The F1 loop of the talin head domain acts as a gatekeeper in integrin activation and clustering

Cells spread, adhere and move with the help of integrins. Integrins are allosterically regulated heterodimeric transmembrane receptors forming 24 distinct combinations of α- and β-integrins in mammals (Hynes, 2002; Bachmann et al., 2019). Cytoplasmic adaptor proteins, along with extracellular divalent cations and tensional forces, affect integrin conformation and interactions with extracellular ligands (Sun et al., 2016; Bachmann et al., 2019). One of the key intracellular adaptors is talin, a large, modular protein that binds and activates integrins with its N-terminal head domain, while connecting them to actin fibers via its C-terminal rod domain, which harbors binding sites for further adaptor and signaling molecules required to form cell–matrix adhesion complexes (Sun et al., 2016; Haining et al., 2016). Humans have two talins, talin-1 and talin-2, which have 76% amino acid sequence identity. Of these two talin forms, talin-1 is ubiquitously expressed and is critical for cell adhesion and morphogenesis (Monkley et al., 2000, 2011; Calderwood et al., 2013; Ellis et al., 2013), whereas talin-2 is highly expressed in brain and striated muscle (Senetar and McCann, 2005; Senetar et al., 2007; Debrand et al., 2012). Hereafter, the term talin refers to talin-1.

The talin N-terminal head domain (defined by calpain II cleavage at residues 433–435) contains two ubiquitin-like subdomains (F0 and F1), an acyl-coenzyme A-binding protein-like subdomain (F2) and a phosphotyrosine binding-like integrin-binding subdomain (F3) (Roberts and Critchley, 2009), followed by a calpain II processable linker (residues 406–481) (Rees et al., 1990). The C-terminal rod domain (residues 482–2541) is composed of 13 α-helical bundles that bind actin filaments and adaptor proteins, and a C-terminal helix involved in talin dimerization (Roberts and Critchley, 2009; Gingras et al., 2008; Calderwood et al., 2013). Association with phosphatidylinositol (4,5)-bisphosphate (PIP2)-containing membranes regulates talin activity and integrin binding by competing with auto-inhibitory rod bundle–head domain interactions (Goksoy et al., 2008; Saltel et al., 2009; Goult et al., 2013; Zhang et al., 2016). The talin head domain is a FERM domain by homology (residues 84–405), and talin-1 has been crystallized in a linear arrangement of the subdomains (Elliott et al., 2010; Chinthalapudi et al., 2018). A talin-2 structure adopting a novel cloverleaf-like conformation was presented recently (Rangarajan et al., 2020).

In addition to talin, knockout studies of kindlin-1, -2 and -3 (also known as FERMT1, 2 and 3, respectively), as well as studies of patients with kindlin-3 deficiencies have shown that kindlins are also required for integrin activation and clustering (Montanez et al., 2008; Moser et al., 2008; Qu et al., 2011; Theodosiou et al., 2016). However, in most experimental settings, endogenous kindlin is present in cells, and hence integrins can be activated by talin head overexpression (Tadokoro et al., 2003). Integrins can also be allosterically activated by ligand binding and manganese (Mn²⁺) (Mould et al., 1995; Eng et al., 2011) (Fig. 1A). Upon binding to bent integrins on the cell surface, Mn²⁺ switches integrins to an extended conformation, allowing interaction with immobilized extracellular ligands, leading to integrin clustering in a talin-head-dependent manner (Cluzel et al., 2005; Saltel et al., 2009; Changede et al., 2015; Bachmann et al., 2019). For this talin-head-dependent integrin clustering to occur, an acidic (D723/E726), membrane-proximal motif in β3 integrins interacts with basic residues (K316/K324) in the talin F3 subdomain, while other basic residues of the F2 and F3 subdomains interact with acidic membrane lipids, such as PIP2 (Anthis et al., 2009; Saltel et al., 2009). The talin head subdomains F2 and F3 (referred to collectively as F2–F3) are hence sufficient for activation of αIIbβ3 integrin, as recorded using the PAC-1 antibody in detached cells (Tadokoro et al., 2003), but fail to activate α5β1 integrin, when probed with a fibronectin fragment (Bouaouina et al., 2008). This suggests that, at least in the latter case, the entire talin head domain is needed for the binding of soluble ligand. Importantly, when kindlin-1 and kindlin-2 are completely removed from cells, talin no longer activates integrins and cells do not spread (Theodosiou et al., 2016). In addition, Rap1 (RAP1A and RAP1B in mammals) binding to the F0 and/or F1 subdomains of the talin head has been proposed to recruit talin to the plasma membrane, contributing to integrin activation and Rap1-mediated cell–matrix adhesion (Camp et al., 2018; Katzemich et al., 2019; Gingras et al., 2019). Thus, talin, Rap1 and kindlin play important roles in the integrin activation step, while kindlin also induces clustering of activated integrins (Feigelson et al., 2011; Schmidt et al., 2011; Feng et al., 2012; Ye et al., 2013).

The F1 subdomains of both talins and kindlins contain a long loop (38 residues in talin) that binds to negatively charged membrane lipids (Goult et al., 2010; Bouaouina et al., 2012). The talin F1 loop is important for αIIbβ3- and α5β1-integrin activation, but not for β1 integrin binding to talin (Goult et al., 2010; Bouaouina et al., 2008). Charge inversion of basic residues in the talin F1 loop inhibits cell spreading (Elliott et al., 2010), membrane association of the F1 subdomain and integrin activation (Goult et al., 2010). Nevertheless, the F2–F3 subdomains alone bind efficiently to membranes, and the F0–F1 subdomains do not further enhance this interaction (Moore et al., 2012). In addition, the F1 loop is reportedly flexible and does not form stable contacts within the talin head, and its removal has not been found to affect the structure of the folded subdomains (Elliott et al., 2010). Thus, it is not fully clear how the F1 loop contributes to talin-head-induced integrin activation, since it also acts downstream of the Rap1-mediated mechanism of integrin activation and membrane recruitment (Gingras et al., 2019). Nevertheless, it appears to be a site of important posttranslational modifications, because phosphorylation of the F1 loop at T144 and T150 has been associated with reduced membrane binding (Goult et al., 2010), and also affects talin recruitment to muscle attachment sites in Drosophila (Katzemich et al., 2019).

Talin is, so far, the only FERM family protein that has not been shown to fold into a conserved cloverleaf-like shape, and current models propose that the membrane recruitment of the talin head domain in an F1 loop- and Rap1-dependent manner is the key to integrin activation. Here, we validated the FERM fold of the full-length talin head by small-angle X-ray scattering and used molecular dynamics (MD) analysis to evaluate the functional role of the F1 loop in integrin activation. We show that the F1 loop interacts with the β integrin cytoplasmic tail and interferes with the inner membrane clasp via conserved acidic residues, leading to integrin activation and clustering in a talin- and kindlin-mediated manner. Taken together, our data support a new mechanism for kindlin- and talin-induced integrin activation, in which the FERM-folded talin head tightly interacts with the β integrin tail and projects the F1 loop to act as a gatekeeper to control the talin–integrin association and regulate the activation and clustering of integrins.

The talin head can induce integrin activation (as measured by soluble ligand binding) and integrin clustering (Tadokoro et al., 2003; Goult et al., 2010; Cluzel et al., 2005) (Fig. 1A). However, we found that the integrin-binding F2–F3 fragment of the talin head (t186–435), when probed with a soluble, RGD-containing snake venom (SKI-7; Ballestrem et al., 2001), did not efficiently activate β3 integrin (Fig. 1B,C). Similarly, the talin F2–F3 fragment is reportedly not sufficient to induce α5β1 integrin activation, when measured by soluble fibronectin binding (Bouaouina et al., 2008), whereas it activates αIIbβ3 integrin when probed with the multivalent PAC-1 IgM antibody (Calderwood et al., 2002; Tadokoro et al., 2003; Bouaouina et al., 2008). On the other hand, the talin head F2–F3 fragment induces αvβ3 integrin clustering on vitronectin-coated surfaces in the presence of Mn²⁺ (Saltel et al., 2009). Because integrin activation and clustering are tightly interlinked (Bunch, 2009), the conformational activation of the extracellular domain of integrins by Mn²⁺ may therefore mask the physiological role of the intracellular adaptors in the regulation of the integrin function at the cell surface.

In order to develop a more physiological Mn²⁺-independent test that also takes into account the role of kindlin in integrin activation (Ma et al., 2008; Moser et al., 2009; Harburger et al., 2009), and therefore its potential influence on integrin clustering, we analyzed the clustering of β3 integrin with a C-terminal GFP tag (β3–GFP integrin) in the presence of fluorescently labeled talin head and kindlin-1 in B16F1 melanoma cells spread on serum-coated glass coverslips (Fig. 1B,C,D) (Saltel et al., 2009). The degree of clustering of integrin receptors was calculated from TIRF images, by determining the percentage of the cell area covered with integrin clusters (Fig. 1E). The complete talin head domain (t1–435) enhanced β3–GFP integrin clustering (P<0.001), but the co-expression of kindlin-1 with the talin head and β3–GFP integrin was needed for increasing the integrin clustering further (P<0.001) to a level similar to that observed with the talin head and Mn²⁺. Interestingly, kindlin-1 did not enhance talin F2–F3-mediated clustering, whereas this did occur when Mn²⁺ was used for integrin activation (Saltel et al., 2009) (Fig. 1D,E). Importantly, co-expression of talin head and kindlin-1 was critical for full clustering. Although the exact role of kindlin in the integrin activation process is still unknown (Rognoni et al., 2016), kindlin-mediated integrin clustering requires the presence of the talin F0–F1 subdomains. In addition, the large loop insert in the F1 subdomain is a feature of talins and kindlins, and has been shown to be important for integrin activation by both adaptors (Goult et al., 2010; Bouaouina et al., 2012). Because antibody interference of the talin–integrin association using an F1 loop-directed monoclonal antibody, Ta205 (Xing et al., 2006), suggested proximity of the F1 loop to the integrin-binding F3 subdomain in the talin head, we decided to further investigate the role of the talin F1 loop in kindlin-1-assisted integrin clustering by assaying a series of loop deletions. In addition, we created a model of a FERM-folded talin head domain, containing the F1 loop, to study the biological role of the F1 loop in silico (Fig. 2).

We completely removed the talin F1 loop using two different strategies: by deleting residues 139–168 (del30), as was done for the previously published elongated structures of the talin head (Goult et al., 2010; Elliott et al., 2010), and by replacing residues 134–170 with a Gly-Ala-Gly stretch (del37/GAG) to yield a loop similar in sequence and equal in length to Dictyostelium TalB (Fig. 2G,H; Fig. S1). Both loop deletions led to defects in β3 integrin activation and clustering (Fig. 2F,H,I), but enhanced β3 integrin binding in vitro (Tables S1, S2). We then further narrowed down the deletion to the central part of the loop (del17), and replaced the residues 151–154 (LLRD) with four alanines (151–154AAAA), preserving the length, the previously identified cluster of basic residues (Goult et al., 2010) and the net charge of the loop. Neither of the mutants showed significant integrin activation (Fig. 2F), and the 151–154AAAA mutation alone was sufficient to decrease clustering as compared to the wild-type head domain (t1–435), suggesting that the central stretch of residues 151–154 (LLRD) is important for the functions of the F1 loop. Despite the presence of a full-size F1 loop, the 151–154AAAA mutant displayed decreased in vitro binding to β3 integrin compared to that of the F1 loop deletion mutants, and this was even slightly lower than measured for the wild-type F1 loop (Table S2). These data suggest that the F1 loop interferes with β3 integrin binding to the talin head, to some extent, but at the same time appears to stimulate integrin activation and clustering.

Because all talin head structures published so far lack the F1 loop, we decided to create a F1 loop-containing talin head model based on the compact FERM fold and then explored its structural stability and potential lipid interactions using MD simulations. Compared to the existing extended crystal structure of the talin head [t1–400(del139–168/del30); PBD ID 3IVF] (Elliott et al., 2010), our talin model is rotated at its F1–F2 linker to yield a compact FERM domain (Fig. 2E), similar to the recently published crystallographic structure of kindlin-2 (PDB ID 5XPY) (Li et al., 2017). The F1 domain loop conformation, adopted from a talin F1 subdomain NMR structure (PDB ID 2KC2) (Goult et al., 2010), is clearly different from the conformation of the F1 loop of kindlin-2. Similiar to the F1 loop in our talin model, the kindlin F1 loop is in contact with the F3 and F2 subdomains; however, the kindlin F1 loop is very long (∼125 residues) compared to the talin loop (∼40 residues).

Transmembrane and cytoplasmic domains of αIIbβ3 integrin and a lipid membrane were included in the MD simulation (Table S3). The talin head model remained in a stable FERM domain fold in all simulations except one, in which the F1–F2 interface slightly opened after 0.42 µs of simulation (PIP-A simulation; see Table S3). Generally, the F1, F2 and F3 subdomains remained in the FERM-like triangular assembly, with the F1 and F3 subdomains in contact and the F1 loop in close proximity to the integrin-binding site in the F3 subdomain throughout the 0.5–1 µs simulations. In this compact conformation, the talin head sat on the lipid membrane surface with lipid interaction sites in the F2 and F3 subdomains and in the F1 loop (Fig. S1). Whereas the F1 loop of an isolated F1 domain is highly flexible (Goult et al., 2010), the compact, FERM-folded talin head stabilized the F1 loop conformation, with conserved regions of the loop contributing to interactions with integrin, talin head and lipids, as summarized in Fig. S1.

A salt bridge connects the αIIb and β3 integrin chains as part of the inner membrane clasp that controls integrin activation (Hughes et al., 1996; Anthis and Campbell, 2011). In the MD simulations, the talin F1 loop residue D154 competed with D723 of β3 integrin for binding to R995 of αIIb integrin (Fig. 3A–C), and a D154–R995 salt bridge formed after 3–180 ns (mean±s.d. distance between charged groups 3.5±0.3 Å). As a consequence, the contact between D723 and R995 was lost in all talin–integrin–lipid simulations (charged group distance 6.3±2.1 Å), whereas in control simulations without talin, the D723–R995 contacts were more stable (charged group distance 4.8±0.8 Å).

Based on the MD simulations, we hypothesized that a D154A substitution could block integrin clustering in cells. However, D154A had no significant effect on clustering (Fig. 3E), and closer inspection of the simulation data revealed an additional salt bridge connecting the F1 loop residue E155 with the αIIb integrin residue R997 in three out of four simulations (charged group distance 3.8±0.8 Å, n=3 simulations), suggesting that either of the D154 and E155 residues might interact with αIIb integrin (Fig. 3D). Indeed, although a single E155A mutation had no effect on clustering, replacement of both D154 and E155 with alanine residues was as harmful to integrin clustering as the complete deletion of the F0–F1 subdomains. In contrast, replacement with asparagine and glutamine (D154N, E155Q) preserved wild-type-like integrin clustering (Fig. 3E).

A previous report has shown that the combined charge inversion of adjacent R153E and K156E residues, as well as that of R146E, prevented lipid interaction of the F1 domain, as well as integrin activation in the context of the talin head domain (Goult et al., 2010). We therefore also tested talin head mutants R153A (LLR¹⁵³D-LLAD) and L151A, L152A, D154A (LLR¹⁵³D-AARA), to detect potential interference of integrin clustering. Similarly to the individual D154A mutant, both of these mutants showed normal talin-induced integrin clustering (Fig. S1C), suggesting that membrane binding of the F1 loop is shared by multiple basic residues, as is the potential to activate and cluster integrins by the acidic motif.

Both molecular modeling and integrin clustering assays thus suggested that acidic residues in the F1 loop regulated β3 integrin activation and clustering. To assess whether this regulation takes place via direct interaction of the talin F1 loop with the integrin heterodimer in living cells, we introduced cysteine residues into sites along the inner membrane clasp of β3 integrin and along the mid-part of the F1 loop. We avoided the mutation of residues directly involved in the stabilization of the inner membrane clasp, such as D723 and R995, which might have modified the integrin activation and clustering state independently of kindlin and talin. COS7 cells were transfected with plasmids encoding mouse αv integrin (also known as ITGAV) and human kindlin-1, along with cysteine-mutated talin heads and β3 integrins. To prevent palmitoylation of these newly added cysteines, cells were incubated overnight with bromopalmitate prior to permeabilization with saponin. During saponin permeabilization, cysteine thiol groups were oxidized with copper phenanthroline to promote covalent integrin–talin binding via intermolecular disulfide bonds within integrin–talin complexes persisting in the cells during the permeabilization and oxidation treatment (Fig. 4A). After cell extraction, covalently linked proteins were subsequently identified using non-reduced SDS–PAGE followed by western blotting. Of the five different integrin mutants tested, only the β3(H722C) and β3(I719C) integrin variants yielded covalent heterodimers with the t1–435(L152C) and t1–435(D154C) talin head mutants, representing a restricted subset of the five different F1 loop variants tested (Fig. 4B–E; Fig. S2). In addition, β3(H722C) formed a heterodimer with t1–435(L151C). It is important to note that in the extended talin head conformation (Elliott et al., 2010), the F1 loop and integrin would not reach sufficiently close proximity for disulfide bonding, whereas the FERM-like conformation would enable the observed contacts between the F1 loop and the β3 integrin, and would therefore also explain an interference of the F1 loop with β3 integrin binding to the F3 domain in vitro (Table S2).

In order to obtain further insight into the role and position of the F1 loop in the compact FERM-folded talin structure, we closely analyzed the interface between subdomains F2 and F3 (Fig. 5A). In previously published talin head structures that either did not include the F0 and F1 domains (Anthis et al., 2009) or showed the elongated talin head fold (Elliott et al., 2010), the F2–F3 domains are closely associated and linked together by a salt bridge connecting E269 and K345. However, this salt bridge was not present in our FERM-based model, or in the FERM-folded talin head structure (P.Z., L.A., S.K. et al., unpublished). This led to large solvent-exposed surfaces in the model, which were centered around a solvent-exposed Leu314 in the F3 subdomain and Gly226 in the F2 subdomain, forming potential surfaces for interaction with the F1 loop (Fig. 5A). Indeed, observations of our MD simulations suggested stable hydrophobic and charge–charge interactions of C-terminal F1 loop residues with both the F3 (Fig. 5B) and F2 subdomain surfaces (Fig. 5C).

To investigate whether such F1 loop interactions with the F2 and F3 subdomains would be stable, and whether they contribute significantly to domain–domain interactions within the talin head, we applied two mass spectrometry methods to study the compactness and subdomain interactions of the talin head in solution. Firstly, we characterized the inter-subdomain interactions in the talin head by crosslinking lysine residues using disuccinimidyl suberate (DSS). Monomeric talin head was separated from its oligomeric species by electrophoresis, then trypsinized and analyzed by mass spectrometry. The analysis confirmed that the F1 loop readily interacts with the F1, F2 and F3 subdomains (Fig. 5D; Table S4). However, a further classification of the crosslinks was difficult, because the potential flexibility of the F1 loop may dynamically sample the subdomain surfaces to induce multiple crosslinks from one given loop residue. For example, K149, which in our F1 loop model was solvent exposed, formed crosslinks with both K268 from the F2 subdomain and K320 from the F3 subdomain (Fig. 5D; Table S4). The latter crosslink further suggests that the F1 loop can indeed contact regions of the F3 domain that are otherwise relevant for β integrin binding to talin (Anthis et al., 2009) and integrin clustering (Saltel et al., 2009), indicative of a gatekeeper function for the F1 loop.

Secondly, because the crosslinking approach provided limited information about the specific positions of the F1 loop, we analyzed the compactness of three talin head forms using hydrogen–deuterium exchange mass spectrometry (HDX-MS). The HDX-MS profiles of the talin head variants t1–405, t1–400(del30) and t1–405(del37/GAG) showed major differences in their deuterium intake versus exchange time (Fig. 5E). The native form containing an intact F1 loop (t1–405) clearly showed the lowest HDX rate among the talin forms (up to ∼60% in 2500 s), and had a population of very rapidly exchanging (burst phase) hydrogens, which represented labile side-chain or non-hydrogen-bonded, solvent-exposed backbone hydrogens. In addition, ∼40% of all hydrogens in t1–405 were tightly hydrogen bonded or located in the hydrophobic core and were not exchanged within 1 h of reaction time. The HDX rate was, however, clearly faster for the loop-deleted form t1–405(del37/GAG) and even faster for t1–400(del30), which corresponds to the construct that formed the extended talin head structure during crystallization. Thus, the results suggest that the F1 loop is an integral part of the talin head structure, where it covers parts of the F2 and F3 domains. On the other hand, the C-terminal lysines, absent in the tl–400(del30) construct, may connect the F3 and F1 subdomains, thereby stabilizing the compact structure. Whereas the removal of the F1 loop results in new solvent-exposed talin residues, moderately increasing the HDX rates, the additional C-terminal truncation of the talin head (t1–400) induces an open and flexible structure that is reflected by a further considerable increase in HDX rates (Fig. 5E).

Lastly, we studied the t1–405, t1–405(del30) and t1–405(del37/GAG) talin head variants using small-angle X-ray scattering (Fig. S3). Inline size exclusion chromatography was utilized to ensure monodisperse scattering data derived only from monomers. Kratky analysis indicated that all of the talin forms studied were partially flexible, with t1–405(del37/GAG) being slightly more compact than the other forms (Fig. S3E). To account for the apparent flexibility of talin in solution, we analyzed the talin head solution conformation using the Ensemble Optimization Method (Eom) (Tria et al., 2015). In this analysis, the Eom program generated a pool of talin head domain models from the individual subdomain 3D structures (PDB ID 3IVF; Elliott et al., 2010) with the missing F1 loop and linkers ab initio-modeled as dummy residues. The program then selected an ensemble of conformations that best fitted the scattering data using a genetic algorithm, and this conformational ensemble was represented by a few conformations. The analysis suggested that most of the talin head molecules in the samples were in a FERM-like conformation (Fig. S3). Moreover, the calculated conformational ensembles of all the talin head forms showed narrow R_g distributions, and again the t1–405(del37/GAG) was observed to be more compact than the other talin head forms or a random pool (Fig. S3). The results thus propose that FERM-shaped talin head folds are the major constituent of the talin samples, irrespective of the presence or absence of the F1 loop.

All of our cell biological experiments reported thus far were carried out with the talin head domains in a β3 integrin context. Talin head alone does not become autoinhibited and, hence, is more potent in integrin activation and clustering than full-length talin (Goksoy et al., 2008; Saltel et al., 2009). However, the head domain alone is not properly coupled to the cytoskeleton (Cluzel et al., 2005), because two of the three actin-binding sites in the talin protein are located in the rod domain (Hemmings et al., 1996). To analyze the function of the F1 loop within the complete protein, we expressed wild-type (t1–2541), loop-deleted [t1–2541(del37/GAG)] and point-mutated talins in talin-1^−/− talin-2^−/− mouse kidney fibroblast cells (Theodosiou et al., 2016) and detected the extended and activated forms of β1 integrin by FACS using the conformation-specific 9EG7 monoclonal antibody. Whereas no 9EG7 staining was detected on talin-null cells, the expression of wild-type talin t1–2541 protein rescued the 9EG7 staining to levels comparable to those in talin-floxed cells (data not shown). The loop-deleted talin t1–2541(del37/GAG) only partially activated β1 integrins, comparable to the effects of an R358A talin mutant [t1–2541(R358A)], which reportedly reduces β3 integrin binding (García-Alvarez et al., 2003). The loop deletion was, however, less harmful to integrin activation than the membrane-proximal β integrin binding site mutant t1–2541(K316E,K324E) (Saltel et al., 2009) (Fig. 6A). Furthermore, the previously proposed phosphomimetic F1 loop mutant t1–2541(T144E,T150E) activated β1 integrin less efficiently than wild-type t1–2541, which is also consistent with the reduced β3 integrin binding of a talin head construct of this mutant in vitro (Table S2). In contrast, the phosphorylation-deficient variant t1–2541(T144A,T150A) behaved like wild-type talin in the β1 integrin activation assay (Fig. 6A), and showed kindlin-1-independent integrin clustering in B16F1 cells (Fig. S1D). This suggests that the loop structure and integrin activation function is modulated by phosphorylation at these residues.

Thus, the loop deletion in full-length talin impaired the activation of β1 integrin in suspended cells. Although the ‘inside-out’ activation by loop-deleted talin works poorly in this situation, tension from integrin attachment to a rigid fibronectin-coated surface may still provide ‘outside-in’ activation of integrins, leading to integrin clustering and cell spreading. Indeed, on a rigid and saturated fibronectin substrate, besides a reduction in cell spreading area, talin-knockout cells expressing wild-type or F1 loop-deleted talin (t1–2541) constructs spread comparably (data not shown).

To explore cell spreading in a more natural setting, we used cell-derived 3D fibronectin-containing matrices. After an incubation of 6 h, cells expressing loop-deleted talin had an area of 870±335 µm² (mean±s.d.), whereas cells expressing the intact talin form had spread out to an area of 1359±503 µm² (Fig. 6B,C). Intriguingly, cells expressing wild-type talin formed longer ‘fibrillar-like’ adhesions, whereas in cells transfected with loop-deleted t1–2541(del37/GAG) talin, adhesions were shorter (Fig. 6D). Thus, β1 integrin activation was partially inhibited by the loop deletion, affecting cell spreading in complex 3D matrices and the elongation of adhesions along 3D extracellular matrix fibrils.

Talin plays a key role in integrin-mediated cell–matrix interactions. The function of the F1 loop in the talin head domain has been found important in previous studies, where it was proposed that it either worked through a fly-cast mechanism (Goult et al., 2010) or by continuing the lipid-binding ridge of the F3 and F2 subdomains towards the N-terminal part of an extended form of the talin head domain (Kalli et al., 2013). The reduction of integrin activation and clustering by phosphorylation of F1 loop residues (Fig. 6; Goult et al., 2010) may indeed suggest that differential membrane interaction of the F1 loop controls integrin function. Similarly the loss of kindlin-mediated control of integrin clustering by T144A/T150A substitutions in the F1 loop (Fig. S1D) is consistent with altered muscle attachment site remodeling observed for a T152A mutant in Drosophila (Katzemich et al., 2019), and may also indicate a link between kindlin and the phosphorylation state of the F1 loop. However, independent of the regulation of the talin–membrane interaction, the identification of conserved acidic residues critical for integrin activation and clustering that are localized immediately adjacent to the inner membrane clasp reveals a much more direct way of controlling integrin activation and clustering than by differential membrane recruitment of the talin head.

The crosslinking and SAXS data suggest that the head domain of talin assumes a compact FERM fold, which positions the talin F1 loop next to the integrin-binding site in the F3 subdomain of talin (K316, K324 and L325; see Fig. 5B). This site in talin interacts with the membrane proximal acidic (Saltel et al., 2009) and hydrophobic (Wegener et al., 2007) integrin-activity regulating motif D⁷²³R[K/R]EFAKF to take a central role in the β1- and β3-integrin activation and clustering process regulated by talin and kindlin. Selected acidic residues of the F1 loop appear to target the positively charged juxtamembrane residues in the integrin α subunit thereby interfering with the inner membrane clasp and inhibiting α−β subunit association in order to stabilize the extended conformation of the integrin. The recently reported binding of the Rap1 adaptor to the talin F1 or F0 subdomains (Zhu et al., 2017; Bromberger et al., 2018; Lagarrigue et al., 2018; Gingras et al., 2019) might, in turn, help to stabilize the orientation and membrane tethering of the FERM-folded talin head in order to strategically position the F1 loop and to regulate integrin activation in an F1 loop-dependent manner (Gingras et al., 2019).

While talin and kindlin binding to integrins rearranges the integrin juxtamembrane- and ecto-domains, talin itself is also regulated through conformational changes (Goksoy et al., 2008; Song et al., 2012; Goult et al., 2013; Ye et al., 2016). Our results suggest that additional subtle regulations of the talin–integrin interface may take place within the talin head. The deletion of either the entire F0–F1 subdomains or only the F1 loop enhanced talin head binding to wild-type β3 integrin, whereas phosphomimetic mutations (T144E, T150E) in the F1 loop reduced this binding (Table S2), suggesting that the phosphorylation of the F1 loop can regulate β3–F3 interactions. Accordingly, the phosphomimetic T144E, T150E mutations impaired β1 integrin activation by full-length talin (Fig. 6A) and β3 integrin clustering by the talin head (Fig. S1D). Moreover, the mutagenesis of T144 and T150 (T144A, T150A) enhanced integrin clustering independent of additional kindlin-1 expression. In contrast, both the neutral F1 loop mutant (151–154AAAA) and the mutant with increased net charge (D154A, E155A) clustered integrin poorly (Figs 2F,G and 3E), suggesting that the F1 loop has functions beyond charge interactions with membrane lipids. We observed that the L152 residue of the LLRD151–154 region can stretch and enter into the hydrophobic core of the membrane in all talin–integrin–lipid simulations (Fig. S1A). The LLRD stretch may thus help to guide the D154 or E155 residues to membranes and even to interact directly with integrin at the level of the inner membrane clasp (Fig. 4). The F1 subdomain, and the F1 loop in particular, may therefore function as a gatekeeper in raising the threshold for initial talin–integrin association, while promoting the stability of the unclasped, activated form of the bound integrin heterodimers.

This model is particularly interesting in the light of mechanical tension exerted on the integrin bond. When the talin head induces integrin activation and clustering in a low-force regime, the presence of the F1 loop is required for maintaining the conformational changes of cell-surface-exposed integrins for rapid binding of soluble ligands (inside-out activation). In contrast, when integrins are exposed to mechanical forces, such as those encountered within a rigid 3D matrix scaffold, the integrin catch bond is able to maintain the extracellular domain in a ligand-bound conformation, reducing the requirement for the F1 loop to keep the integrin activated. Therefore, the equilibrium between different integrin conformations at the cell surface, and their functional roles, could be regulated by the F1 loop and the affinity of the talin head for the integrin. Thus, the different types of ECM adhesion and differential remodeling capacity of the extracellular matrix, as recently shown for the β1A- and β1D-integrin splice variants (Soto-Ribeiro et al., 2019), could be explained by different affinities for the talin head domain (Anthis et al., 2010) as well as changes in integrin signaling capacities that potentially affect talin F1 loop residues (Katzemich et al., 2019).

In conclusion, structural and mutational analysis shows that the conformation of the talin head in solution adopts the shape of a FERM domain. In cells, the talin F1 loop directly interacts with residues of the inner membrane clasp of integrins, enabled by the specific projection of the F1 loop from a compact FERM-folded talin head. The FERM configuration of talin provides a scaffold that allows the docking of upstream integrin regulators such as Rap1, in order to position the F1 loop for the talin-dependent regulation of β integrin activation and clustering (Fig. S4). This integrin activation process is assisted by a kindlin-dependent mechanism that leads to integrin clustering, which is required for efficient linkage of integrins to the actin cytoskeleton, and by regulation of integrin signaling, which in turn can be modulated by the phosphorylation state of the F1 loop of talin.

Talin head domains (Table S1) were expressed using the pTrcHis C vector (Thermo Fisher Scientific, Carlsbad, CA), which includes an N-terminal hexahistidine tag (MGGSHHHHHHGMASMTGGQQMGRDLYDDDDKDRWIRPRA). The sequence-confirmed plasmids were introduced into E. coli BL21-Star cells (Thermo Fisher Scientific). Bacteria were cultured in LB medium containing 100 µg/ml of ampicillin at 37°C. IPTG (1 mM) was used to induce protein expression at an optical density of 0.4, and cells were harvested after 5 h of incubation. E. coli extracts were prepared by homogenization (Emulsiflex C3) in 20 mM NaPO₄ pH 7.4, 1 M NaCl and 20 mM imidazole. Clarified lysates were applied to a HisTrap FF 5 ml affinity column (GE Healthcare) using an ÄKTA purifier (GE Healthcare). Talin head domains were eluted from columns using 20 mM sodium phosphate buffer pH 7.4 containing 1 M NaCl with gradually increasing imidazole concentration (20–700 mM). Eluted fractions were analyzed by SDS–PAGE and Coomassie staining and were pooled when appropriate. Furthermore, talin head domains were purified using a HiTrap SP FF 1 ml column (GE Healthcare) after dilution (1:10) in 50 mM NaPO₄, 20 mM NaCl, pH 7.2 buffer. Elution was performed by preparing a linear NaCl gradient with 50 mM NaPO₄, 1 M NaCl pH 7.2 buffer.

GST–β3-integrin-tail fusion constructs (WT-beta3, GST–GSKLLITIHDRKEFAKFEEERARAKWDTANNPLYKEATSTFTNITYRGT; and beta3-VE, GST–GSKLLITIHDRKEFAKFEEERARAKWVENPLYKEATSTFTNITYRGT) were expressed in E. coli BL21-Star cells using a similar approach. Cells were lysed by homogenization (Emulsiflex C3) in phosphate-buffered saline (PBS) pH 7.4 buffer, and proteins were purified with Glutathione Sepharose (GE Healthcare 4 Fast Flow) using buffer containing 50 mM Tris-HCl pH 8 and 20 mM reduced glutathione as eluent. Eluted fractions were analyzed with SDS–PAGE, Coomassie staining, and western blotting using mouse anti-GST B-14 (Santa Cruz) and goat anti-mouse antibodies. Fractions containing the desired proteins were dialyzed into 50 mM sodium phosphate buffer, pH 7.2, containing 150 mM NaCl and stored at −20°C.

The homogeneity of the purified proteins was evaluated using dynamic light scattering (Malvern Zetasizer ZS). The purified talin proteins were also subjected to high-resolution mass spectrometric analysis. The mass spectra indicated that the protein samples were pure and homogeneous, and the determined accurate masses were in a full agreement with those calculated from the sequences.

Talin head molecular weight determination and hydrogen–deuterium exchange mass spectrometry experiments were performed on a Bruker Apex-Qe FT-ICR instrument (Bruker Daltonics, Billerica, MA), equipped with an Apollo-II ESI source and an actively shielded 12-T superconducting magnet. First, all talin samples were buffer exchanged into 200 mM ammonium acetate (pH 7) using Sephadex G-25 columns (PD-10; GE Healthcare Life Sciences). The protein concentrations were determined by absorbance at 280 nm using sequence-derived extinction coefficients. The samples were further diluted with acetonitrile:water (50:50 v/v) containing 1% acetic acid to a final concentration of 1–10 μM. The produced protein ions were accumulated in the collision cell for 100 ms and transferred to the Infinity ICR cell for trapping, broadband (m/z 500–3000) excitation and detection. For each spectrum, a total of 200 time-domain transients (1 MWord) were co-added. All mass spectra were externally calibrated by using ESI-L Tuning Mix calibration mixture (Agilent Technologies). The spectral acquisition and data post-processing were accomplished using Bruker DataAnalysis 4.4 software. Neutral, most abundant protein masses were obtained through a standard charge deconvolution and compared to those calculated from the protein sequences (Table S1).

In hydrogen–deuterium exchange mass spectrometry (HDX-MS) experiments, all talin samples were first buffer-exchanged to 200 mM ammonium acetate (pH 7) by using Sephadex G-25 columns (PD-10; GE Healthcare Life Sciences). The protein concentrations were determined by absorbance at 280 nm using sequence-derived extinction coefficients. The determined protein concentrations varied between 5 and 50 μM. Native HDX-MS measurements were subsequently performed by diluting 40 μl of each protein stock solution with 160 μl of D₂O (99.9% D) to initiate HDX. Therefore, the maximum theoretical deuterium intake was 80%. In order to obtain sufficient spectral intensities for the least concentrated talin samples, greater dilution factors could not be used. After mixing, the samples were immediately delivered to a 250 μl Hamilton glass syringe and continuously infused (1.5 µl/min) to the ESI source using a syringe pump. The ESI-generated ions were accumulated in the collision cell and transferred to the Infinity ICR cell for trapping, broadband (m/z 1000–5000) excitation and detection. All ion source and ion transmission parameters were optimized for native MS experiments. The HDX mass spectra were measured in an automated fashion by 125-s intervals (the first spectrum was acquired after a period of 107.5 s to ensure that the spray current had been stabilized) up to ∼1 h. For each spectrum, a total of 40 time-domain transients were co-added and fast-Fourier transformed prior to magnitude calculation. Each recorded transient had only a 128-kWord data size (equal to the transient length of ∼0.11 s) to increase the time-resolution of the experiments (at the expense of spectral resolution). The final transformed data were downsampled to 32 kWord using Gaussian multiplication to improve peak shapes and allow straightforward peak centroid location, and finally charge deconvoluted. The control experiments (0% D) were performed by diluting the protein samples with H₂O instead; these experiments were performed with two different time-domain data sizes (128 and 1024 kWord), the latter providing isotopically resolved spectra for a more accurate molecular mass determination. All mass spectra were externally calibrated by using ESI-L Tuning Mix calibration mixture (Agilent Technologies). The spectral acquisition and data processing was accomplished using Bruker XMASS 7.0.8 software.

The data analysis for HDX-MS experiments was as follows. The maximum theoretical number of exchangeable hydrogens (D_max) in each protein construct was calculated as: D_max=3n_Arg+2n_Lys+2n_Gln+2n_Asn+n_His+n_Ser+n_Thr+n_Cys(SH)+n_Tyr+n_Asp+n_Glu+n_total+n_charge−n_Pro−1, where n_Xxx is the number of the respective residue in the sequence, n_total is the total number of residues (counting for backbone amide hydrogens) and n_charge is the average number of charges in the observed protein ions. The relative deuterium intake (% D) at each time point was calculated as % D=(m_D−m_H)/(0.80×D_max), where m_D and m_H are the masses of the deuterated and undeuterated protein samples, respectively. For each time point, the HDX reaction time (t_react) was calculated as t_react=(t_start−t_end)/2, in which t_start and t_end are the starting and ending time points of the spectral acquisition. This calculation assumes roughly linear deuterium intake for the given time window. Finally, % D was plotted against t_react, and the data were fitted with a kinetic model of three exponential terms, representing fast, slow and very slow exchanging hydrogens. The fitting was accomplished using OriginPro 15 software (OriginLab Corporation).

A talin head sample (t1–405, 1 mg ml⁻¹ in NaPO₄ pH 7.2 and 150 mM NaCl) was crosslinked with 50× molar excess of disuccinimidyl suberate (DSS) according to the manufacturer's instructions (Thermo Fisher Scientific). The crosslinked talin particles were separated by electrophoresis on a Bolt 4–12% Bis-Tris Plus gel (Thermo Fisher Scientific). The ∼50 kDa bands were excised, and peptides were extracted from the gel slices by in-gel digestion according to methods previously described (O'Connell and Stults, 1997). Cysteine bonds were reduced with 0.045 M dithiothreitol (DTT; D0632 Sigma-Aldrich) for 20 min at 37°C, followed by alkylation with 10 mM iodoacetamide (57670 Fluka, Sigma-Aldrich) at room temperature. The proteins were then digested to peptides using sequencing-grade modified trypsin (V5113, Promega) at 37°C overnight. After quenching with 10% trifluoroacetic acid (TFA), the peptides were desalted by C18 reversed-phase spin columns according to the manufacturer's instructions (Harvard Apparatus). The eluted peptide sample was dried in vacuum centrifuge and reconstituted to a final volume of 30 μl buffer containing 0.1% TFA in 1% acetonitrile (ACN).

Liquid chromatography–mass spectrometry (LC-MS) analysis was performed on an EASY-nLC 1000 system (Thermo Fisher Scientific) connected to a Q-Exactive mass spectrometer (Thermo Fisher Scientific) with nano electrospray ionization sprayer (Thermo Fisher Scientific). In detail, peptides were eluted and separated with a C18 pre-column (Acclaim PepMap 100, 75 μm×2 cm, 3 μm, 100 Å; Thermo Fisher Scientific) and analytical column (Acclaim PepMap RSLC, 50 μm×15 cm, 2 μm, 100 Å; Thermo Fisher Scientific), using a 60 min buffer gradient ranging from 5 to 35% buffer B, followed by a 5 min gradient from 35 to 80% buffer B and 10 min gradient from 80 to 100% buffer B at a flow rate of 300 nl/min (buffer A: 0.1% formic acid in 98% HPLC grade water and 2% ACN; buffer B: 0.1% formic acid in 98% ACN and 2% water). For direct LC-MS analysis, 4 μl peptide sample was injected to the LC-MS/MS and analyzed. Data-dependent FTMS acquisition was in positive ion mode for 80 min. A full scan (200–2000 m/z) was performed with a resolution of 70,000, followed by top10 CID-MS2 ion trap scans with resolution of 17,500. Dynamic exclusion was set for 30 s.

The search engine pLink (v2.3.3; http://pfind.ict.ac.cn/software/pLink) (Lu et al., 2015) was used to identify the intra-protein crosslinked peptides. The precursor mass tolerance was set at 10 ppm and 0.05 Da for the fragment mass tolerance, and the FDR was set to <1% at spectral level. All spectra of putative crosslinked peptides were manually controlled before positive identification. The extended and FERM conformations were compared in terms of inter-residue distances of the identified crosslinks using MNXL analysis (Bullock et al., 2016).

Proteins were analyzed using a liquid chromatography instrument (CBM-20A, Shimadzu Corporation, Kyoto, Japan) equipped with an autosampler (SIL-20A), UV-VIS (SPD-20A) and a fluorescence detector (RF-20Axs). Molecular weight was determined using size-exclusion chromatography with a Malvern Zetasizer µV instrument (Malvern Instruments Ltd, Worcestershire, UK) measuring inline Static Light Scattering (SLS) and Dynamic Light Scattering (DLS). Data were processed using Lab Solution Version 5.51 (Shimadzu Corporation) and OmniSec 4.7 (Malvern Instruments Ltd) software. Samples (50 µg) were injected into the column (Superdex 200 Increase 5/150, GE Healthcare, Uppsala, Sweden) using the autosampler. The column was equilibrated with 50 mM Na₃PO₄ pH 7.2, 150 mM NaCl running buffer. The measurements were run with a flow rate of 0.1 ml/min at 20°C. BSA was used for calibration of the system to calculate molecular weight from the measured SLS intensity.

Talin samples were analyzed using the VP-DSC instrument (MicroCal, Malvern Instruments Ltd) with protein concentrations of ∼0.2 mg/ml in 50 mM NaH₂PO₄ pH 7.2, 150 mM NaCl containing 1 mM EDTA and 1 mM DTT. Concentrations were measured before the analysis by NanoDrop using a theoretical molar extinction coefficient calculated using ProtParam software. Samples were stored at 4°C prior to analysis. Solutions were degassed prior to measurements. Samples were heated from 20°C to 130°C at a scanning rate of 2°C/min. Feedback mode was set to low, and the filter period was 5 s. Temperature transition midpoint (T_m) and calorimetric heat change (ΔH) were obtained by subtracting the baseline and applying the Levenberg–Marquardt non-linear least-squares method to the data using MicroCal Origin 7.0 software (MicroCal, Malvern Instruments Ltd).

Biosensor analysis was carried out on a ForteBio Octet RED384 instrument (Pall Life Sciences) using Ni-NTA sensors following a strategy similar to the one described by Pinon et al. (2014). Samples or buffers were prepared into 96- or 384-well plates at a volume of 200 µl or 80 µl per well, respectively. Sensors were pre-wetted with buffer (50 mM Na₃PO₄ pH 7.2, 150 mM NaCl) in order to obtain baseline measurements prior to protein immobilization. An operating temperature of 27°C and a stirring speed of 1000 rpm were used throughout the experiment. Sensors were chemically activated by immersion in 0.05 M EDC [1-Ethyl-3-(3-dimethylaminopropyl)carbodiimide] and 0.1 M NHS (N-Hydroxysuccinimide) in water for 100 s. Different talin head mutant samples (50 µg/ml) were immobilized on the sensor's surface for 300 s. The excess sulfo-NHS esters were quenched by 1 M ethanolamine pH 8.5 for 100 s. Serially diluted GST–β3-integrin-tail fusion proteins were applied on the talin-coated sensors in concentrations of 20, 80, 320, 1250 and 5000 nM to obtain the relative affinity of talin to GST–β3. Each concentration of GST–β3-integrin-tail fusion protein binding to the sensor was measured for 300 s before moving on to a higher concentration.

Talin–wild-type β3 integrin (WT-β3) and talin–high-affinity β3 integrin (VE-β3) binding data from Octet biosensor (20, 80, 320, 1250, and 5000 nM) were treated as follows: background (GST control) was subtracted, and all the data were normalized to t1–405 at concentration 1250 nM to allow comparison between experiments. The normalized data were analyzed by linear regression analysis in GraphPad Prism 5.02 (GraphPad Software, La Jolla, CA), assuming one-site specific binding. Bmax was set to 8.6 for WT-β3 binding data and to 1.265 for VE-β3 binding data.

Integrin-αIIb (residues 955–1008) and -β3 (residues 684–762) were modeled as described by Orłowski et al. (2015). A FERM-like fold for the talin head domain was generated using the experimental structures of talin F0, F1, F2, and F3 subdomains [PDB IDs 3IVF (Elliott et al., 2010), 2KC2, 2KMA (Goult et al., 2010) and 2H7E (Wegener et al., 2007)]. The F1, F2 and F3 subdomains were arranged into a FERM-like fold using merlin as a template (PDB ID 1H4R; Kang et al., 2002). The model was prepared in an iterative process of homology modeling, manual adjusting of contacts, energy minimizations and short molecular dynamics simulations (up to 23 ns). The loop missing from the crystallographic structure of talin head, residues 139–168, was added to the FERM fold by selecting the least clashing conformer from the NMR structure of talin F1 subdomain (PDB ID 2KC2; conformer 8). Modeling of the integrin dimer has been described previously (Orłowski et al., 2015). MODELLER (Šali and Blundell, 1993) was used for protein modeling. Quality assessment of the talin–integrin complex used for the molecular dynamics simulations with MolProbity (Chen et al., 2010) gave a MolProbity score of 1.71 (89th percentile) and a clash score of 0.61 (99th percentile).

Molecular dynamics simulations were run at 310 K using Gromacs 4.5.5 (Van Der Spoel et al., 2005) and the OPLS force field (Jorgensen et al., 1996) in 0.15 M KCl and TIP3P water using a setup described previously (Orłowski et al., 2015). The simulated systems are described in Table S3. The resulting MD data has been deposited into IDA research data storage and can be accessed via https://ida.fairdata.fi/s/NOT_FOR_PUBLICATION_AfayjpjdWxyL.

Mouse melanoma B16F1 cells (initially obtained from Garth L. Nicolson, Texas M.D. Anderson Cancer Center, Houston, TX) (Ballestrem et al., 1996) and COS7 cells (ATCC CRL-1651) were cultured in DMEM, 10% fetal calf serum (FCS) and antibiotics and passaged by trypsin treatment as described previously (Cluzel et al., 2005). Talin-1^−/− talin-2^−/− cells were obtained from Reinhard Fässler, Max Planck Institute of Biochemistry, Martinsried, Germany (Theodosiou et al., 2016), and cultured in DMEM, containing 10% FCS and antibiotics. At each passage, substrate-adhering cells were detached with trypsin treatment and combined with suspended cells, prior to dilution and resuspension in new culture medium.

Wild-type and mutant human talin1 head domain constructs used for mammalian cell transfection were cloned in pcDNA3 [Invitrogen (Thermo Fischer V79020)] as N-terminal cyan fluorescent protein (CFP) fusion proteins, following standard molecular biology protocols, and confirmed by automated DNA sequencing (Saltel et al., 2009). His-tagged human talin1 head domain constructs were cloned into pTrcHis C vector. GFP-tagged talin1 full-length constructs were cloned into pcDNA3 as an N-terminal tagged human–mouse chimera comprising the human head domain (Saltel et al., 2009) and the mouse rod domain (kindly provided by Anna Huttenlocher, University of Wisconsin–Madison, WI). N-terminal TagRFP-labeled human kindlin-1 (kindly provided by Drs Hongquan Zhang and Staffan Strömblad, Karolinska Institutet, Sweden) was cloned into pcDNA3. Wild-type and VE-mutant forms of the GST–β3-integrin cytoplasmic tail constructs have been described previously (Pinon et al., 2014). C-terminal GFP-tagged mouse β3 integrin constructs were cloned in pcDNA3 and the linker region modified by two additional amino acids, in comparison to previously used constructs (old linker, SPVAT; new linker, DGSPVAT). The DG-linker variant performed identically in β3 integrin spreading assays (Pinon et al., 2014), but showed slightly better clustering behavior in response to co-transfected kindlin-1 constructs (data not shown). A non-tagged mouse αv-integrin construct was amplified by PCR from a mouse placenta cDNA and cloned into pcDNA3 using the following primers (forward, 5′-ATTATGGATCCACCATGGCTGCTCCCGGGCGCCTGCT-3′; reverse: 5′-ATATTAGGGCCCTCAGGTTTCAGAGTTCCTT-3′) as described previously (Wiedle et al., 1999).

Adherent (B16F1, COS7) or non-adherent cells (talin knockout), were transfected in 6-well culture dishes with the JET PEi transfection reagent (Polyplus-transfection) according to the manufacturer’s protocol. For multiple transfections, different vector DNAs were mixed at equimolar concentrations. For adherent cells, the transfection mixture was removed and replaced with fresh medium after 6 h. For talin-knockout cells, the transfection mixture was left on the cells.

After 6 h in transfection mixture, melanoma B16F1 cells were detached using trypsin-EDTA and plated on ethanol-sterilized glass-bottom dishes in FCS-containing culture medium. After 48 h on the FCS-coated glass surface, cells were either fixed directly, or after 20 min exposure to 0.5 mM Mn²⁺ in DMEM containing FCS, with 4% PFA (paraformaldehyde in PBS) for 10 min. Cells were subsequently immerged in PBS for total internal fluorescence microscopy. Integrin clustering analysis was then performed from TIRF images according to methods reported by Saltel et al. (2009). In brief, after background removal, an arbitrary threshold that allowed clustered and membrane-inserted integrins to be distinguished was used to quantify the area exhibiting integrin clusters relative to the total cell surface.

48 h after transfection, mouse melanoma B16F1 cells were detached using non-enzymatic cell-dissociation solution (SIGMA C5788), and talin-knockout cells were collected using trypsin-EDTA, as described above, in culture medium. Cells were then washed twice in PBS containing 0.5 mg/ml BSA and split into two equal populations. Antibody staining or soluble-ligand incubation was performed on ice in PBS containing 0.5 mg/ml BSA. For one cell population, the total amount of cell-surface integrin was revealed by staining with either a hamster anti-β1-integrin antibody (clone HM β1-1; 550530, Pharmingen, BD) or hamster anti-β3-integrin antibody (anti-mouse CD61, clone 2C9.G2; 553344, Pharmingen, BD). The other cell population was stained with a conformation-specific rat monoclonal antibody (rat anti-mouse CD29, clone 9EG7, 553715, Pharmingen, BD) to detect the extended form of the β1 integrin (Lenter et al., 1993), or incubated with an RGD-containing fusion protein of the snake venom disintegrin (kistrin) and the first Ig domain of CD31 (SKI-7), followed by incubation with a rat anti-CD31 monoclonal antibody (GC51) as described previously (Saltel et al., 2009). Subsequent detection of the conformation-specific antibody or the bound β3 integrin ligand SKI-7–GC51 complex were achieved using PE-labeled affinity-purified anti-hamster antibody (127-115-160; Jackson ImmunoResearch Europe), or F(ab′)₂ fragments of goat anti-rat antibodies, respectively and analyzed in an Accuri flow cytometer. Integrin activation was determined by the ratio of the signal between the activation-specific and total integrin-binding antibodies as described in Saltel et al. (2009).

To prepare cell-derived matrix in ethanol-sterilized glass-bottom dishes, the dishes were coated for 60 min at 37°C with 0.2% gelatin in PBS then washed three times in PBS. The gelatin coating was fixed for 30 min in 1% glutaraldehyde, washed in PBS and blocked with 1 M glycine in PBS for 20 min, both at room temperature. After washing in PBS and equilibration in DMEM containing 10% FCS for 30 min at 37°C, 3T3 cells were plated at 50,000 cells/cm² and incubated until they reached confluency (typically, 1 day). The medium – DMEM containing 10% FCS, antibiotics and ascorbic acid at 50 µg/ml – was then changed daily. After 7 days of culture, the wells were washed with PBS and incubated in extraction buffer at 37°C (20 mM NH₄OH containing 0.5% Triton X-100 in PBS) for periods of 2 min. DNA of lysed cells was removed by DNAseI (Amersham, 10 U/ml final) treatment in PBS containing 0.5 mM Mg²⁺ and 1 mM Ca²⁺ at 37°C for 30 min. After gentle washing with PBS containing 0.5 mM Mg²⁺ and 1 mM Ca²⁺, the cell-derived matrix was blocked in DMEM containing 10% FCS for 30 min at 37°C, prior to seeding with talin-knockout cells transfected with GFP–talin constructs. After incubation for 6 h, invaded and spread cells were fixed in 4% PFA in PBS for 10 min and subsequently imaged in PBS by confocal or epifluorescence microscopy.

COS7 cells were transfected in a 6-well plate using JetPEi reagent (1.5 µl; Polyplus-transfection) containing 0.3 µg of mouse αv integrin expression vector, 0.3 µg of mRFP–kindlin-1 vector, 0.3 µg of wild-type or cysteine-mutated β3–GFP integrin constructs (all in pcDNA3) and 0.3 µg of wild-type or cysteine-mutated CFP–talin head constructs, as described above. After 1 d, the culture medium was supplemented with 20 µg/ml 2-bromopalmitate for an additional 17 h, in order to prevent palmitoylation of juxtamembrane cysteine residues. Prior to cell extraction, cells were oxidized for 10 min at 4°C in 400 µl of DMEM medium containing 200 µM CuSO₄, 100 µM phenanthroline and 0.02% saponin. Cells were lysed after gentle removal of the oxidation solution in 300 µl of Tris-buffered saline containing 2% Triton X-100, 0.1% NP40, 5 mM N-ethylmaleimide, 1 mM PMSF and protease inhibitors at 1 µg/ml (chymotrypsin, leupeptine, aprotinin and pepstatin; Sigma) for 10 min at 4°C. 40 µl of the lysate was subsequently run on an SDS–PAGE gel (6%) under non-reducing conditions, then transferred to nitrocellulose and revealed using a mouse monoclonal anti-GFP antibody (clone B34, MMS-118R, LOT/147936001; Covance, Berkeley, CA) according to standard protocols. Analysis of the cysteine-crosslinked samples by 2D-SDS–PAGE (non-reduced/reduced) was performed as follows. The sample to be analysed in 2D was run alongside a pre-stained molecular weight marker on a 6% SDS–PAGE gel under non-reducing conditions. The sample band and part of the marker were then cut from the gel and placed on top of the stacking gel of a 10% SDS–PAGE gel. The gel slice was then overlaid with 1× sample buffer containing reducing agents. SDS–PAGE was stopped for 10 min once the front had just passed the gel slice, in order to reduce the included samples, and subsequently resolved. Transfer to nitrocellulose and anti-GFP detection was then performed as described above.

Small angle X-ray scattering (SAXS) data for talins t1–405, t1–405(del30) and t1–405(del37/GAG) were collected at the P12 beamline at the PETRA III storage ring (EMBL Hamburg/DESY, Hamburg) (Blanchet et al., 2015). Analysis was performed on the monomeric fraction of the sample, as separated by in-line size exclusion chromatography. 100 µl of talin head in a concentration of 4.8–7.6 mg/ml in 50 mM sodium phosphate (pH 7.2) containing 150 mM NaCl was injected onto a Superdex 200 10/300 GL column with a flow rate of 0.25 ml/min. From the column, the sample flowed through the right-angle light-scattering detector at 26°C and through the SAXS sample cell at 10°C. SAXS data were collected on a Pilatus 2 M detector at a 3-m distance from the sample cell, at a wavelength of 1.24 Å, covering the momentum transfer range of 0.025<s<4.8 Å⁻¹. Data were collected at 1-s intervals with an exposure time of 995 ms. The details of the experiment and data analyses are described in Table S5. The SAXS data has been deposited in the Small Angle Scattering Biological Data Bank and can be accessed at https://www.sasbdb.org/project/796/.

Statistical analyses were carried out in GraphPad Prism 5.02 for Windows (GraphPad Software, La Jolla, CA). All clustering data were analyzed using one-way analysis of variance (ANOVA) and a Tukey's multiple comparison test for differences between the groups. The cell spreading data in Fig. 6B were analyzed using a two-tailed, unpaired t-test.

Ritva Romppanen, Ulla Kiiskinen, Niklas Kähkönen, Tatiana Fomekong and Monica Julio-Barreto are acknowledged for excellent technical help, and Michael Bachmann for help in figure preparations. Hongquan Zhang and Staffan Strömblad (Karolinska Institutet, Sweden) kindly provided the kindlin-1 construct. CSC – IT Center for Science Ltd. is acknowledged for computational resources, and Biocenter Finland is acknowledged for infrastructure support.