Focal adhesion kinase (FAK) controls adhesion‐dependent cell motility, survival, and proliferation. FAK has kinase‐dependent and kinase‐independent functions, both of which play major roles in embryogenesis and tumor invasiveness. The precise mechanisms of FAK activation are not known. Using x‐ray crystallography, small angle x‐ray scattering, and biochemical and functional analyses, we show that the key step for activation of FAK's kinase‐dependent functions—autophosphorylation of tyrosine‐397—requires site‐specific dimerization of FAK. The dimers form via the association of the N‐terminal FERM domain of FAK and are stabilized by an interaction between FERM and the C‐terminal FAT domain. FAT binds to a basic motif on FERM that regulates co‐activation and nuclear localization. FAK dimerization requires local enrichment, which occurs specifically at focal adhesions. Paxillin plays a dual role, by recruiting FAK to focal adhesions and by reinforcing the FAT:FERM interaction. Our results provide a structural and mechanistic framework to explain how FAK combines multiple stimuli into a site‐specific function. The dimer interfaces we describe are promising targets for blocking FAK activation.
Crystal structure and functional data show how activatory autophosphorylation of focal adhesion kinase (FAK) is facilitated in physiological FAK dimers, which form upon paxillin‐mediated local enrichment.
This study elucidates the molecular mechanisms that allow FAK to be specifically activated at focal adhesions
Following FAK clustering at focal adhesions, FAK dimerizes through FERM:FERM and FERM:FAT interactions
Paxillin promotes FAK dimerization at focal adhesions by clustering and reinforcing the interaction of FAT with a basic patch on the FERM domain essential for localization and regulation.
The site‐specific dimerization promotes autophosphorylation of FAK and thus activates its kinase‐dependent functions
Compounds that alter FAK dimerization might be useful for therapeutic purposes, for example to block cancer invasiveness or metastasis.
Focal adhesion kinase (FAK) is a highly conserved non‐receptor tyrosine kinase (Corsi et al, 2006) enriched at focal adhesions (FAs) (Hanks et al, 1992; Schaller et al, 1992). It plays a critical role in the response to integrin‐mediated cell adhesion (Parsons, 2003; Mitra & Schlaepfer, 2006) and sensing of environmental rigidity (Geiger et al, 2009; Moore et al, 2010). FAK is also involved in signaling by other receptors, including G protein‐coupled receptors, Deleted‐in‐colon‐cancer (DCC) netrin receptors, and transmembrane tyrosine kinases (Zhao & Guan, 2009; Schaller, 2010). FAK is essential for embryonic development; FAK‐deficient mouse embryos die at E8‐9 (Ilic et al, 1995). Mice that express a FAK mutant lacking its autophosphorylation site die a few days later, pointing to the existence of autophosphorylation‐dependent and independent functions (Corsi et al, 2009). FAK is important for the formation of many organs including heart, brain, and blood vessels (Ilic et al, 1995; Rico et al, 2004; Shen et al, 2005; Braren et al, 2006) but the phenotype of kinase‐dead knock‐in mutants recapitulates only part of the complete knock‐outs (Zhao et al, 2010; Chen et al, 2012). Although FAK expression decreases in most adult non‐neuronal tissues (Burgaya et al, 1995), it is augmented in many tumors (McLean et al, 2005). Enhanced FAK expression allows tumor cells to override apoptotic signals triggered by cell detachment and to acquire increased invasiveness (Owens et al, 1995). Thus, FAK promotes cancer cell metastasis and resistance to chemotherapy, making it a major drug target in oncology (McLean et al, 2005).
FAK is a versatile multidomain scaffold (Fig 1A) with distinct functions in different subcellular environments through kinase‐dependent and kinase‐independent mechanisms (Corsi et al, 2009; Frame et al, 2010; Schaller, 2010; Arold, 2011). An intramolecular interaction between the N‐terminal 4.1–ezrin–radixin–moesin homology (FERM) domain (Girault et al, 1999) and the central kinase domain inhibits FAK kinase activity (Toutant et al, 2002; Ceccarelli et al, 2006; Lietha et al, 2007). Once recruited to FAs through an interaction between the C‐terminal FA targeting (FAT) domain and the integrin‐associated proteins paxillin (Hildebrand et al, 1993; Arold et al, 2002; Hayashi et al, 2002; Liu et al, 2002; Hoellerer et al, 2003) and talin (Lawson et al, 2012), FAK autophosphorylates on Y397 (Schaller et al, 1994). Y397 is located in the FERM‐kinase linker that also encompasses a proline‐rich motif (PR1) (Thomas et al, 1998). Phospho‐Y397 and PR1 recruit and activate Src family kinases (SFKs) by interacting simultaneously with their Src homology 2 (SH2) and SH3 domains (Schaller et al, 1994; Xing et al, 1994; Thomas et al, 1998; Arold et al, 2001). The activated SFK provides most of the FAK‐associated kinase activity and phosphorylates tyrosine residues of FAK and FAK ligands recruited by its multiple binding sites (Hall et al, 2011). In a sense, FAK acts as a scaffolding protein with an autophosphorylation‐dependent switch between sets of functions (Arold, 2011). A basic cluster (K216AKTLRK) in the FERM domain, distinct from its kinase interaction site, is involved in ligand‐mediated co‐activation of FAK at the cell membrane (Dunty et al, 2004; Chen & Chen, 2006; Cai et al, 2008). Intriguingly, the same cluster also constitutes a nuclear localization signal (NLS), which allows FAK to inactivate the tumor suppressor p53 in the nucleus, thereby promoting cell survival in a kinase‐independent manner (Golubovskaya et al, 2005; Lim et al, 2008). These diverse functions at the cell membrane and in the nucleus allow FAK to sense cell attachment and promote survival (Cance & Golubovskaya, 2008). However, it remains unclear how FAK is precisely activated at focal adhesions. This lack of knowledge limits our capacity to design function‐specific therapeutic agents.
Structural studies showed how paxillin binds to FAT (Hoellerer et al, 2003; Gao et al, 2004; Garron et al, 2008; Lulo et al, 2009) and how the enzymatic activity of the kinase domain is suppressed by its interaction with the FERM domain (Lietha et al, 2007), but these studies failed to reveal the regulation of the full‐length enzyme. We have therefore combined X‐ray crystallography with small angle X‐ray scattering (SAXS) to reconstruct the structure of full‐length FAK. Together with biochemical and functional analyses, our results reveal how FAK localization and kinase activity are regulated by dimerization and the interplay between multiple intramolecular and intermolecular interactions.
FERM domains form dimers in crystals, implicating W266
We determined the 2.8 Å‐resolution crystal structure of the human FERM domain (hFERM; Supplementary Table 1). As expected, the four hFERM molecules found in the asymmetric unit were very similar to each other (r.m.s.d. 0.5 Å) and also to the structure of the avian FERM domain (Ceccarelli et al, 2006; r.m.s.d. 0.7 Å, 94.4% sequence identity). Interestingly, the four hFERM molecules formed two arch‐shaped symmetric dimers in the crystal. The FERM:FERM binding surface was formed by the highly conserved W266 and surrounding residues (Fig 1B). We found the same FERM:FERM dimers in the crystal lattices of all three avian FERM constructs (Ceccarelli et al, 2006) [residues 31–399 (PDB accession number 2AEH), residues 31–405 (2AL6, 3ZDT)] and both avian FERM‐kinase structures (Lietha et al, 2007) (2J0K, 2J0J, Fig 1C), although these constructs crystallized under different conditions and in different crystal forms. Therefore, all currently known FERM domain structures (14 in total; from five different crystal lattices and two distinct species) formed the same dimer in the crystal. Our analysis with the PISA protein interface server (Henrick et al, 2008) showed that the FERM:FERM binding surface covered about 1350 Å2 and displayed features compatible with a bona fide protein‐protein interface, rather than a crystal packing artifact (solvation free energy gain [ΔGint], −11.5 kcal/M). We therefore hypothesized that the high protein concentration during crystal formation stabilized a biologically relevant dimeric arrangement.
FAK self‐association in cells requires W266
To test whether dimerization occurs in cells, we first co‐transfected HEK293 cells with VSV‐ and HA‐tagged FAK. In support of FAK dimerization, antibodies against the VSV tag immunoprecipitated HA‐FAK (Fig 2A). This interaction was strongly decreased when both tagged proteins contained the W266A mutation (Fig 2A and B), in agreement with the key role of W266 in the FERM:FERM interaction suggested by our crystallographic analysis. Because of avidity effects and coupling of interactions, a FAK dimer is expected to associate more strongly with paxillin and other FA‐localized ligands than is a FAK monomer. We therefore expected that the W266A mutation, by disrupting the FAK dimer, would also weaken the recruitment of this mutant to FAs, without directly affecting the FA‐targeting sites located on the FAT domain. To test this hypothesis, we examined the cellular location of HA‐tagged FAK or W266A‐FAK in transfected FAK‐defective (Ptk2−/−) fibroblasts (Ilic et al, 1995). In HA‐FAK‐transfected cells, confocal analysis revealed that HA immunoreactivity was co‐localized with paxillin at FAs (Fig 2C, upper panels), as expected. HA‐W266A‐FAK was also detected at FAs, but a larger proportion was cytosolic than for the wild‐type (WT), as predicted by our model (Fig 2C and D, lower panels). To obtain a more quantitative evaluation, we transfected WT or W266A mCherry‐FAK in Ptk2−/− cells (Fig 2E) and measured the fluorescence at FAs and in the perinuclear cytoplasm (Fig 2F). The recruitment of the W266A mutant FAK to FAs was decreased by 37% as compared to the WT protein (Fig 2F).
Source Data for Figure 2A [embj201386399-SourceData-Fig2FAK-Fig2A.pdf]
We then studied FAK:FAK interactions in Ptk2−/− cells co‐transfected with FAK fused to EGFP or mCherry by acceptor photobleaching. In these experiments, the existence of fluorescence (Föster) resonance energy transfer (FRET) between the EGFP and mCherry was used as an index of the proximity of the FAK fusion proteins. To ensure that FAK dimers were not artifacts of abnormally high expression of the transfected protein, we selected for analysis only cells that showed FAK expression levels < 2‐fold those of endogenous FAK. To do so we compared the endogenous and transfected FAK immunoreactivity in wt and Ptk2−/− cells, respectively (Supplementary Fig S1A and B), and used it as standard for selecting cells based on EGFP fluorescence for FRET experiments. The ratio of FAK concentration in the cytoplasm to FAK concentration at FAs was similar for endogenous and transfected FAK, showing that transfection did not alter the protein distribution. The validity of using transfected FAK as a proxy for endogenous FAK has been demonstrated previously (Webb et al, 2004; Hamadi et al, 2005). A strong FRET efficiency was observed at FAs with WT FAK but not in the rest of the cytoplasm (Fig 3A upper panels, Fig 3B). When cells were transfected with WT FAK‐mCherry and W266A‐FAK‐GFP the FRET efficiency was dramatically decreased (−80%, Fig 3A and C). When the transfection was done with W266A‐mCherry and W266A‐FAK‐GFP, the effect was even more pronounced (−90%, Fig 3A and C). The decreased FRET efficiency of W266A‐FAK could result from its diminished enrichment at FAs and/or from altered dimerization. To distinguish between these two possibilities we examined the interactions of FAK‐mCherry with paxillin‐GFP (Fig 3D, Supplementary Fig S1C). We reasoned that diminution of W266A‐FAK enrichment at FAs should similarly alter FAK:FAK and FAK:paxillin FRET efficiency, whereas impaired homodimerization should preferentially alter FAK:FAK FRET. The FRET efficiency between paxillin and WT FAK was significant (Fig 3D), although less pronounced than for the FAK:FAK interaction (Fig. 3C, D). The W266A mutation decreased the FAK:paxillin interaction much less than it decreased the FAK:FAK interaction. Therefore, we concluded that the effect of the W266A mutation on FAK:FAK FRET was mostly explained by the disruption of FAK dimerization, and only to a minor degree by a decreased recruitment to FAs.
FERM:FERM interactions and FAK dimerization of recombinant purified proteins
We first investigated the ability of purified recombinant FERM to self‐associate. Surface plasmon resonance (SPR) showed that FERM self‐associated in the absence of other partners and that the W266A mutation abolished this interaction (Supplementary Fig S2A). Since determination of accurate affinities by SPR can be hampered by multiple factors, including ligand or analyte dimerization, occurrence of multiple interactions, and accuracy of the fit, we estimated the FERM:FERM affinity by analytical ultracentrifugation (AUC, Supplementary Fig S2B). This method provided an apparent Kd of 29 μM (95% confidence interval, 17–49 μM) for the FERM:FERM interaction.
We then produced full‐length WT FAK as a His6 fusion protein in a baculovirus system, purified it on a Ni2+ column (Supplementary Fig S2C) and examined its ability to self‐associate. Polyacrylamide gel electrophoresis under native conditions revealed a dimeric species of full‐length FAK, which disappeared in the presence of increasing amounts of SDS (Fig 4A) or in W266A‐FAK (Fig 4B). Dynamic light scattering (DLS) experiments indicated the presence of two major FAK species with Stokes radii (RS) of 48 Å (~50%) and 66 Å (~25%). The molecular masses calculated from these radii were 130 and 280 kDa, respectively, in agreement with the masses of FAK monomers (125 kDa) and dimers (250 kDa). Similarly, size‐exclusion chromatography (SEC) revealed two major species for WT FAK, with RS of 52 Å and 69 Å (Fig 4C and D). The 69‐Å peak was markedly decreased for W266A‐FAK (Fig 4E). Both species corresponded to FAK as shown by Coomassie blue‐stained SDS‐PAGE of elution fractions (Fig 4F). These results demonstrated that purified FAK can form dimers in solution, in the absence of other factors. Importantly they also indicated that the W266A mutation, which abolishes the FERM:FERM interaction, does not completely suppress the FAK:FAK interaction. Moreover, the 29 μM affinity we measured by AUC for the FERM:FERM interaction would be too low to be detected by most methods. The observation of FAK dimers in native SDS‐PAGE, DLS, and SEC, using only sub‐micromolar concentrations of FAK, however, suggested that additional interactions, not included in the FERM‐kinase fragment, stabilize FAK dimers. These additional interactions would also explain the small amounts of higher molecular weight species observed in DLS and native SDS‐PAGE gels (Fig 4A and B), as well as the residual interactions of full‐length W266A‐FAK in SPR (not shown). We therefore used additional approaches to investigate FAK self‐association.
Source Data for Figure 4BCG [embj201386399-SourceData-FAK-Fig4BCG.pdf]
SAXS analysis of full‐length FAK confirms FERM:FERM interactions and suggests FAT:FERM interactions
The size, flexibility and low in vitro solubility of FAK have so far precluded structural analysis of the full‐length molecule by X‐ray crystallography, nuclear magnetic resonance or electron microscopy. We therefore used SAXS to investigate the intramolecular interactions occurring in full‐length FAK in solution. We first used an iterative approach to calculate ab initio low‐resolution bead models. The SAXS data were compatible with the particle size expected for a FAK dimer (Fig 5A and B, Supplementary Fig S3) and the averaged structure displayed dimeric features even without symmetry constraints (Supplementary Fig S3A). The central part of the SAXS bead model was highly similar to the arch shape of the crystallographic FERM:FERM dimer. Moreover, the crystallographic FERM‐kinase homodimers (PDB 2J0K or 2J0J) fitted well into a large part of the SAXS bead model (Fig 5A). After this fit, the SAXS envelope provided additional space near the FERM F2 lobe, likely to harbor many of the ~400 residues missing in the FERM‐kinase structure. In a second SAXS analysis, we used the homodimeric FERM‐kinase fragment as a rigid body and placed the FAT domain and the other missing residues (the 38 N‐terminal residues and the 220 kinase‐FAT linker residues) directly according to the SAXS scattering pattern. This algorithm did not take into account the SAXS ab initio bead model. In the 10% best‐scored of the ~500 produced models, the four‐helical FAT domain was consistently placed in contact with the FERM F2 lobe, close to the site where FERM binds to the C‐terminal kinase lobe, suggesting a direct interaction between the FERM and FAT domains (Fig 5A, Supplementary Fig S3C). The RS of 67 Å calculated for a compact dimeric model (Ortega et al, 2011) (featuring FERM:FERM and FAT:FERM interactions and a flexible kinase‐FAT linker) matched the RS measured by DLS (66 Å) and SEC (69 Å) for the dimeric species. Conversely, the average RS for dimeric FAK models in which the FAT domain was not attached to FERM was significantly larger (89 Å). The SAXS results therefore suggested that FAK dimers are stabilized by a FAT:FERM interaction.
Source Data for Figure 5C [embj201386399-SourceData-FAK-Fig5C.pdf]
FAT binds to a basic cluster on FERM and thus stabilizes the FAK dimer
Since the FAT:FERM interaction suggested by our SAXS study had not been reported previously, we tested it in vitro. In pull‐down experiments, purified recombinant FAT displayed a significant interaction with GST‐FERM, as compared to GST alone (Fig 5C). This interaction was confirmed using isothermal titration calorimetry (ITC), yielding a Kd of 0.6 ± 0.2 μM (Supplementary Fig S4A). We also observed the association between immobilized FAT and GST‐FERM or FAT and immobilized GST‐FERM by SPR (Supplementary Fig S4B). The putative FAT binding region identified by SAXS within the FERM F2 lobe encompassed a basic cluster, K216AKTLRK, which is important for FAK function (Dunty et al, 2004; Chen & Chen, 2006; Lim et al, 2008). We mutated the four basic residues in this cluster to alanine (K216AKTLRK to A216AATLAA). It is important to point out that it has been shown that this mutation does not compromise the 3D structure of FERM, as demonstrated by the crystal structure of this FERM mutant (PDB 3ZDT). The A216AATLAA mutation abolished binding of FAT to GST‐FERM in GST pull‐down (Fig 5C) and SPR experiments (Supplementary Fig S4A). These results confirmed that FAT binds directly to the FERM domain, and they revealed the importance of the K216AKTLRK motif for this interaction. In support of a role of this FAT:FERM interaction for the stabilization of FAK dimers, we found that the peak corresponding to dimeric FAK in SEC was dramatically reduced (Supplementary Fig S4C–E) when we incubated FAK with a 10‐fold excess of recombinant FAT prior to chromatography.
The FAT domain includes two binding sites for the LD motifs of paxillin (Hoellerer et al, 2003), an interaction which targets FAK to FAs (Brown et al, 1996). We examined whether the binding of a 15‐residue paxillin LD4 peptide (LD4‐pep) known to bind to these two sites of FAT (Hoellerer et al, 2003), interfered with FAT:FERM interactions. Surprisingly, the LD4‐pep did not decrease FAT binding to GST‐FERM, but enhanced it (Supplementary Fig S4F, Fig 5D). This effect was specific since a control peptide with a LD4 scrambled sequence (LD4‐scr) did not enhance binding and the LD4‐pep did not promote FAT association with GST (Supplementary Fig S4G and H). The LD4‐pep also increased GST‐FERM binding of a shorter FAT construct restricted to the 4‐helix bundle (FAT916–1024), indicating that this bundle is sufficient for both FERM binding and its enhancement by the paxillin peptide (Supplementary Fig S4F, Fig 5D). It is important to note that we did not detect any binding of LD4 to FERM (data not shown). The enhancement of the FAT:FERM interaction by the LD4‐pep, which binds to the two paxillin binding sites of FAT, far from indicating a possible competition, revealed a potential synergy between paxillin binding and FAK dimerization.
FAK autophosphorylation of residue Y397 occurs in trans and requires W266
Having shown that FAK can form dimers through FERM:FERM and FAT:FERM contacts, we examined the functional relevance of this self‐association. Autophosphorylation of Y397 is the central step in activation of kinase‐dependent functions of FAK. Previous studies suggested that in intact cells and in vitro, FAK autophosphorylation can be intermolecular and that it is enhanced by induced dimerization (Katz et al, 2002; Toutant et al, 2002). To test if FAK autophosphorylation requires FAK dimerization, we used purified FAK in a dilution assay. We found that the specific autophosphorylation activity increased with protein concentration, as expected for an intermolecular reaction (Fig 6A). The W266A mutation completely abolished autophosphorylation (Fig 6A–C). To rule out an unspecific effect of the mutation on the enzyme structure, we tested if forced dimerization of W266A‐FAK by the addition of sepharose‐coupled FAK antibodies could restore phosphorylation of Y397. Indeed, in the presence of antibodies, the autophosphorylation of W266A‐FAK was restored and was virtually identical to that of WT FAK incubated under the same conditions (Fig 6B–D).
Source Data for Figure 6BE [embj201386399-SourceData-FAK-Fig6BE.pdf]
We then explored the effects of the W266A mutation on FAK phosphorylation in transfected COS7 cells. Phosphorylation of Y397 and total tyrosine phosphorylation were dramatically reduced in W266A‐FAK as compared to FAK (Fig 6E). These results supported the possibility that full‐length FAK is autophosphorylated mostly through an intermolecular mechanism, in contrast to the FERM‐kinase moiety, which was reported to phosphorylate itself intramolecularly (Lietha et al, 2007). Moreover, our results showed that autophosphorylation of Y397—and hence activation of the kinase‐dependent functions of FAK—is linked to dimerization.
Mutation of FAK W266 alters focal adhesions turnover
Finally, we investigated the consequences of impaired dimerization on FAK function in cells. FAK‐null fibroblasts display more FAs than do WT fibroblasts (Ilic et al, 1995), suggesting that FAK plays a role in FA turnover. As expected, we observed fewer FAs in Ptk2−/− fibroblasts transfected with FAK than in non‐transfected cells (Fig 7A). In contrast, expression of W266A‐FAK was significantly less efficacious in decreasing the number of FAs in Ptk2−/− cells (Fig 7A), indicating that the mutation impaired the role of FAK in FA turnover. We then directly investigated FA stability by live imaging of GFP‐paxillin with a confocal spinning disk microscope (Fig 7B). Ptk2−/− fibroblasts expressing FAK had a FA dissociation rate [(5.4 ± 1) × 10−2 min−1, mean ± SEM, n = 20] comparable with previously published values (Webb et al, 2004; Deramaudt et al, 2011). In cells transfected with W266A‐FAK, the dissociation rate was lower [(3.5 ± 0.4) × 10−2 min−1, n = 20, t‐test P < 0.01] (Fig 7C). The number of stable FAs was lower in cells transfected with WT FAK compared with those transfected with W266A‐FAK (Fig 7D). Thus, FAs were less stable and more mobile in fibroblasts expressing WT FAK than in cells expressing W266A‐FAK.
We next examined the role of FAK dimerization in GFP‐paxillin dynamics by fluorescence recovery after photobleaching (FRAP). We bleached GFP‐paxillin in peripheral FAs or in the cytoplasm at a distance from FAs, in Ptk2−/− fibroblasts co‐transfected or not co‐transfected with FAK or W266A‐FAK (Fig 7E). FAK increased the mobile fraction of GFP‐paxillin at FAs, whereas W266A‐FAK had no effect (Fig 7F). FAK transfection did not alter GFP‐paxillin mobility in the cytoplasm away from FAs (Fig 7G). The effect of FAK on GFP‐paxillin was due to an increased mobile fraction at FAs without significant change in its time constant (Fig 7H). No effect was observed in the cytoplasm (Fig 7I). Altogether, our experiments in living cells showed that FAK's capacity to dimerize transiently is central to its functions at FAs, including FA disassembly and turnover.
FAK exerts kinase‐dependent and kinase‐independent functions. Both types of function are crucial for embryogenesis and tumor invasiveness, albeit through different mechanisms and at different sites. Here, we show that activation of kinase‐dependent functions at FAs is regulated through transient, ligand‐promoted FAK dimerization. We observed FAK self‐association in vitro (using SEC, SPR, AUC, DLS, SAXS, and crystallography) and in cells (using co‐immunoprecipitation of transfected FAK and confocal microscopy under quasi‐endogenous expression levels). We showed that self‐association of full‐length FAK requires the formation of a FERM homodimer and that this FERM dimerization is dependent on W266. Interestingly this residue is conserved in Pyk2 but not in other FERM domain proteins. Pyk2 has been reported to undergo a Ca2+/calmodulin‐induced dimerization, through an unknown structural mechanism (Kohno et al, 2008) and it will be important to investigate the contribution of the FERM:FERM interface we report here. We also found that the FERM:FERM interaction alone is relatively weak (Kd ~ 30 μM), and FAK requires additional FAT:FERM interaction to stabilize the dimeric state. For this, FAT binds the FERM K216AKTLRK sequence presumably in trans, a conformation compatible with constraints from SAXS and stereochemistry. This conclusion is supported by the dissociation of the FAK dimer in the presence of an excess of FAT. This FAT:FERM interaction also provides an explanation for the requirement of the K216AKTLRK motif on the FERM domain to interact with another FAK molecule (Dunty et al, 2004). FAT has also been shown to form helix‐exchange dimers in vitro (Arold et al, 2002). Although we cannot exclude that such FAT:FAT interaction contribute to FAK self‐association in some conditions, their contribution is expected to be minor, since the structural changes necessary to promote these FAT:FAT interactions presumably occur with a low probability.
Importantly, we observed that binding of paxillin LD4 increased the FAT:FERM interaction, an effect expected to stabilize the dimer further. This unsuspected role of FAT may explain its requirement for the dimerization‐induced autophosphorylation of FAK (Toutant et al, 2002). The interaction of the FAT domain with K216AKTLRK clarifies the hitherto mysterious role of this motif in FAK activation (Dunty et al, 2004). This motif also mediates interaction of FERM with other ligands, such as phospho‐cMet (Chen & Chen, 2006) or phospholipids (Cai et al, 2008; Papusheva et al, 2009). These interactions may constitute an additional proofreading mechanism and their possible interdependence with FAT, FERM, and paxillin complex formation remains to be investigated.
The combination of multiple interactions in full‐length FAK results in features that are not observed in the FERM‐kinase fragment alone (Lietha et al, 2007), namely dimer formation and the occurrence of autophosphorylation as a trans reaction. Although, for reasons of feasibility, we could not determine a precise monomer:dimer apparent Kd for full‐length FAK, this Kd appears to be of the order of 0.1–0.5 μM as judged from the FAK concentrations used and amount of dimer found in DLS, SEC, SAXS and our autophosphorylation assay. Interestingly, this suggests that at the estimated average concentration of FAK in cells (on the order of 10 nM, see Methods), FAK dimers, and hence FAK autophosphorylation, are unlikely to occur at a significant level. Indeed, we found by acceptor photobleaching that FAK forms dimers in cells specifically at FAs, where FAK is enriched through membrane‐associated ligands. Our data therefore supports that paxillin plays a key role not only by recruiting FAK, but also by stabilizing the FAT:FERM interaction. The resulting FAK dimerization triggers autophosphorylation and activation of kinase‐dependent functions. This activation mechanism is reminiscent of receptor tyrosine kinases, for which extracellular ligand‐induced dimerization triggers trans‐autophosphorylation (Lemmon & Schlessinger, 2010). For autophosphorylation of Y397 to occur, the kinase domains need to dissociate from the FERM domains. Once dissociated, the FERM‐kinase and kinase‐FAT linkers are long enough to allow phosphorylation in trans without disrupting the FERM:FERM and FAT:FERM interactions. The mechanisms that promote dissociation of the FERM:kinase interaction are currently not completely understood and may be facilitated by additional ligands of the FERM domain.
By identifying paxillin–binding as a potential regulator of FAK dimerization, our results may also provide a rationale for the absence of kinase activation in nascent adhesions, where paxillin interacts with Nudel rather than with FAK (Shan et al, 2009). It is interesting to note that FERM deletion enhances FAK activity (Chan et al, 1994; Schlaepfer & Hunter, 1996) without altering its responsiveness to adhesion (Jacamo & Rozengurt, 2005) or its requirement for trans‐autophosphosphorylation (Toutant et al, 2002). Therefore the presence of the FERM domain is not essential for the “basic” activation of FAK, but brings an additional layer of control on its activity that contributes to an exquisite site‐specific regulation through multiple protein‐protein interactions.
Finally, the interaction of FAT with FERM may also play a role in regulating FAK intracellular localization. Indeed, the K216AKTLRK sequence is part of a nuclear localization signal (NLS) (Lim et al, 2008). The intra‐ or intermolecular interaction of FAT with K216AKTLRK is likely to conceal the NLS and inhibit FAK nuclear localization, as indicated by the nuclear localization of the FERM domain or of truncated forms of FAK missing only the C‐terminal region (Lobo & Zachary, 2000). Interestingly, our results raise the possibility that MBD2, which interacts with FAT to enhance FAK nuclear localization (Luo et al, 2009), may act by disrupting the FAT:FERM interaction and revealing the NLS. In the absence of specific local enrichment, nuclear FAK would be expected to remain monomeric, an hypothesis compatible with its function as a kinase‐independent scaffold inhibiting pro‐apoptotic factors such as p53 (Lim et al, 2008).
In conclusion, our results show how the interactions of the FERM, FAT, and kinase domains, regulated by FA targeting through paxillin, control FAK dimerization and hence autophosphorylation‐triggered activation of kinase‐dependent functions. We propose that this interplay endows FAK with a capacity for coincidental detection of multiple protein‐protein interactions, providing checkpoints to prevent premature or ectopic activation. The close association of distinct conformations with FAK activation also provides a molecular framework for understanding the role of FAK in mechanotransduction (Geiger et al, 2009; Moore et al, 2010). These new insights into the molecular organization and dynamics of FAK activation suggest that specific disruption of inter‐ or intramolecular interactions should be promising therapeutic approaches for selective inhibition of distinct functions.
Materials and Methods
Antibodies and Reagents
Mouse monoclonal antibodies were from the following sources: FAK (4.47, 1:500 for immunoblotting), Upstate Biotechnology, phospho‐Y397‐FAK (1:1,000), Biosource, phosphotyrosine (4G10, 1:2000), Upstate Biotechnology, paxillin (1:1000), Invitrogen. C‐20 FAK antibodies were raised in rabbits against a peptide encompassing the 20 C‐terminal residues (Toutant et al, 2000). Rabbit polyclonal HA antibody was from Zymed (1:500); anti‐VSV antibody was from Sigma (1:500). All other reagents were from Sigma unless otherwise specified. The peptides were: paxillin LD4 domain: SATRELDELMASLSD; scrambled peptide: LSDAMETSSLRDALE.
Molecular Biology Constructs
Recombinant N‐terminally His6‐tagged FAK and His6‐tagged W266A‐FAK were produced in a baculovirus‐based expression system (Invitrogen). Rat FAK cDNA was amplified by PCR from pCMV2‐FAK using a forward primer introducing a BamHI site and a reverse primer bearing a KpnI site, and then they were cloned into the similarly digested pFastBac HT B donor plasmid (Invitrogen). Baculovirus was prepared as recommended by the manufacturer. The FAK coding sequence was inserted into pmCherry‐C1 between Bgl2 and BamH1 sites. EGFP‐FAK was described previously (Toutant et al, 2000). W266A FAK was introduced into mCherry FAK construct using EcoR1 and Xho1 and from this construct to the EGFP vector using EcoRI and BamH1. All constructs were verified by sequencing.
Protein Production and Purification
Recombinant Sf9 cells were cultured in Insect‐XPRESS™ Medium (Cambrex Bio Science). Cells were harvested 48 h after infection and recombinant proteins were extracted and purified in “native conditions” with a nickel‐chelating resin (ProBond system, Invitrogen) according to the manufacturer's protocol. Recombinant GST‐tagged FERM (FAK1–361), FERM mutant, FAT (FAK895–1054) and mutant proteins were produced in BL21 E. Coli cells. Bacteria were grown at 37°C and protein expression was induced by the addition of 0.4 mM IPTG after which cells were grown for 3 h at 30°C. Cell lysis was done by sonication in phosphate‐buffered saline (PBS), 1% Triton X‐100 (v/v) and 0.5 mM dithiothreitol (DTT). After sonication, cell debris was removed by centrifugation and lysates were incubated with glutathione‐agarose beads. After 5 washes, purified proteins were eluted by competition with 10 mM glutathione in 50 mM Tris pH8, 0.5 mM PMSF. After purification, FAT was cleaved with 3c protease (Arold et al, 2002). For crystallization, the F85L and W181G mutations were introduced into recombinant hFERM31–405 to increase solubility. hFERM31–405;F85L,W181G was produced and purified as described (Ceccarelli et al, 2006).
Biochemical and structural studies of purified proteins
Non denaturing polyacrylamide gel electrophoresis was carried out on 3–12% Bis‐Tris NativePAGE gel (Invitrogen) following the manufacturer's recommendations. Dynamic light scattering was carried out with a DynaPro Protein Solutions apparatus (Wyatt, Paris‐Descartes CNRS UMR 8015 facility) using 1 mg/ml His6‐FAK. In the pull‐down experiments, GST‐FERM and mutant proteins were incubated with FAT for 2 h at 4°C in Hepes, 50 mM, NaCl 150 mM, glycerol 10% (v/v), MgCl2 1.5 mM, Triton X‐100 1% (v/v) before precipitation with glutathione sepharose. After three washes with Hepes 50 mM, NaCl 150 mM, Triton X‐100 1%, glycerol 10%, the proteins were subjected to SDS‐PAGE followed by Coomassie or immunoblot analysis. Size exclusion chromatography experiments were done with an Aktä 900 purifier (GE Healthcare) using a Superdex 200 10/300 GL column (GE Healthcare). The column was equilibrated in a buffer containing 200 mM NaCl, 50 mM NaH2PO4 pH 8. Calibration was done using the Native markers set (Sigma). Purified FAK or FAT protein in 500 μl of buffer was injected and 200 μl fractions were collected and subjected to SDS‐PAGE followed by Coomassie staining (Biosafe, Biorad). Analysis of elution profiles was done using the Aktä software. The SPR, ITC, AUC, and SAXS experiments are described in Supplementary Material.
For these assays, His6‐FAK was dephosphorylated with the catalytic domain of receptor‐like protein tyrosine phosphatase‐β fused to GST for 10 min at room temperature before removal of the phosphatase with glutathione sepharose. Autophosphorylation assays were performed at 20°C with various amounts of FAK for 1 min in phosphorylation buffer containing 50 mM Hepes at pH 7.4, 10 mM MnCl2, and 10 μM ATP in the presence of a protease inhibitor cocktail (Complete, Roche) and sodium orthovanadate (1 mM). In the induced dimerization experiments, protein A‐sepharose‐coupled C20 FAK antibodies (Toutant et al, 2002) were added and phosphorylation was carried out in the resuspended pellet. The autophosphorylation assays were stopped by adding a 5‐fold‐concentrated solution at 100°C (150 g/l SDS, 0.3 M Tris‐Cl pH 6.8, 25% glycerol) and placing the sample at 100°C for 5 min.
Immunoprecipitation and immunoblotting
In the FAK immunoprecipitation experiments, 24 h after transfection, COS7 or HEK 293 cells were lyzed in modified RIPA buffer as described previously (Toutant et al, 2002). Briefly, lysates were precleared by incubation for 1 h at 4°C with 100 μl of a mixture (50% v/v) of Sephacryl and protein G Sepharose beads (GE Healthcare). Immunoprecipitation was carried out overnight at 4°C with 5 μl of polyclonal anti‐HA antibody and 50 μl protein G Sepharose. After 3 washes, beads were resuspended in Laemmli loading buffer, placed at 95°C for 5 min and subjected to SDS‐PAGE. Proteins were transferred to nitrocellulose (GE Healthcare or Bio‐rad), immunoblotted with the appropriate antibodies and visualized either with IR‐Dye 800 CW or IR‐Dye 700 CW donkey anti‐mouse/anti‐rabbit IgG antibodies (1:4,000, Rockland) and detection by infrared fluorescence with an Odyssey Li‐Cor scanner or anti‐rabbit IgG coupled with horseradish peroxidase (1:5,000, Cell Signaling) and revealed with a horseradish peroxidase‐enhanced chemiluminescence system (Immobilon Western – Chemiluminescent HRP Substrate – Millipore).
Cell culture, transfection, and immunocytochemistry
Ptk2−/− fibroblasts (Ilic et al, 1995) were cultured in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal calf serum (FCS) plus 0.001% (v/v) βME. COS7 and HEK cells were cultured in DMEM supplemented with 10% FCS. Cells were transfected with 2 μg DNA per 14‐mm diameter culture dish, and 4 μg DNA per 60‐mm dish with Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions. In the co‐transfection experiments, 0.3 μg GFP‐paxillin and 0.9 μg FAK plasmids per 3.5‐cm dish, or 3 μg VSV‐FAK and 3 μg HA‐FAK plasmids per 10‐cm dish were used. The total quantity of DNA was maintained with an empty pcDNA3 plasmid. Twenty‐four hours after transfection, the cells were plated on 24‐well plates coated with fibronectin and then fixed 24 h later for 20 min in PBS containing 4% (w/v) paraformaldehyde. After three rinses in PBS, the cells were permeabilized for 10 min with 0.2% Triton X‐100 in PBS, treated with blocking buffer (PBS containing 5% w/v BSA) for 30 min and then incubated overnight with antibodies for HA and paxillin at 4°C in PBS containing 1% BSA. After the rinses, the cells were incubated sequentially for 1 h at room temperature with goat Cy3‐coupled anti‐rabbit and Alexa‐488‐coupled anti‐mouse antibodies. After three rinses, the cells were mounted under coverslips using Vectashield‐containing DAPI (Vector Abcys). Images were acquired at the Institut du Fer à Moulin Tissue and Cell Imaging Facility with a laser scanning confocal microscope (Leica and SP5). The cells with low expression of transfected protein (s), comparable to endogenous FAK levels, were chosen for analysis in all imaging experiments.
Live cell imaging
Transfected cells were plated at low density on 35‐mm μ‐dish (Ibidi) coated with fibronectin (10 μg/ml, Sigma). Twenty‐four hours later, the cells were placed at 37°C in HBSS medium on a DMI 4000 inverted microscope (Leica Microsystems) with a confocal spinning‐disk head (CSU 22; Yokogawa) and a 491 nm laser (MAG Biosystems). Images were acquired with a 63× HCX‐PL‐APO (1.4 NA) objective every 5 min on an EMCCD camera (QuantEM 512 SC, Photometrics) and analyzed using Metamorph software (Molecular Devices). Up to five different fields were sequentially recorded during each experiment using a Märzhäuser (Wetzlar) automated stage piloted by Metamorph. Ratios of fluorescence at FAs at each time point to that of the same FAs at time 0 were calculated. The disassembly of FAs was linear on a semi‐logarithmic plot of the fluorescence intensity as a function of time. The apparent rate constants for FA disassembly were determined from the slope of these graphs. The number of stable FAs was determined by analysis of the merged image combining the images of the cell visualized at t = 0 in red and the same cell at t = 60 min in green. Disassembled FAs appeared in red, stable FAs in yellow and newly formed FAs in green.
The cells were placed at 32°C in HBSS medium on a Leica SP5 II upright microscope (Leica Microsystems). Images were acquired with a 40× HCX APO (0.80 NA) water immersion objective and the FRAP experiment was performed with the FRAP Wizard software from Leica Microsystems. Five images where taken at low laser intensity (~ 5%) before the bleach at the rate of 1 Hz for measuring basal fluorescence intensity. Photo‐bleaching was done at 100% of the 488 nm laser line with four iterations. Recovery was followed with the same laser power as in the pre‐bleached session at the same imaging rate for 60 s. For each time point, the intensity of the bleach area was normalized to the pre‐bleached intensity. FRAP recovery curves and analysis where generated using Igor Pro software (WaveMatrics).
Fixed cells were acquired using a Leica TCS SP5 upright confocal microscope using the FRET acceptor photobleaching wizard and a 40× 0.8 NA water immersion objective (Leica Microsystems). Pre‐bleach and post‐bleach images were serially recorded by excitation of GFP at 488 nm (donor channel) with an argon laser and mCherry at 561 nm (acceptor channel). Low laser intensities were used to avoid bleaching effects during acquisition. Cells were selected by visualizing only the donor channel to prevent premature partial bleaching of the acceptor. The acceptor was bleached with high intensity (100%) power at the 543 nm laser line for two iterations. Images were analyzed using Matlab (Mathworks). The change in the fluorescence intensity between pre‐ and post‐bleach donor values (efficiency, E) was calculated using the formula E = (donor after−donor before) × 100/donor after, and was shown as a percentage; pseudo‐colored images showing FRET efficiency values were also generated.
Evaluation of FAK concentration in cells
NIH3T3 cells were cultured in DMEM medium supplemented with 10% FCS. At confluence, the cells were trypsinized, counted, and solubilized at 100°C in 1% SDS solution. The amount of FAK was determined by immunoblotting in comparison with known and increasing amounts of recombinant FAK protein. The FAK concentration in cells was then estimated based on the known number of cells in the sample multiplied by a calculated cell volume (approximated as a 20‐μm diameter sphere). By this method, the mean FAK concentration in cells was estimated at 5–6 nM.
Biophysical and structural methods
The procedures for surface plasmon resonance (SPR), isothermal titration calorimetry (ITC), analytical ultracentrifugation (AUC), X ray crystallography, SAXS analysis, and construction of FAK models are described in the Supplementary Information. The coordinates of hFERM have been deposited at the PDB, with accession number 4NY0.
KB‐C and NG, conceived and carried out experiments, interpreted results, and participated in manuscript writing. STA planned and coordinated the study, conceived and carried out experiments, interpreted results, and wrote the manuscript. J‐AG planned and coordinated the study, conceived experiments, interpreted results, and wrote the manuscript. DA, KW, M‐CB, AO, PGL, BS, LG, TB, and GK carried out specific experiments and/or calculations.
Conflict of interest
The authors declare that they have no conflict of interest.
Supplementary Figure S1 [embj201386399-sup-0001-FigS1.pdf]
Supplementary Figure S2 [embj201386399-sup-0002-FigS2.pdf]
Supplementary Figure S3 [embj201386399-sup-0003-FigS3.pdf]
Supplementary Figure S4 [embj201386399-sup-0004-FigS4.pdf]
Supplementary Table S1 [embj201386399-sup-0005-TableS1.pdf]
Supplementary Information [embj201386399-sup-0006-Data.pdf]
We thank S. Lachkar for help with SEC, K. Muller and V. Unkefer for editorial assistance, J.E. Ladbury for access to the MicroCal iTC200, L. Ponchon for help with DLS, D. Svergun and his colleagues for assistance with SAXS data recording and analysis at the ×33 beamline at the European Molecular Biology Laboratory/Deutsches Elektronen‐Synchrotron, G. Meigs for help with X‐ray crystallography data recording at the Advanced Light Source beamline bl.8.3.1., Berkeley, CA, D. Ilic for the gift of Ptk2−/− fibroblasts, and A. Sobel and R.M. Mège for critical reading of the manuscript. We acknowledge support with data recording at the European Synchrotron Radiation Facility beamline ID14 and from the European Community Research Infrastructure Action under the Sixth Framework Program (RII3/CT/2004/5060008) for access to the European Molecular Biology Laboratory/Deutsches Elektronen‐Synchrotron. This work was supported by Agence Nationale de la Recherche (ANR‐05‐2_42589), Association pour la Recherche sur le Cancer (ARC, A05/3/3138), Fondation pour la Recherche Médicale, European Research Council, Inserm, the University Cancer Foundation via the Institutional Research Grant program at the University of Texas MD Anderson Cancer Center, by NIH/NCI grant R03 CA169969‐01, and, in part, by the National Institutes of Health through MD Anderson's Cancer Center Support Grant (CA016672). KBC was recipient of fellowships from ARC and Région Ile de France (NeRF). J.A. Girault's group is affiliated with the Ecole des Neurosciences de Paris‐Ile‐de‐France and the Bio‐Psy Laboratory of Excellence.
FundingAgence Nationale de la RechercheANR‐05‐2_42589
- © 2014 The Authors