FAK engages multiple pathways to maintain survival of fibroblasts and epithelia – differential roles for paxillin and p130Cas

Different cell types interpret their distinct extracellular matrix (ECM) environments to bring about specific cell fate decisions, and can differentiate or undergo apoptosis depending on their local adhesive interactions. Apoptosis in response to an inappropriate ECM environment is termed `anoikis', or homelessness. Several studies, utilising a variety of cell types, have indicated a common, crucial role for focal adhesion kinase (FAK) in suppressing anoikis. A wide range of different integrins can activate FAK, raising the question of how cell type specific effects are regulated. In this study, we have used a constitutively active form of FAK to examine the mechanism of FAK-mediated survival signalling in cell types from distinct embryonic lineages that show differing sensitivities to anoikis. We demonstrate that both fibroblasts and epithelial cells prevent anoikis through FAK activation. We show that FAK activates multiple downstream pathways in order to suppress anoikis. However FAK regulates survival through a more restricted set of pathways in the more anoikis-sensitive epithelial cells. Furthermore, we identify a novel role for paxillin in apoptosis suppression.


Introduction
Integrin-mediated adhesion to the extracellular matrix (ECM) is essential for survival of most cells (Meredith and Schwartz, 1997;Reddig and Juliano, 2005). In the absence of an appropriate ECM, normal cells undergo a form of apoptosis termed anoikis (Frisch and Screaton, 2001;Gilmore, 2005). Loss of sensitivity to anoikis correlates with anchorage-independent tumour growth and the ability of a cell to metastasise and survive in diverse ECM environments (Douma et al., 2004).
Integrins are adhesion receptors that monitor the cellular microenvironment through interactions with ECM proteins (Hynes, 2002). They have a dynamic intracellular signalling role, recruiting numerous enzymes and scaffold proteins to focal adhesions. The nonreceptor tyrosine kinase focal adhesion kinase (FAK) is a key mediator of integrin signalling (Mitra and Schlaepfer, 2006;Parsons, 2003). FAK interacts with signalling and adaptor proteins, including p130Cas, Grb2, paxillin, Shc and PI-3 Kinase, many of which are involved with cellular survival. FAK is recruited to sites of integrin/ECM attachment via a focal adhesion-targeting (FAT) domain at its C-terminus. FAK recruitment induces autophosphorylation at tyrosine 397, providing a binding site for Src. Src activates upon recruitment to FAK and phosphorylates key tyrosines within both FAK and adaptor molecules such as paxillin and p130Cas, generating multiple binding sites for SH2-domain proteins, and thereby initiating intracellular signalling cascades (Playford and Schaller, 2004).
Shortly after the first descriptions of anoikis, an activated form of FAK was shown to suppress cell death in detached MDCK cells (Frisch et al., 1996b). This effect was dependent upon both FAK kinase activity and its phosphorylation on tyrosine 397. Other studies have shown that disrupting FAK signalling using anti-FAK antibodies, antisense oligonucleotides, or the expression of dominant-negative FAK, induces apoptosis in adherent cells (Gilmore et al., 2000;Hungerford et al., 1996;Ilic et al., 1998;Xu et al., 1996;Xu et al., 1998). Moreover, genetic deletion of FAK sensitises keratinocytes and endothelial cells to apoptosis in vivo (Braren et al., 2006;Essayem et al., 2006). Disruption of FAK signalling not only has effects in normal cells but is also implicated in tumour progression. FAK is overexpressed in approximately 90% of breast tumour cell lines and is amplified in around 80% of primary breast tumours (Agochiya et al., 1999;Lark et al., 2003;Lightfoot et al., 2004;Watermann et al., 2005). Aberrant FAK signalling may facilitate the neoplastic process by providing cells with an anchorage independence, allowing them to survive in an otherwise hostile ECM environment. Thus, understanding how FAK interprets ECM into a survival signal in normal cells is vital in identifying key points at which aberrant signalling in cancer can be targeted. However, current knowledge of which FAK-interacting proteins integrate adhesion with survival pathways is often contradictory (Almeida et al., 2000;Frisch et al., 1996a;Gilmore et al., 2000;Ilic et al., 1998;Khwaja and Downward, 1997).
Here we examine the role of FAK in cell survival utilising a myristoylated form of FAK, (myrFAK). MyrFAK is active independent of adhesion, localises to adhesion complexes and suppresses apoptosis following loss of integrin function. We demonstrate that cells from different embryonic lineages have distinct strategies for interpreting integrin-mediated survival stimuli. Both epithelia and fibroblasts rely on FAK for their survival but Different cell types interpret their distinct extracellular matrix (ECM) environments to bring about specific cell fate decisions, and can differentiate or undergo apoptosis depending on their local adhesive interactions. Apoptosis in response to an inappropriate ECM environment is termed 'anoikis', or homelessness. Several studies, utilising a variety of cell types, have indicated a common, crucial role for focal adhesion kinase (FAK) in suppressing anoikis. A wide range of different integrins can activate FAK, raising the question of how cell type specific effects are regulated. In this study, we have used a constitutively active form of FAK to examine the mechanism of FAK-mediated utilise different yet overlapping sets of FAK adaptors and signalling pathways.

Fibroblasts and epithelial cells show distinct survival requirements for ECM but both transmit this requirement via FAK
Mouse embryonic fibroblasts (MEFs) did not undergo anoikis when they were maintained in suspension in the presence of growth factors, whereas detachment from ECM coupled with withdrawal of growth factors did cause anoikis (Fig. 1a, lower panel) (Hungerford et al., 1996;Meredith et al., 1993). By contrast, mammary epithelial cells (MECs) were sensitive to anoikis regardless of whether growth factors were present (Fig. 1a,upper panel). In response to detachment from ECM, both cell types showed rapid dephosphorylation of FAK and paxillin (Fig. 1b). Adherent mesenchymal and epithelial lineages both showed discrete patterns of tyrosine phosphorylation, with major components at approximately 130 kDa (FAK) and 70 kDa (paxillin). Phosphorylation of both FAK-Y397 and paxillin-Y31 was lost within 15 minutes of detachment from the ECM. Expression of a dominant-negative form of FAK (DN-FAK), consisting of the FAK FAT domain, induced apoptosis in both adherent MECs and MEFs, although the latter also required removal of serum growth factors (Fig. 1c). Thus, despite the distinction between growth factor requirements in MECs and MEFs, both cell types required FAK signalling to provide the ECMdependent anti-apoptotic signal.
FAK suppresses anoikis via its interaction with p130Cas in rabbit synovial fibroblasts (Almeida et al., 2000;Ilic et al., 1998). We therefore asked whether p130Cas has a role in MEC or MEF survival. We utilised a number of p130Cas deletion constructs (Fig.  2a), all of which expressed at the predicted molecular weights (Fig.  2b). Expression of the CasSH3 domain, previously used as a dominant negative, inhibited the phosphorylation of endogenous p130Cas on tyrosine 410 (Fig. 2c). We next asked whether inhibiting endogenous p130Cas induced apoptosis in fibroblasts or MECs, compared with expression of DN-FAK (Fig. 2d). Both MEFs and MECs were transfected with expression vectors for DN-FAK, FL-p130Cas and p130Cas-SH3. Apoptosis was quantified 24 hours post transfection (MEFs in the absence of serum growth factors). DN-FAK induced apoptosis in both adherent MECs and MEFs, whereas FL-p130Cas did not. By contrast, the dominant-negative p130Cas-SH3 increased apoptosis in MEFs by a small but significant (P<0.05) amount. This was notably less than previously seen in rat synovial fibroblasts (Almeida et al., 2000). There was no increase in apoptosis in MECs expressing p130Cas-SH3. Together, these data support previously published results indicating that p130Cas has a role in the suppression of anoikis in fibroblastic cells but indicated no similar role in epithelial cells. Furthermore, the much greater amount of apoptosis induced by DN-FAK indicated that in both cell types additional signalling pathways were likely to play a pivotal role in suppression of anoikis.

Myristoylated FAK suppresses anoikis in MECs and MEFs
We wanted to identify what other pathways downstream of FAK may be involved in survival. DN approaches can potentially inhibit or sequester key molecules in multiple pathways. We therefore decided to use a constitutively active (CA) FAK to suppress anoikis in detached cells. CA-FAK would remain active following cell detachment, when endogenous adhesion-dependent signalling was turned off.
Myristoylated FAK (myrFAK) was generated by cloning the v-Src myristoylation sequence N-terminal to the FAK coding sequence (Klippel et al., 1996). Autophosphorylation-site (myrFAKY397F) and kinase-inactive (myrFAKK454R) mutants were also expressed ( Fig. 3a,b). All three myrFAK constructs localised to focal adhesions in adherent cells (Fig. 3c). MyrFAK was phosphorylated on tyrosine 397 but this was significantly reduced in myrFAKK454R and undetectable with myrFAKY397F ( Fig. 3b). Moreover, myrFAK, but not endogenous FAK or myrFAKY397F, remained phosphorylated following loss of cell/ECM attachment (Fig. 3d). Together, these data indicate that although myrFAK is recruited to focal adhesions in adherent cells, its activation and phosphorylation are independent of cell/ECM attachment.
We asked whether myrFAK suppresses anoikis. MEFs and MECs transiently expressing myrFAK, myrFAKY397F or myrFAKK454R were detached from ECM for 24 hours and maintained in the absence of serum (in the case of MEFs), or in complete growth medium (in the case of MECs) (Fig. 3e). Transfected cells were identified by immunostaining with anti-V5, and apoptosis determined by nuclear morphology. In both cell types, the apoptosis occurring as a result of ECM detachment (mock-transfected controls) was rescued significantly in cells expressing myrFAK (P<0.05). The ability of myrFAK to protect against anoikis required both its kinase activity (K454R) and its tyrosine phosphorylation (Y397F).
To determine whether recruiting FAK to the membrane was required for its anti-apoptotic function, we expressed either myrFAK or FAK minus the myristoylation sequence (WT-FAK). Subcellular fractionation demonstrated that although both myrFAK and WT-FAK were detected in the cytosolic fraction, only myrFAK was present in the membrane fraction (Fig. 4a). Both myrFAK and WT-FAK were phosphorylated on tyrosines 397 and 577 in cells detached from ECM, indicating that overexpression alone was sufficient to activate FAK (Fig. 4b). However, when we examined apoptosis in detached cells, although WT-FAK appeared to reduce anoikis in both MECs and MEFs, this was not significantly different (P>0.05) than cells expressing myrFAKY397F (Fig. 4c).
These data show that both activation and recruitment to the membrane is required for FAK to suppress anoikis in both fibroblasts and epithelial cells.
The FAK-p130Cas signalling axis only provides an antiapoptotic signal in fibroblasts To further explore the role for p130Cas in survival, we asked whether a form of myrFAK that was unable to bind p130Cas could suppress anoikis. FAK contains two proline rich (PR) domains, which can interact with the SH3-domain-containing proteins (Harte et al., 1996;Hildebrand et al., 1996;Polte and Hanks, 1995). We generated myrFAK constructs lacking either or both PR domains (Fig. 5a). Mutation of the PR domains did not prevent phosphorylation on Y397 or focal adhesion localisation in adherent cells (Fig. 5b,c). We were unable to detect endogenous p130cas binding, presumably due to the levels of its expression. Thus, to confirm the PR-domain mutations were nonfunctional for SH3-domain binding, we co-expressed either p130Cas⌬SD (which contains the SH3 domain) or p130Cas⌬SH3, along with myrFAK and the mryFAK PR-domain mutants. myrFAK was immunprecipitated with anti-V5 and immunoblotted for the coexpressed p130Cas. Deletion of PR-1, reported as the primary binding site for p130Cas (Harte et al., 1996;Polte and Hanks, 1995), resulted in a noticeable decrease in the amount of p130Cas⌬SD myrFAK or the myrFAK PR-domain deletion mutants were expressed in MEFs and MECs, cells were detached for 24 hours in the absence (MEFs) or presence (MECs) of serum growth factors, and apoptosis was quantified (Fig. 5e). In agreement with previous studies on fibroblasts (Almeida et al., 2000), mutation of PR-1 abolished the ability of myrFAK to protect MEF anoikis. myrFAK with a deletion of the PR-2 domain appeared to suppress anoikis to some extent, but this did not reach statistical significance (P>0.05). This may reflect the different ability of each site to interact with p130Cas (Fig. 5d). In marked contrast to the results in fibroblasts, all of the FAK PR-domain mutants suppressed anoikis in epithelial cells to the same extent as the mutated myrFAK (P<0.001).
Thus, whereas intact PR-domains are required for FAKdependent survival in fibroblasts, they do not contribute to epithelial cell survival.
FAK interactions with paxillin and SH2-domain proteins are required for anoikis suppression p130Cas was not required for MEC survival. Furthermore, using the DN-p130CasSH3 indicated that it was unlikely to be the sole mediator of survival in MEFs. We therefore examined other potential interactions with FAK that may transduce survival signals. The C-terminus of FAK contains two other major interaction sites for signalling adaptors. Tyrosine 925, when phosphorylated, can recruit SH2-domain proteins such as Grb2 (Schlaepfer and Hunter, 1996). The FAT domain contains a binding site for paxillin, consisting of two discontinuous sub domains, in which I936 and I998 are essential (Hayashi et al., 2002). Paxillin shares some features of p130Cas, in that it is an adaptor protein for downstream signals, which interacts with the C-terminus of FAK (Hildebrand et al., 1995;Turner et al., 1990). The FAK/Src complex phosphorylates paxillin at Y31 and Y118, creating binding sites for SH2-domain-containing proteins (Bellis et al., 1995;Deakin and Turner, 2008;Playford and Schaller, 2004).
We generated myrFAK mutants in which either the paxillin binding site or Y925 were disrupted (myrFAKI936E/I998E and Journal of Cell Science 122 (3) (c) MEFs transiently expressing myrFAK, myrFAKY397F or myrFAKK454R were immunostained with antibodies against the V5 epitope or a focal adhesion marker (anti-paxillin). (d) MECs transiently expressing myrFAK or myrFAKY397F were left adherent or detached and maintained on poly-HEMA for 30 minutes. Whole-cell lysates were immunoblotted for anti-V5, anti-phosphoFAK Y397, anti-total FAK and actin. Note that phospho-FAK Y397 is reduced in detached cells transfected with empty vector or expressing myrFAKY397F but is not reduced in cells expressing myrFAK. (e) MECs and MEFs expressing myrFAK, myrFAKY397F or myrFAKK454R, or transfected with empty vector, were detached from ECM for 24 hours. Apoptosis was quantified by nuclear morphology following Hoechst staining. MECs were maintained in the presence of growth factors, whereas MEFs were without growth factors. Data represent the mean of three experiments. Error bars indicate standard error. *, Significant reduction in apoptosis relative to the mock-transfected cells (ANOVA with Bonferroni's multiple comparison test); n/s, not significant. myrFAKY925F) (Fig. 6a). Both mutants expressed and were phosphorylated on Y397 (Fig. 6b, lanes 4 and 5). myrFAK and myrFAKI936E/I998E were phosphorylated on Y925 (Fig. 6b, lanes 7 and 10), but myrFAKY397F and myrFAKY925F were not (Fig.  6b, lanes 8 and 9). To confirm disruption of the paxillin-binding site, lysates of cells expressing wild-type myrFAK, myrFAKI936E/I998E or myrFAKY925F were incubated with glutathione agarose beads coated with GST-paxillin. Immunoblotting with anti-V5 indicated that both myrFAK and myrFAKY925F bound to paxillin (Fig. 6c, lanes 2 and 4), whereas myrFAKI936E/I998E did not (Fig. 6c, lane 3). We examined binding to endogenous paxillin by expressing the myrFAK mutants in MEFs and crosslinking complexes with DSS before immunoprecipitating with anti-V5. Both myrFAKY925F and myrFAKY397F coprecipitated with endogenous paxillin, whereas myrFAKI936E/ I998E did not. In adherent cells, myrFAKI936E/I998E did not localise to focal adhesions in MEFs, whereas myrFAKY925F did (Fig. 6d). Thus, a functional paxillin binding site appears to be required for focal adhesion localisation.
To determine whether either the paxillin or SH2-domain interacting sites were required for myrFAK to suppress anoikis, myrFAKI936E/I998E and myrFAKY925F were expressed in MECs and MEFs, and apoptosis was quantified following detachment from ECM. The anoikis suppression afforded by myrFAK in both cell types was abolished when either the paxillin or SH2-domain binding sites were deleted (Fig. 6e). Thus, these data show that functional paxillin and SH2-domain binding sites within the Cterminus of FAK are required for adhesion-dependent survival signalling in both MECs and MEFs.

Paxillin signalling is required for FAK to suppress epithelial anoikis
To confirm that paxillin was required to suppress anoikis, we expressed paxillin in which tyrosine 31 and 118 were substituted with phenylalanine (GFP-PaxY31/118F) (Fig. 7a) in MECs. Previous studies have shown that PaxY31F/Y118F functions as a dominant negative and can block fibroblast migration (Petit et al., 2000). GFP alone, GFP-paxillin, and GFP-PaxY31/118F all expressed to similar levels in MEC (Fig. 7b). Mutation of the both phosphorylation sites did not affect the ability of paxillin to target to focal adhesions in adherent cells, although it did displace endogenous paxillin from these sites, seen by loss of phosphopaxillin immunostaining (Fig. 7c). When we examined survival, adherent MECs expressing wild-type GFP-paxillin showed no increase in apoptosis compared with cells expressing GFP alone (Fig. 7d). By contrast, expression of GFP-PaxY31/118F resulted in a marked and significant increase in apoptosis (P<0.05).
Thus, even though wild-type and mutant paxillin are both recruited to focal adhesions, its phosphorylation on Y31/Y118 is critical for MEC survival.
Multiple signalling complexes recruited to the C-terminus of FAK are required to suppress anoikis As myrFAKY925F could still bind paxillin and myrFAKI936E/ I998E was still phosphorylated on Y925, we asked whether FAK suppressed anoikis via multiple signalling complexes. MyrFAK, myrFAKY397F, myrFAKI936E/I998E and myrFAKY925F were expressed in HEK293T cells. These were detached from ECM and crosslinked with DSS prior to immunoprecipitating the expressed proteins with an antibody against the V5 epitope tag. The immunoprecipitates were then separated by SDS-PAGE and immunoblotted for FAK and paxillin (Fig. 8a). Immunoblotting with an anti-FAK antibody identified that myrFAK and myrFAKY397F formed two distinct crosslinked complexes, at approximately 200 kDa and 300 kDa (Fig. 8a, lanes 1 and 3, marked with an asterisk). Neither complex was observed without crosslinking (Fig. 8a, lanes 2 and 4) or in cells not expressing myrFAK (lanes 9 and 10). Paxillin was identified in the 200-kDa complexes with myrFAK, myrFAKY397F and myrFAKY925F, but not with myrFAKI998/936E. Conversely, the 300 kDa complex was not observed with myrFAKY925F (Fig. 8a, lane  7). As these data suggested that FAK formed a number of independent signalling complexes, we asked whether co-expression of the paxillin binding and Y925 mutants could complement each other and restore survival signalling. Whereas neither myrFAKI936E/I998E nor myrFAKY925F alone protected MECs from anoikis, co-expression of both suppressed it to a similar level as seen with wild-type myrFAK (Fig. 8b). This suggests that the ability of FAK to suppress anoikis depends upon its ability to bind both paxillin and a second molecule that interacts with Y925, but that these do not form in the same signalling complex on the same FAK molecule.
Western blotting of 293T whole-cell lysates expressing myrFAK constructs identified that both myrFAK and myrFAKY925F maintained paxillin phosphorylation on Tyr31 and Tyr118 in detached cells (Fig. 8c, lanes 1 and 4). Paxillin was not phosphorylated in cells expressing myrFAKY397F, despite the observation that the two interacted (Fig. 8c, lane 2). Thus, wildtype FAK forms two distinct complexes, only one of which (at approximately 200 kDa) contains paxillin. The paxillin interaction and signalling is maintained by myrFAKY925F, but not myrFAKI936E/I998E. The components of the higher complex are at present unknown, and are a current focus of research.
Finally, we were interested to see whether there were any obvious cell-type differences in the signalling components immediately downstream of FAK. A number of the signalling pathways examined showed no obvious difference between MECs and MEFs. However, one notable survival pathway, Akt (protein kinase B) was seen to be adhesion dependent in MECs, whereas it was predominantly growth factor dependent in MEFs (Fig. 8d). To determine whether this could represent a FAK-dependent, cell-type-specific difference, myrFAK- expressing MEFs and MECs were detached from ECM for 1 hour and immunoblotted for phosphorylated Akt (Fig. 8e). Akt phosphorylation was maintained in MEC that expressed myrFAK following detachment from ECM, but was lost in untransfected cells. By contrast, growth factor-deprived MEFs showed no changes in Akt phosphorylation under the same conditions.
Together, these data indicate that FAK forms multiple signalling complexes. Furthermore, depending on the cellular context, these complexes transmit distinct survival signals.

Discussion
In this paper we have examined the proximal survival signalling mechanisms downstream of FAK in murine cells from distinct lineages. We show that although FAK is a common transducer of ECM-derived survival signals in fibroblasts and epithelia, its activation is interpreted differently in each cell type. Thus, in agreement with previous studies, we find that FAK suppresses anoikis in fibroblasts via its SH3-interacting PR domains but that these are not required in epithelial cells. However, we also find that both paxillin and SH2-domain interactions via tyrosine 925 are also required for adhesion-dependent survival. As we have utilised a model whereby FAK is constitutively activated in cells detached from ECM, the downstream signalling pathways instigated are themselves independent of integrin engagement. This implies that how activated FAK interacts with downstream signalling components is in part an intrinsic feature of a particular cell type. Our study demonstrates the importance of cellular context to determine how the same survival signal is interpreted intracellularly.
FAK suppresses apoptosis in a variety of anchorage-dependent cells (Frisch et al., 1996b;Gilmore et al., 2000;Hungerford et al., 1996; Xu et al., 2000). However, a confusing and often contradictory array of pathways are implicated downstream of FAK. For example, three independent studies have indicated that c-Jun N-terminal kinase (Jnk) can be either pro-apoptotic, anti-apoptotic or have no role downstream of FAK. In its pro-apoptotic role, Jnk was shown to be activated in MDCK cells following loss of adhesion, and dominant-negative Jnk-kinase inhibited anoikis (Frisch et al., 1996a). However, another study indicated that Jnk activation was dispensable for anoikis and instead activation of PI3-kinase was required (Khwaja and Downward, 1997). Yet another study demonstrated that Jnk activation downstream of FAK was actually required for cell survival, rather than apoptosis, in fibroblasts (Almeida et al., 2000). Our data, which show that myrFAK can maintain Akt phosphorylation in detached MECs, agree with a role for PI3-kinase in epithelial cells.
These contradictions may be explained by the cell-type-dependent signalling differences we highlight here. Thus, in its pro-survival guise, Jnk is activated in synovial fibroblasts downstream of FAK via p130Cas. Upon recruitment to FAK, p130Cas is phosphorylated on several tyrosine residues that themselves recruit SH2-domain containing adaptor proteins, including Crk and Nck (O'Neill et al., 2000;Playford and Schaller, 2004). This can activate a Ras-Rac pathway, which controls Jnk pro-survival signalling via p21activated kinase 1 (PAK1) and MKK4 (Almeida et al., 2000). This pathway is consistent with our data showing that myrFAK⌬PR1/2 cannot suppress MEF anoikis.
Studies on cell migration support our data that cellular context determines via which downstream adaptors FAK will signal. In MDCK cells, paxillin controls cell migration, whereas in other cell types p130Cas is required (Lamorte et al., 2003;Petit et al., 2000;Schlaepfer et al., 1994;Yano et al., 2000). Our data show that p130Cas is not required for FAK-dependent survival in epithelial cells. Instead, interactions with FAK via both paxillin and tyrosine 925 are necessary. Paxillin is an adapter molecule that can link FAK to multiple pathways (Deakin and Turner, 2008). The N-terminus of paxillin contains two phosphorylation sites, tyrosines 31 and 118, which bind Crk and can control Ras and ERK signalling (Dolfi et al., 1998;Igishi et al., 1999). Mutation of tyrosines 31 and 118 to phenylalanine block Crk-mediated fibroblast migration (Petit et al., 2000). Paxillin also interacts with paxillin kinase linker (PKL), integrin-linked kinase (ILK), α-parvin and PTP-PEST (Cote et al., 1999;Nikolopoulos and Turner, 2000;Nikolopoulos and Turner, 2001;Turner et al., 1999). The paxillin-PKL interaction is critical for PAK-mediated cell motility. PAK can activate Jnk and ERK pathways and could potentially be involved in FAK/paxillin survival signalling (Almeida, 2000;Hood et al., 2003;Howe, 2001;Igishi et al., 1999).
Although well established in cell migration, the role of paxillin in survival signalling is not well understood. One previous study has directly implicated paxillin as a mediator of survival signals (Subauste et al., 2004). In that study, competition between vinculin and FAK for paxillin binding determined the amount of paxillin available for downstream signalling via ERK. Thus, vinculin-null F9 cells were resistant to apoptosis because more paxillin was available for interacting with and being phosphorylated by FAK. Expressing either the paxillin-binding region of vinculin, or paxillinY31/118F, could reduce resistance to apoptosis. Here, we have shown that in normal cells, paxillin provides a necessary link in the FAK-dependent survival pathway, and that this may be common in multiple cell types. Thus, we clearly demonstrate a role for paxillin in survival signalling.
Paxillin alone is not sufficient for FAK to transmit survival signals, as mutation of tyrosine 925 also blocked survival signalling. Phosphorylated tyrosine 925 is a binding site for the SH2 domain of Grb2 (Schlaepfer et al., 1994;Schlaepfer and Hunter, 1996), although it is possible that other SH2-domain-containing proteins can bind to this site. A role for tyrosine 925 in FAK-dependent suppression of anoikis has not been demonstrated previously. It is of particular interest that signalling from paxillin and tyrosine 925 each act independently. Thus, co-expression of myrFAKY925F, which binds paxillin, and myrFAKI936E/I998E, which is still phosphorylated on tyrosine 925, rescued cells from anoikis to the same extent as wild-type myrFAK alone. Together, our data show Journal of Cell Science 122 (3) that FAK forms multiple signalling complexes, each of which functions independently but all of which can contribute to survival signalling. Depending on the cell type, more than one of these pathways is required for ECM-dependent survival.
Our in vitro results show that FAK signals to suppress anoikis using multiple pathways, and that distinct cell types utilise different subsets of these pathways. However, it is important to consider the relevance of this in vivo. Tissue-specific FAK knockouts support a model whereby the role of FAK to suppress anoikis shows cell lineage specificity. For example, FAK deletion in keratinocytes resulted in defects in epidermal thickness and hair growth, but there were no apoptosis defects in vivo. However, keratinocytes isolated from these animals died rapidly in vitro, indicating that FAK was required to support survival under certain conditions (Essayem et al., 2006). By contrast, deletion of FAK in endothelial cells resulted in an increase in apoptosis in vivo, leading to lethality before day 10.5 due to extensive haemorrhaging (Braren et al., 2006). Interestingly, whereas fibroblasts isolated from FAK -/embryos had defects in migration but not survival, FAK -/endothelial cells in vivo did not show defects in migration but did undergo apoptosis (Judson et al., 1999;Lark et al., 2003;Lark et al., 2005;Lightfoot et al., 2004;Owens et al., 1995;Owens et al., 1996). Together, these data indicate that how FAK controls cell survival depends very much on cellular origin as well as environmental context. This has implications for how FAK signalling may alter during tumour metastasis, when epithelial cells become more mesenchymal and invade diverse ECM types. Our data suggest that alterations in how FAK signals can affect cell survival in different ECM environments. FAK is upregulated in a range of tumours, including breast, colon, thyroid and ovarian (McLean et al., 2005). Indeed, the FAK locus is amplified in 79% of sporadic breast cancers (Naylor et al., 2005). Deletion of FAK in a mouse model of skin cancer blocked malignant progression (McLean et al., 2004).
In conclusion, our data indicate that FAK can signal via multiple, divergent pathways to suppress apoptosis. These results suggest that even though FAK may represent a common adhesion-dependent kinase activated by a wide range of integrins and extracellular matrices, it shows a complexity of downstream interactions that fine-tune cell-type-specific phenotypes.

Reagents
Anti-V5 was from Invitrogen. Anti-phospho-paxillinY31, anti-phospho-paxillinY118, anti-phospho-FAKY397, anti-phospho-FAKY925 and anti-phospho-FAKY577 were from Biosource (Nivelles, Belgium). The polyclonal anti-FAK was a gift from Andy Ziemiecki (University of Berne, Switzerland). Anti-actin was from Sigma and the anti-mtHsp70 was from Affinity Bioreagents (Golden, CO). Anti-p130Cas, antipaxillin and anti-phosphotyrosine (PY20) were from BD Transduction Labs. Antiphospho-p130Cas, anti-phospho-Akt and anti-phospho-paxillin were from Cell Signalling Technology (Danvers, MA). Secondary antibodies (anti-rabbit, anti-goat and anti-mouse peroxidase conjugates) were from Jackson Laboratories. Fluorescent   Fig. 8. MyrFAK signals through multiple, independent complexes to suppress apoptosis. (a) Mock-transfected HEK293T or HEK293T cells transiently expressing myrFAK constructs were detached for 15 minutes and treated with and without the membrane permeable crosslinker DSS. Cell lysates were immunoprecipitated with anti-V5. Immunoprecipitates were immunoblotted with anti-FAK Ab and the same immunoblot re-probed with anti-paxillin Ab. *, Position of the ~200 and ~300 kDa complexes seen following crosslinking. (b) Suppression of anoikis in MECs expressing myrFAK, myrFAKY925F or myrFAKI936/998E alone, or both myrFAKY925F and myrFAKI936/998E. Results are the mean of three independent experiments. Error bars indicate standard error. *, Significant suppression of apoptosis compared with either myrFAKY925F or myrFAKI936/998E alone. (c) HEK293T cells transiently expressing myrFAK constructs were detached for 15 minutes as indicated. Whole-cell lysates were immunoblotted as indicated. (d) MECs and MEFs were left adherent or detached for the indicated times. MECs were left in complete growth medium. MEFs were deprived of serum growth factors, except for the indicated adherent cells. Whole-cell lysates (WCL) were prepared and immunoblotted for total Akt or phospho-Akt serine 473. (e) MECs or MEFs expressing myrFAK (+), or non-expressing controls (-), were left adherent or detached as indicated. WCL were prepared and immunoblotted for phospho-Akt serine 473, total Akt, or myrFAK (anti-V5). conjugated secondary antibodies (anti-rabbit FITC and anti-mouse RRX) were from Alexis Corp. (Lausen, Switzerland). Hoechst was from Sigma. Rhodamine-phalloidin was from Molecular Probes.

Expression constructs
Constitutively active FAK was created by cloning a viral-Src myristoylation tag to the N-terminus of FAK. Oligonucleotides encoding the myristoylation sequence were annealed and ligated into pcDNA6/V5-His. Full-length wild-type FAK was then cloned in frame 3Ј to the myristoylation sequence. The oligonulceotides encoding the myristoylation tag were: Sense: 5Ј-CTAGCATGGGGAGCAGCAAGAGCAAGCCTAAGGACCCCAGC -CAGCGCCGGCGCCATATGTGGTAC-3Ј; antisense: 5Ј-CACATATGGCGC -CGGCGCT GGCTGGGGTCCTTAGGCTTGCTCTTGCTGCTCCCCATG.
Specific mutations in myristoylated FAK were created using site-directed PCR (QuikChange, Stratagene) and mutagenic oligonucleotides. The mutated bases are underlined and the sense strand is shown.
Cell culture and transfection FSK-7 mammary epithelial cells (MECs) (Kittrell et al., 1992) were cultured and transfected as previously described (Gilmore et al., 2000). Mouse embryonic fibroblasts (MEFs) were cultured in Dulbeccos modified Eagle's medium (DMEM)/10% fetal calf serum. Human embryonic kidney 293T cells were cultured in DMEM with 10% foetal bovine serum. Cells were plated at 1ϫ10 5 cells/cm 2 18 hours prior to transfecting with LipofectAMINE Plus (Invitrogen). For detachment assays, the cells were trypsinised and replated onto dishes coated with poly-HEMA (Sigma). The cells maintained on poly-HEMA were collected by centrifugation (5000 g, 30 seconds). For immunostaining, cells were centrifuged onto polysine slides (Merck, Whitehouse Station, NJ) using a Cytospin cytological centrifuge (Shandon). For apoptosis assays, cell death was quantified by nuclear morphology as previoulsy described (Gilmore and Streuli, 2002). Cells were stained with Hoechst and counted blind using an Axioplan2 microscope (Carl Zeiss MicroImaging). Apoptosis was analysed by ANOVA with Bonferroni's multiple comparison test to obtain P-values, using Prism 4 software (GraphPad, La Jolla, CA).

Subcellular fractionation
Adherent cells were scraped, and detached cells were collected by centrifugation (1200 g, 3 minutes). Following washing in ice-cold PBS, cells were resuspended in hypotonic buffer (10 mM HEPES-Cl, pH 7.5, 10 mM NaCl, 1.5 mM MgCl 2 , 1 mM Na 3 VO 4 , 4 mM NaF) supplemented with protease inhibitors (Calbiochem), and processed with a Dounce homogenizer (Wheaton) on ice. NaCl was then added to a final concentration of 150 mM. The lysates were centrifuged at 100,000 g for 30 minutes at 4°C. Supernatant (cytosol fraction) was saved, and the pellet (membrane fraction) was resuspended in an equal volume of NET buffer (50 mM Tris.Cl pH 7.6, 150 mM NaC, 1% NP-40, 2 mM EDTA, 10 mM NaF, 1 mM Na 3 VO 4 , protease inhibitors). After brief sonication, the pellet was centrifuged as before, and the supernatant was saved. Equivalent amounts of cytosol and membrane fractions were analysed.

Immunofluorescence and microscopy
Fixed cells on coverslips or polysine slides were permeabilised (PBS/0.5% Triton X-100). Cells were immunostained with primary Ab diluted in blocking solution (PBS, 0.2% Triton X-100, 0.05% Tween 20, 1% horse serum) at room temperature for 60 minutes. After being washed, the appropriate fluorescent-tagged secondary Ab was incubated as before. Nuclei were stained with Hoechst 33528, and F-actin was stained with Rhodamine-conjugated phalloidin. Coverslips were mounted with Prolong ® Antifade reagent and sealed prior to viewing with a Leica SP2 AOBS confocal system equipped with a ϫ63 objective lens. Images were processed in Photoshop (Adobe). Changes in brightness and contrast were applied to the entire image equally.

Immunoprecipitations
Cells were lysed in NET buffer and pre-cleared by incubating with 20 μl of protein A-sepharose or protein G-agarose beads (50:50 suspension) and rotating at 4°C for 30 minutes. Lysates were incubated with the precipitating Ab (rotating at 4°C for 2 hours) followed by the addition of 40 μl beads (rotating at 4°C for 1 hour). Immune complexes were sedimented (5000 g for 1 minute at 4°C) and washed three times with ice-cold lysis buffer. The pelleted immune complexes were resuspended in 2ϫ sample buffer and boiled before being resolved by SDS-PAGE.

GST fusion proteins pull down assays
GST-fusion proteins were expressed and purified from Escherichia coli onto glutathione agarose as previously described (Gilmore and Romer, 1996). Transiently transfected cells were lysed in NET buffer and incubated with 40 μl (50 % slurry) of the purified GST-fusion protein-coated glutathione-agarose (4°C/2 hours). Protein complexes were sedimented (5000 g for 1 minute at 4°C), and precipitated proteins were washed three times with 1 ml ice-cold lysis buffer. To recover the bound protein, the pelleted bead complexes were resuspended in 2ϫ SDS-PAGE sample buffer and boiled before being resolved by SDS-PAGE.

Protein crosslinking
Transfected cells were detached onto polyHEMA for 15 minutes. Cells were pelletted (1200 g for 3 minutes) and resuspended in 200 μl PBS. The membrane permeable crosslinker DSS (Pierce) was added to a final concentration of 0.75 mM. After 30 minutes, crosslinker was quenched with 50 mM Tris.Cl pH 7.0 for 15 minutes. Cells were pelleted and washed with 200 μl PBS before lysing in NET buffer. Cell lysates were resolved on 3-7.5% gradient SDS-PAGE gels.