels of integrin are sufficient to recruit ILK, but high levels are needed for functional adhesion); and that the amount of ILK recruited by integrins is not set by the quantity of integrin at MASs, suggesting that there is an amplification mechanism so that a single integrin can recruit multiple ILK molecules (although our experiments do not reveal the relative number of molecules). A similar, but less extreme, lack of correlation between the levels of integrin and an intracellular integrin-associated protein at MASs was found previously for talin (Devenport et al., 2007). This apparent amplification could be explained by a multidocking scaffolding protein that is able to recruit, either directly or indirectly, several ILK molecules. Alternatively, ILK could have the ability to employ a variety of additional interactions with different binding partners in the integrin-containing junctions, so that even if integrins or other components are much reduced, multiple ILK molecules are recruited. The mechanism of the amplification is unknown, with a number of the integrin-associated proteins being potential components; the kindlin proteins are particular good candidates, because they bind both integrins and ILK (Karaköse et al., 2010; Rogalski et al., 2000), and Wech has been shown to bind both talin and ILK and to contribute to ILK recruitment (Löer et al., 2008).

Talin binding to integrins is one of the first molecular events, and is a prerequisite, in the formation of the integrin–cytoskeleton link (Tadokoro et al., 2003). Therefore, we examined whether talin is required to recruit components of the ILK-containing complex at sites of integrin adhesion. As with integrins, it was essential to remove both the maternal and zygotic production of talin to reveal defects in ILK, PINCH and paxillin recruitment. Consequently, these results show an important function for talin in recruiting the ILK-containing complex. A similar role for talin in ILK recruitment at myotendinous junctions in mice was reported recently (Conti et al., 2009). Currently there are no data to support a direct association between components of the ILK complex and talin, but Wech binds to talin and ILK suggesting that it could be the key linker between the two proteins (Löer et al., 2008), at least in muscles. Therefore we favour a model whereby talin, in concert with integrins, recruits Wech and additional proteins. These proteins then in turn recruit ILK and paxillin, and, as shown here, ILK indirectly recruits PINCH (Fig. 9).

In vertebrate cells ILK and PINCH have mutually dependent functions. Their association is required to protect each other from a proteasome-dependent degradation (Fukuda et al., 2003) and to facilitate each other's localization to focal adhesions (Zhang et al., 2002). The direct interaction between ILK and PINCH has also been confirmed in Drosophila (Clark et al., 2003; Kadrmas et al., 2004), and PINCH-mediated stabilization of ILK in mice has been confirmed genetically (Li et al., 2005). In Drosophila, we found that neither ILK protein stability nor its recruitment to MASs was affected in the absence of PINCH. This is in agreement with the persistent subcortical localization of ILK in the pinch-1-knockout mouse embryoid bodies (Li et al., 2005), as well as in unc-97 mutants in C. elegans (Norman et al., 2007). By contrast, loss of ILK function reduced PINCH levels in Drosophila embryos. The instability of PINCH in the absence of ILK was overcome by overexpressing PINCH but, under these conditions, PINCH still was not localized to MASs. The isolated ILK ANKRs did localize to MASs in the absence of endogenous ILK, but were not able to recruit PINCH. Thus, rather than ILK recruiting PINCH by binding of its ANKRs to the LIM1 domain of PINCH, it appears to do so indirectly. This finding raises two important points. First, the interaction between the ILK ANKRs and LIM1 of PINCH might not be always maintained at sites of integrin adhesion. Our finding that LIM1 is crucial for recruitment, but apparently not through interaction with ILK, suggests alternative interactions might be required. This is consistent with our observation that high levels of PINCH–GFP can be recruited to MASs, even when only the endogenous levels of ILK are present. Second, recent work in mammalian cells provided strong evidence that the primary function of PINCH was the recruitment of ILK through the ANKRs–LIM1 interaction (Stanchi et al., 2009). Notably, the recruitment of tensin and ILK at focal adhesions in cells lacking PINCH could be rescued by a direct fusion of the ILK kinase domain to integrin. However, our result suggests that, in Drosophila muscles, the reverse is true, and a primary function of ILK might be the stabilization and recruitment of PINCH.

Finally, we examined whether ILK and PINCH influence the subcellular localization and protein stability of paxillin. Previous studies have demonstrated an interaction between the kinase domain of mammalian ILK and paxillin, and further suggested that this interaction controls ILK localization at focal adhesions (Nikolopoulos and Turner, 2001). These studies identified key residues in ILK that contribute to paxillin interaction and ILK recruitment, such that ILK mutations at E359 and T384 impaired recruitment of ILK to focal adhesions. However, the ILK E359K mutation does not impair its ability to rescue the null phenotype or alter its recruitment in flies or mice (Sakai et al., 2003; Zervas et al., 2001). Here, we showed that mutating T384 also failed to impair the biological activity or recruitment of ILK, and neither ILK nor PINCH were required for paxillin recruitment. Thus, in MASs, the stability of paxillin and its subcellular localization does not depend on the ILK–PINCH complex.

In this study, we examined the functions of the two main domains that constitute ILK (i.e. the ANKRs and the kinase domain), as well as specific binding sites that have been proposed to mediate interactions with phosphoinositides or paxillin and have been implicated in the regulation of the protein. From this work, it became evident that deletion of either the ANKRs or the kinase-like domain eliminates the essential functions of ILK in the developing organism. These two domains are required in the same molecule to execute the essential function of ILK. The proposed phosphoinositide-binding motif is dispensable for embryonic development but is required in the epithelial tissues of the wing, and therefore does contribute to ILK function. This suggests that the highest levels of ILK function are required in the wing. This is consistent with the finding that elimination of tensin only causes a wing-blister phenotype, even though it is also concentrated at MASs (Torgler et al., 2004). By contrast, hypomorphic mutations in the gene encoding the αPS2 integrin subunit were found to cause muscle defects but not cause wing blisters (Bloor and Brown, 1998), indicating that integrin adhesive sites are differentially sensitive to loss of different components of the integrin adhesion complex. The residue F436, which is located at the far end of the C-terminal region, was found to be crucial in vertebrate cells, but its mutation only mildly affected recruitment of the overexpressed protein and was dispensable for ILK function. It is possible that the F436A mutation reduces the binding affinity of ILK with Wech.

We found that both the isolated ANKRs and the kinase domain were recruited to MASs, in contrast with the recruitment to focal adhesions seen in mammalian cells (Zhang et al., 2002). This suggests that recruitment is more robust at MASs, so impaired ILK recruitment can still be visualized. We were surprised to find that the recruitment of the ANKRs of ILK did not involve an interaction with PINCH, as removing PINCH did not impair recruitment. Given that there is only one Drosophila pinch gene, in contrast with two in mammals, the ANKRs must be binding to another type of protein. One alternative candidate is Mlp84B, a protein that contains five LIM domains and is localized at MASs in an integrin-independent manner; it is known that the related zebrafish protein CSRP3 binds ILK (Clark et al., 2007; Postel et al., 2008).

In summary, our results have demonstrated that, at the robust integrin adhesive sites that form in the developing muscles, ILK recruitment is much less sensitive to perturbation compared with recruitment in focal adhesions in cells in culture. Our results show that integrins and talin are essential for the recruitment of ILK, and ILK in turn is essential for the recruitment of PINCH. How much of the function of ILK is mediated by PINCH, and whether it also functions independently through its interaction with parvin, will be the focus of our future research.

The ilk¹, Df(3L)Pc-14d, and the genomic ILK–GFP rescue construct were as described previously (Zervas et al., 2001). The allele ilk⁵⁴ was isolated in an ethyl methanesulphonate (EMS) screen carried out in the laboratory of Danny Brower (University of Arizona, Tucson, AZ) and was kindly provided to us. This allele contains a premature stop codon after the sixth amino acid of ILK. The stck^T2 allele was generated by the imprecise excision of the P-element EY06672, inserted within the stck locus. The molecular lesion in the stck^T2 allele is a deletion of 1615 bp, removing 697 bp upstream of the ATG and 918 bp downstream, resulting in the complete absence of the region encoding the amino acid residues 1–228 of PINCH protein. The alleles ilk⁵⁴ and stck^T2 were recombined with FRT sites in order to make germline clones (Chou and Perrimon, 1992). Additional alleles used were stck^3R-17, stck^3R-18, if^B4, if^sef, mys^XG43, rhea^79A, and the Gal4 lines mef2 and 24B (see FlyBase for references).

The genomic rescue fragment for ilk was subcloned in the P-element transformation vector pWhiteRabbit (Zervas et al., 2001) and used as a template for PCR mutagenesis, verified by sequencing. All transgenes were fused at the C-terminus to mGFP6 (Kaltschmidt et al., 2000) or mRFP1 [provided by Roger Tsien (Howard Hughes Medical Institute and Department of Pharmacology, University of California, San Diego, CA)] and were subcloned downstream of the Gal4-dependent UAS promoter in the pGreenRabbit vector derived from pUAST (Brand and Perrimon, 1993) or in the genomic ilk rescue construct. The ILK deletion constructs generated had the following modifications: Δ34–177, which removes ANKRs 2–4 (amino acids 33–128) and the middle part of the protein (amino acids 129–177); Δ178–448, a deletion of the C-terminus part of the ILK protein (amino acids 177–448), which contains the putative phosphoinositide-binding motif (amino acids 180–212) and the entire kinase domain (amino acids 192–448); R208D, R210D; the Δ178–448 deletion with R208D, R210D; T384A; T384D; S176G, T180G; S176D, T180D; and F436A. All the engineered transgenic lines selected for rescue experiments were introduced in the same genetic background carrying the Df(3L)Pc-14d, e¹ third chromosome and crossed with the ilk¹ or ilk⁵⁴ alleles at 25°C.

Whole-mount labelling of embryos up to stage 16 was performed as described previously (Zervas et al., 2001), whereas stage 17 embryos were heat-fixed (Miller et al., 1989), and was followed by standard antibody labelling. Primary antibodies used were against: αPS2 (PS2HC2) (rat monoclonal; 1:10) (Bogaert et al., 1987); MHC (myosin heavy chain) (mouse monoclonal; 1:60) (Kiehart et al., 1990); PINCH (rabbit polyclonal; 1:500) (Clark et al., 2003); and paxillin (rabbit polyclonal 1:500) (Chen et al., 2005). The secondary antibodies were conjugated to Alexa-Fluor-488, -568 and -633 (1:500, Molecular Probes). F-actin was probed using either Rhodamine or Alexa-Fluor-635-labelled Phalloidin (R-415 and A34054; Molecular Probes). Images were obtained with a Radiance2000 microscope (Bio-Rad Laboratories), using an oil 60×/1.4 NA objective, or a Leica SP5 confocal microscope, using the 20×/0.7 NA objective or an oil 63×/1.4 NA objective. Two types of confocal optical sections of muscle attachment sites are shown. In some cases, single horizontal optical sections were obtained, which allows discrimination between signals coming from muscles and tendon cells. This view was used to demonstrate the loss of the examined molecule from the muscle ends (Fig. 1, Fig. 2 and Fig. 7A,B). When the examined molecule was localized to some degree, a surface view was used to assess recruitment over a larger number of MASs. This surface view is a stack of images over a depth of 4–5 μm, starting 2–3 μm below the surface of the epidermis, and focused on either the lateral longitudinal (Fig. 5, Fig. 6B–D, Fig. 7D–J and Fig. 8C) or dorsal oblique muscles (Fig. 3, Fig. 4 and Fig. 7G,H). Leica SP5 software was used for quantification. The confocal laser power and gain was set appropriately to avoid pixel intensity saturation. Selected areas at MASs or nuclei, or the entire muscle area, were outlined, and the total intensity was measured and plotted using Excel. Images were assembled in Photoshop 7 and labelled in Corel Draw 12.

Protein lysates were prepared from selected stage 17 embryos with the appropriate genotype and were analysed by western blotting using antibodies against PINCH, paxillin and tubulin (Sigma) as a loading control. Blots were then probed with horseradish peroxidase (HRP)-conjugated anti-(rabbit Ig) or anti-(mouse Ig) secondary antibodies (Jackson Laboratories; 1:10,000) and were developed using enhanced chemiluminescence (Amersham).

We thank John Overton and Georgios Chrysanthis for excellent technical assistance and Maithreyi Narasimha for critical reading of the manuscript. We are grateful to Danny Brower for providing us with the EMS ilk alleles. We also thank Mary Beckerle, Mike Taylor, Sheila Thomas, Roger Tsien and the Bloomington Drosophila Stock Center, Developmental Studies Hybridoma Bank at the University of Iowa (DSHB) for sharing several antibodies, fly stocks and DNA plasmids. This research was supported by Greek Secretariat for Research and Technology (PENED 03ED777), Association Française contre les Myopathis (12641), intramural funds of BRFAA to C.G.Z. and the Wellcome Trust (64098) to N.H.B. Deposited in PMC for immediate release.