The role of integrin-linked kinase in the molecular architecture of focal adhesions

Cells adhere to the extracellular matrix (ECM) through integrin-mediated and cytoskeleton-associated structures known as focal adhesions (FAs) (Burridge et al., 1988; Campbell, 2008; Geiger et al., 2009). Beyond their scaffolding function, FAs also serve as sensory structures, which probe the chemical and physical state of the surrounding microenvironment and consequently activate signaling processes that affect multiple cellular features, including survival, proliferation, differentiation and migration (Legate et al., 2009). The structural integrity of FAs, and their scaffolding and sensory functions, depend on large multi-protein complexes linking integrins to F-actin, and regulating both the assembly of the adhesion structures and their signaling activities. Over 180 adhesion-associated proteins have thus far been identified in these adhesion sites (Byron et al., 2011; Schiller et al., 2011; Zaidel-Bar et al., 2007b); yet their precise organization, mode of action, and contributions to the adhesion process are still poorly understood (see: http://www.adhesome.org).

A prominent component of the adhesome is the protein integrin-linked kinase (ILK), a multi-functional molecule that binds directly to the cytoplasmic domains of β1 and β3 integrins, and may associate indirectly with actin through its binding partners, including parvin (Hannigan et al., 1996; Legate et al., 2006; Vakaloglou and Zervas, 2012), paxillin (Nikolopoulos and Turner, 2001) or PINCH (Stanchi et al., 2009). ILK is also reported to play a role in regulating actin and microtubule dynamics and mechanics (Legate et al., 2006; Sayedyahossein et al., 2012; Wickström et al., 2010) by suppressing RhoA-induced actomyosin contractility (Montanez et al., 2009; Aspenström et al., 2004) and stabilizing microtubule tips at nascent FAs (Wickström and Fässler, 2011; Efimov and Kaverina, 2009). Consequently, ILK-deficient fibroblasts have poorly organized actin bundles, associated with attenuated cell spreading. Loss of ILK expression also impairs fibronectin fibrillogenesis (Stanchi et al., 2009; Vouret-Craviari et al., 2004). Knockout of Ilk in mice results in lethality at the preimplantation stage, characterized by adhesion and polarity defects of the epiblast, a failure to form the amniotic cavity (Sakai et al., 2003), and involvement in muscle attachment sites (Zervas et al., 2011). However, despite this information, the molecular mechanisms underlying these structural manifestations of ILK depletion have yet to be resolved.

Recently, the fine architecture of FAs in cultured fibroblasts was visualized using super-resolution fluorescence microscopy (Kanchanawong et al., 2010) and cryo-electron tomography (cryo-ET) (Patla et al., 2010). These findings demonstrated that FAs are laminated structures, composed of a dense bundle of actin filaments, well-aligned with their major axis, which is linked to the ventral cell membrane through a layer of large macro-molecular complexes (termed focal adhesion-related particles; FARPs) (Patla et al., 2010).

The results presented here show that the attenuated spreading of ILK-null cells is associated with a loss of focal complexes and fibrillar adhesions. Furthermore, labeling for multiple adhesome components indicated that Ilk knockout leads to a reduction in vinculin levels and an accumulation of the fibrillar adhesions molecule tensin in FA sites. Cryo-ET of Ilk knockout cells revealed major disarray of the FA-associated actin filaments, and a 1.7-fold increase in the density of FARPs at the membrane–actin interface. Interestingly, FA formation by ILK-null cells required a similarly higher density of the adhesion-mediating c(-RGDfK) peptide, compared with control cells. These results shed light on the key roles of ILK in regulating FA formation and remodeling, and cytoskeletal organization.

To determine the function of ILK in the development and maturation of integrin-mediated adhesions, we compared the spreading and polarization of cultured mouse kidney fibroblasts carrying a floxed (loxP-flanked) ILK gene (Ilk^f/f; expressing normal levels of ILK) to their ILK-deficient counterparts (adenoviral Cre deleted; Ilk^−/−) (Azimifar et al., 2012). When plated on a fibronectin (FN)-coated surface, the Ilk^f/f cells underwent progressive spreading (supplementary material Fig. S1A) and then polarized (Fig. 1A,B), just like wild-type mouse fibroblasts. The Ilk^−/− cells, however, showed marked delays in spreading and remained generally round (Fig. 1C,D, and supplementary material Fig. S1B). Double-fluorescence labeling for vinculin (Fig. 1; red) and actin (Fig. 1; green) indicated that the ILK-deficient cells initially form aberrant FAs, which are almost exclusively located at the cell periphery, associated with poorly organized F-actin bundles, and clearly distinct from the robust FAs and stress fibers found in the Ilk^f/f controls.

In cultured fibroblasts, FA assembly is initiated by the formation of small (typically <1 µm apparent diameter), transient nascent adhesions and focal complexes (FXs) underneath the peripheral lamellipodium (Choi et al., 2008; Geiger et al., 2009; Geiger and Yamada, 2011; Vicente-Manzanares and Horwitz, 2011). A subset of these FXs grows and matures into FAs, which become the primary anchorage sites of contractile stress fibers (Vicente-Manzanares et al., 2009). Careful examination of vinculin immunolabeling in Ilk^f/f cells revealed multiple FXs along the leading edges of the cells (Fig. 1A,B), whereas the Ilk^−/− cells lacked such peripheral structures (Fig. 1C,D). In these cells, FAs were formed under retracted lamellae rather than from FXs, as is commonly observed in control cells. This deficiency in FX formation was apparent both during initial radial spreading (Fig. 1C; 30 minutes) and at later stages, when the cells began to polarize (Fig. 1D and supplementary material Fig. S4; Movies 1, 2, for Ilk^−/− and Ilk^f/f, respectively). The lack of FXs in Ilk^−/− cells probably accounts for the poor alignment of F-actin within the mutant cells. Indeed, whereas the Ilk^f/f control cells developed robust stress fibers (supplementary material Fig. S1A, and Fig. 1A,B), actin bundles in the Ilk^−/− cells were less prominent, and poorly organized (supplementary material Fig. S1B; Fig. 1C,D).

A prominent subset of integrin-mediated adhesions formed by cultured fibroblasts (including Ilk^f/f cells) with extracellular FN fibrils are known as fibrillar adhesions (FBs) (Vicente-Manzanares et al., 2011; Zaidel-Bar et al., 2004; Zamir et al., 1999). These elongated adhesions extend from the edges of FAs, point toward the cell center, and are enriched with tensin, α5β1 integrin, and dephosphorylated paxillin (Pankov et al., 2000; Zaidel-Bar et al., 2007b). However, the causal interplay between FN fibrillogenesis and the formation of distinct FBs remained unclear.

In line with previous reports (Stanchi et al., 2009; Torgler et al., 2004), we show here that knockout of the ILK gene leads to the apparent inhibition of FN fibrillogenesis, as well as to the loss of tensin-rich FBs (Fig. 2A–H). Furthermore, the absence of ILK (and of FBs) was accompanied by pronounced retention of tensin in FAs, and by a marked reduction in vinculin levels. As shown in Fig. 2I, tensin intensity in FAs was elevated by ∼50%, while vinculin intensity decreased by ∼50%. Accumulation of tensin in FAs (which are considerably larger than FBs) also resulted in a 60% increase in the apparent area of tensin-rich adhesions in the ILK-null cells, compared with the Ilk^f/f controls (Fig. 2J). Interestingly, the expression of vinculin was not effected upon Ilk knockout (supplementary material Fig. S3A), whereas the level of tensin expression increased in the ILK-null cells (supplementary material Fig. S3B)

To further explore the functional hierarchical interplay between ILK, tensin and FN in the formation of FBs, we knocked down FN or tensin in a human fibroblast line, WI-38, and determined the effect of this perturbation on the formation of FBs. As shown in Fig. 3, knockdown of FN led to a complete loss of FN fibrils (despite the fact that FN was present in the culture medium), accompanied by the confinement of both ILK and tensin to bona fide FAs (Fig. 3B,E,H). These cells also failed to form FBs, which were present in control cells (transfected with RNA-induced silencing complex-free siRNA; ‘RISC-free’; Fig. 3A,D,G). In contrast, knockdown of tensin did not affect FN fibrillogenesis (Fig. 3C′,F′) or the association of ILK with FBs (Fig. 3C,I). As shown in Fig. 3H′,H, vinculin colocalized with ILK at FAs when FN levels were reduced, but its localization was not affected by tensin knockdown (Fig. 3I′,I). We therefore conclude that ILK-induced assembly of FN fibrils is essential for the formation of FBs, whereas tensin associates with these adhesions, but is not essential for their formation.

Using cryo-ET (Fridman et al., 2012), we previously demonstrated that FA-associated actin is densely packed into bundles, aligned along the major axis of FAs, and linked by short (∼50 nm) interconnecting filaments (Patla et al., 2010). To determine the role of ILK in actin association with FAs, we examined the Ilk^−/− fibroblasts by means of cryo-ET, and compared them to both control Ilk^f/f fibroblasts, and to untreated cells. Correlative fluorescence microscopy and cryo-ET was used to identify FA locations in Ilk^−/− cells expressing YFP–paxillin growing on an FN-coated EM grid. As shown in Fig. 4A, we identified FAs by means of fluorescence microscopy (Fig. 4A, inset), froze the samples, inserted them into the cryo-electron microscope, and identified the same FAs at low magnification. An angular set of images was then collected, and subjected to three-dimensional (3D) image reconstruction. One tomographic slice through the reconstructed volume of the FA, positioned a few nanometers above the cell–matrix interface at the cytoplasmic side of the membrane, is shown in Fig. 3A. A surface-rendered view of the same area, in which all observed actin filaments were tracked, is shown in Fig. 4C. A corresponding tomographic slice and a surface-rendered view of an FA from control Ilk^f/f fibroblasts are shown in Fig. 3B,D, respectively.

The FAs in Ilk^f/f fibroblasts were normal: and similar to those previously reported for REF52 cells (Patla et al., 2010), with an F-actin bundle, consisting of largely parallel filaments, associated with the central region of the adhesion site. The actin cytoskeleton of Ilk^−/− fibroblasts, however, was extremely disorganized, and the number of actin filaments per unit area was considerably reduced (by ∼70%), compared to wild-type or Ilk^f/f cells (Fig. 4A,C, compared to B and D). The reduction in the number of filaments was accompanied by a major loss of filament alignment (Fig. 4E) compared with control cells, where the actin filaments were largely aligned along the FA major axis. Quantification of filament alignment indicated a standard deviation of 15° in the Ilk^f/f cells, compared with ∼25° in the Ilk^−/− cells.

FARPs, indistinguishable from those previously reported for REF52 cells (Patla et al., 2010), were observed at the membrane–cytoskeleton interface of both Ilk^−/− and Ilk^f/f fibroblasts (Fig. 4C,D, green). Quantification of these particles, however, indicated major differences in their distribution density in control and mutant cells. In control cells, the inter-particle spacing (mean distance between the centers of mass of neighboring particles) was 47±29 nm, similar to the inter-particle spacing measured in REF52 cells (Patla et al., 2010), but in Ilk^−/− cells the mean inter-particle spacing was 36±17 nm (Fig. 4F).

Notably, despite the apparent 1.3-fold decrease in average FARP spacing in Ilk^−/− cells, the morphological characteristics of the particles in these mutated cells were highly similar to those found in the WT and Ilk^f/f cells. We analyzed the structure of the adhesion-related particles in two dimensions by extracting the relevant volumes, and projecting them onto a common plane (Fig. 4G). These images were then aligned, and classified according to principle component analysis. Class averages revealed ring-shaped densities with radiating filaments, which presumably connect the FARPs and the actin bundles (Fig. 4G). In both mutant and control cells, three sources constitute the apparent structural variability between sub-classes of FARPs: the particle diameter, the number of ‘whiskers’ extending from the particles, and the angular positions of the whiskers. Particles of both mutated and control cells similarly vary with respect to these three structural features (Fig. 4G and supplementary material Fig. S2). The particles can be classified into three groups of distinct ring diameter, with ratios of 55%, 24% and 21% in Ilk^−/− cells, and 30%, 42% and 28% in Ilk^f/f cells (Fig. 4G, left to right, respectively; supplementary material Fig. S3A).

The properties of the whiskers in Ilk^−/− and Ilk^f/f cells were also similar: FARPs of both cell types normally contained 1–5 whiskers per particle, with mean values of 2.89±1.04 and 2.63±0.98 in Ilk^−/− and Ilk^f/f cells, respectively (supplementary material Fig. S2B). Moreover, the orientations of these whiskers were largely isotropic, although in the Ilk^f/f cells, occasional alignment of whiskers along the FA major axis was seen (supplementary material Fig. S2C). Taken together, these data indicate that the altered actin network in Ilk^−/− cells is accompanied by higher FARP packing densities; however, the FARPs in both types of cells were indistinguishable, suggesting that ILK is not a crucial component of these structures.

To explore the functional significance of the molecular and structural changes induced in FAs by ILK depletion, we tested the adhesion ligand density required for FA formation by Ilk^−/− and control cells. It has previously been shown that cells cultured on nano-patterned surfaces, in which the adhesion ligand (cyclic RGD peptide) is positioned at precise inter-spacings (within the range of 20–200 nm), require a particular threshold density to effectively adhere to the surfaces (Arnold et al., 2008; Cavalcanti-Adam et al., 2007). To compare the ligand density requirements of Ilk^−/− cells and control Ilk^f/f cells, we plated both cell types on nanostructured glass surfaces on which the adhesive peptide ligand [c(-RGDfK); see Materials and Methods] was linked to nanogold particles positioned at fixed inter-spacings. As shown in Fig. 5A,D, Ilk^f/f cells plated on surfaces with 32 nm and 52 nm inter-ligand spacings and incubated for 4 hours, spread well on both surfaces (Fig. 5B,E), whereas Ilk^−/− cells spread exclusively on surfaces with a 32 nm inter-ligand spacing (Fig. 5C), and failed to adhere and spread on the 52 nm pattern (Fig. 5F). These results indicate that Ilk^−/− cells require closer ligand spacing than control cells in order to form functional FAs. Interestingly, these ligand densities are highly similar to the differences in inter-particle spacing found in the mutant and control cells plated on uniform ECMs, and visualized by cryo-ET.

ILK is a canonical component of the integrin cell adhesion machinery. Based on this and previous studies, it appears that ILK participates in three distinct stages of integrin adhesion organization: (1) nucleation of FX under the lamellipodium; (2) organization of mature FAs and their associated stress fibers; and (3) induction of FN fibrillogenesis and, consequently, formation of FBs. Specifically, in the absence of ILK, all forms of integrin adhesions will assemble inefficiently, leading not only to slow cell spreading, but also to limited polarization and poor actin alignment.

FN fibrillogenesis is driven by mechanical forces applied to extracellular FN through integrin adhesions (Klotzsch et al., 2009; Ulmer et al., 2008). It was further demonstrated (Zamir et al., 2000) that this process is accompanied by the formation of a special type of integrin adhesion, namely FBs, which extend centripetally from FAs and are transported towards the cell center. The relationship between FBs and FN fibrils has been well established; yet the functional hierarchy between them remained unclear. Here, we show that in the absence of ILK, FN fibrils do not form. As a consequence, tensin, the hallmark of FBs, is confined to and enriched in FAs, although its knockdown has no apparent effect on FN fibrillogenesis, or on the formation of ILK-rich FBs. These results indicate that ILK is an ‘upstream regulator’, inducing FN fibrillogenesis that, in turn, drives FB formation and the recruitment of molecules such as tensin to these adhesion sites.

The geometry of the ECM has a major effect on the capabilities of cells to adhere.

In previous studies on cell adhesion using nano-patterned surfaces it was shown that the formation of genuine FAs is highly sensitive to adhesion ligand density, and occurs when the ligand-to-ligand spacing is on the order of ∼50 nm or less (Cavalcanti-Adam et al., 2007). Interestingly, the characteristic spacing between adhesion-related particles within FAs, formed on a uniform matrix is, essentially, the same (Patla et al., 2010). Whether this correlation between the apparent densities of the extracellular molecular tethers to the ECM, and the FARPs interconnecting the membrane and the actin cytoskeleton, has some functional significance, was not clear. Nevertheless, our present findings, showing that in the absence of ILK, both the adhesive ligand [c(-RGDfK)] density needed to induce cell adhesion, spreading and FA formation, and the density of FARPs, increase by ∼1.7-fold (from ∼450 to 770 FARPs/µm²), lend further support to the hypothesis that the particles actually bridge between the ECM-bound integrin and the actin network, and that their density is functionally related to the ligand density outside the cell.

The mechanism underlying this ILK-dependent density sensing is still unclear; yet it may be related to the proposed role of ILK in regulating the assembly of adhesome components (Vouret-Craviari et al., 2004; Wang et al., 2008). This is in line with our current results, showing that ILK regulates FARP density, probably without being an essential, intrinsic component of these particles. This conclusion is based on data showing that the FARPs of Ilk^−/− and control cells are essentially indistinguishable when examined by cryo-ET. The regulation of FARP density by ILK may be related to a variety of factors, including an overall alteration in FA molecular composition, such as the increase in tensin levels, or the reduction in vinculin levels shown here (possibly due to the arrest of tensin exit during FB formation), as well as to the direct and specific effect of ILK on cytoskeletal organization at the adhesion site.

Taken together, the recent structural mapping of integrin adhesions at nano-scale resolution (Kanchanawong et al., 2010; Patla et al., 2010), the characterization of adhesome complexity and interactivity (Byron et al., 2012; Geiger and Yamada, 2011; Kuo et al., 2011; Schiller et al., 2011; Zaidel-Bar et al., 2007a), and the identification of distinct ‘molecular signatures’ of subdomains within FAs at micro-scale resolution (Zamir et al., 2008), suggest that integrin adhesions are modular structures, with various molecular complexes displaying distinct distributions and functional roles. Our previous siRNA screen (Winograd-Katz et al., 2009), as well as the findings reported here, demonstrate that specific adhesome components have distinct structural and regulatory roles in the assembly of integrin adhesion structure, and in control of their function. It is conceivable that systematic perturbation of integrin adhesions by gene modulation or different environmental cues, combined with structural analyses such as the one presented here, will become a mainstream approach for characterizing the functional molecular architecture of these sites.

Ilk^f/f and Ilk^−/− fibroblasts, stably expressing similar levels of paxillin–EGFP, were prepared by infecting the mutant cells with a retroviral expression vector encoding paxillin–YFP (kindly provided by E. Danen, Leiden University). The VSV-pseudotyped retroviral vectors used for that purpose were produced as previously described (Sakai et al., 2003; Azimifar et al., 2012), and the concentrated viral supernatant, thus produced, was used to infect Ilk^f/f fibroblasts. Single clones of Ilk^f/f fibroblasts stably expressing paxillin–EGFP were isolated, and subsequently treated with an adenovirus expressing Cre recombinase, in order to delete the ILK gene and to obtain Ilk^−/− paxillin–EGFP-expressing fibroblasts.

Mouse fibroblasts (Ilk^f/f) expressing EGFP–paxillin, cells deficient in the ILK protein (Ilk^−/−), and WI-38 cells from human embryonic lung tissue were plated on glass coverslips coated with FN (as indicated). The cells were cultured in Dulbecco's modified Eagle's medium (DMEM; Biological Industries) supplemented with 10% (v/v) fetal calf serum (FCS; Biological Industries), L-glutamine (0.2 M; Biological Industries), and penicillin–streptomycin–nystatin (Biological Industries), at 37°C and in a 5% CO₂ humidified atmosphere.

Cultured cells were permeabilized, fixed and double-immunolabeled, as previously described (Volberg et al., 2001). Antibodies used included monoclonal antibodies anti-vinculin, rabbit anti-FN and rabbit anti-tensin (produced in the Antibody Production Laboratories, Department of Biological Services, Weizmann Institute of Science), anti-ILK (Santa Cruz Biotechnology, Inc.) and anti-zyxin (kindly provided by Daniel Louvard, Institut Curie, Paris, France). Secondary antibodies used included anti-mouse Cy3 (Jackson ImunoResearch; cat. no. 115-166-072), and anti-rabbit Alexa Fluor 488 (Molecular Probes; cat. no. A11034). To visualize the F-actin cytoskeleton, cells were treated with phalloidin Alexa Fluor 488 (Sigma; cat. no. P1951).

WI-38 cells, 3×10⁵ cells/well, were cultured in six-well plates coated with 5 µg/ml bovine plasma FN, in DMEM supplemented with 10% FCS (Biological Industries) containing sodium pyruvate, non-essential amino acids, and antibiotics (Sigma) for 24 hours at 37°C.

Cells were transfected with siRNA specific for FN (siFN) or tensin (siTNS; SMARTpools; Dharmacon RNAi Technologies).

Carbon-coated 200-mesh gold grids (Quantifoil) were rinsed in PBS (Biological Industries) and overlaid on a drop of 50 µg/ml bovine plasma fibronectin (Biological Industries) for 45 minutes. Cells were typically applied to grids in concentrations of 100 cells per grid, and cultured for 24 hours. Next, specimens were examined by fluorescence microscopy, washed with PBS, and freeze-plunged instantly into liquid-nitrogen-cooled ethane (see below). FAs identified by fluorescence microscopy were indexed according to their precise positions on the grid.

The coordinates of each FA site were recorded and identified under the electron beam. Fluorescence images were acquired with a DeltaVision system (Applied Precision, USA) equipped with a CoolSnap HQ camera (Photometric, USA) operated by SoftWorx and Resolve3D software (Applied Precision), using an Olympus PlanApoN 60×/1.42 NA objective on an Olympus IX70 inverted microscope (Olympus, Japan).

A 5 µl drop of BSA-coated 15 nm gold colloids in PBS was added to the grids before plunging them into liquid-nitrogen-cooled ethane, as previously described (Dubochet et al., 1988). Specimens were then transferred into a 300 FEG Polara microscope (FEI, Eindhoven) equipped with a Gatan post-column GIF 2002 energy filter. Tilt series were then collected, typically covering an angular range of −60° to +66°, sampled in 2° tilt increments, and at a 14 µm underfocus. Pixel size was 0.82 nm at the specimen level.

Projection images (2048×2048 pixels) were aligned to a common origin, using 15 nm-sized fiducial gold particles, and reconstructed by means of weighted back-projection, as implemented by a TOM toolbox software package (Nickell et al., 2005). All tomograms were reconstructed with a binning factor, to yield a 1.64 nm voxel size. For visualization purposes, the reconstructed volumes were processed according to an anisotropic de-noising algorithm (Frangakis and Hegerl, 2001). Individual adhesion complexes were identified and extracted in silico (952 and 946 from mutant and wild-type cells, respectively), from the five best tomograms from each cell line. Two-dimensional images of these elements were calculated by projecting the volumes, 32×32×11 voxels, along the z-axis, using the EM software package. The stack of particles was then masked and normalized prior to PCA, and followed by K-means classification (SPIDER). The most detailed class averages were chosen to be the first references for a multi-reference alignment of the data set. This strategy was iterated for five rounds, until no major changes occurred in the classes, and in the alignment of single images. Surface-rendered visualizations were constructed using AMIRA 5.3 software (Mercury).

Gold surfaces were prepared as previously described (Cavalcanti-Adam et al., 2007).

Glass samples were sputter-coated with a carbon layer, in preparation for imaging with a field-emission scanning electron microscope (LEO-1530, LEO). To visualize the gold nanoparticles on the surface, an acceleration voltage of 3 kV was applied under a pressure of 5×10⁻⁶ mbar.

To passivate the substrate area between the gold nanoparticles, polyethylene glycol (PEG; molecular weight 2000) was used (Blümmel et al., 2007). The gold nanoparticles were then functionalized with the cyclic RGD-based peptide c(-RGDfK-), using the spacer aminohexanoic acid, to mercaptopropionic acid. Next, the PEG-functionalized substrates were immersed for 24 hours in a 25-mM c(- RGDfK-)-thiol/water solution. The substrates were then thoroughly rinsed with MilliQ® water for 24 hours.

Cell samples were sterilized in 70% ethanol for 20 minutes, and washed three times with phosphate-buffered saline at room temperature. Cells were then plated at a density of 40 cells/mm² in DMEM containing 1% FBS and 1% L-glutamine. Experiments were performed three times. Images were obtained after 4 hours and after 24 hours with an Olympus IX inverted microscope.

The authors express gratitude to B. Morgenstern for expert help in preparing this manuscript. B.G. holds the Erwin Neter Professorial Chair in Cell and Tumor Biology. J.S. holds a Weston Visiting Professorship at the Weizmann Institute of Science, Israel.