Finding the weakest link – exploring integrin-mediated mechanical molecular pathways

Mechanical stimuli that are transmitted between a cell and its environment determine cell functions, such as proliferation (Nelson et al., 2005), differentiation (Engler et al., 2006) and motility (Janmey et al., 2009), as well as key processes in development (Gorfinkiel et al., 2009; Mammoto and Ingber, 2010), tumorigenesis (Kumar and Weaver, 2009; Paszek et al., 2005) and wound healing (Fenteany et al., 2000), among others. Mechanical transmission takes place through different cell structures, but one of the main components, and definitely the best studied one, is that of integrin-based cell adhesions. These adhesions connect extracellular matrix (ECM) proteins to the cell cytoskeleton through the transmembrane integrins, and a molecular complex that can contain over a hundred different types of adaptor molecules (Zaidel-Bar et al., 2007a). Adhesions withstand and exert forces on the ECM even though they and the cytoskeleton are dynamic structures. In addition, adhesions are mechanosensitive elements that convert mechanical stimuli into biochemical responses (Bershadsky et al., 2003; Riveline et al., 2001) by sensing and responding to parameters such as matrix rigidity (Moore et al., 2010). Precisely because of their complexity, however, the mechanisms behind the mechanical function of cell–ECM adhesions have been challenging to unravel. The first step is to dissect where biophysics ends and biochemistry starts; that is, we must identify the molecules that exert, withstand, transmit and respond to forces. For those that respond to force (by changing binding kinetics or by unfolding, for instance) we must then evaluate whether and how this change can lead to true force sensing by triggering biochemical signaling cascades. Only after this is achieved will we be able to distinguish the primary molecular events that detect forces from ensuing biochemical signaling networks.

To address this issue, this Commentary will not focus on signaling networks in adhesions, their assembly and dynamics, or their detailed composition, topics which have recently been reviewed elsewhere (Costa and Parsons, 2010; Dubash et al., 2009; Parsons et al., 2010; Ridley, 2011; Wolfenson et al., 2009). Rather, we will follow the path that force follows as it is transmitted from the ECM to the cytoskeleton, analyzing the molecules likely to be under mechanical stress. We will first briefly review the molecular machinery that is responsible for force generation and the biophysical effects that force has on molecular conformation and binding. We will then analyze the ECM molecules, integrins and adaptor molecules step-by-step. For all the components of this protein network, we will review current knowledge with regard to how they respond to force and the possible links through which they could participate in force transmission throughout the cell. We will then discuss the factors that affect the collective mechanical behavior of all these molecules that are present in adhesions. Finally, we will discuss the signaling events that could be triggered by putative mechanosensors. Owing to space limitations, we will not cover adhesions that are mediated by transmembrane proteins other than integrins (such as syndecans) or cell–cell adhesions.

Cellular forces are primarily developed by the acto-myosin cytoskeleton (Box 1) and have a profound effect on molecular conformation and binding (Box 2). Starting at the cell exterior, the first elements that forces encounter in their path are ECM molecules, which form meshworks of polymerized fibers that constitute the major components of the cell microenvironment. Among ECM molecules, fibronectin has been the most studied from a mechanical point of view. Forces of 80–200 pN can stretch all the fibronectin FNIII domains, which contain the integrin-binding sites (Table 1). Upon stretching, fibronectin fibers can extend up to eightfold before breaking (Klotzsch et al., 2009), which leads to unfolding of the FNIII domains and exposure of cryptic sites. This unfolding, however, might account for only a fraction of the fibronectin stretch, which could be mostly determined by a global conformational change (Lemmon et al., 2011). In any case, stretching of FNIII domains is a form of mechanotransduction, because the exposure of the cryptic sites leads to downstream effects, such as the regulation of the fibrillogenesis of fibronectin (Sechler et al., 2001), stimulation of cell growth and contractility (Hocking and Kowalski, 2002), and stimulation of the digestion of gelatin, type IV collagen, α- and β-casein and insulin β-chain (Schnepel and Tschesche, 2000). Force can also change the conformation and distance between the RGD and PHSRN motifs (two synergistic integrin-binding sites within FNIII domains), altering integrin recognition and specificity (Grant et al., 1997; Vogel, 2006). This stretching and unfolding can be achieved in vivo by cell-generated forces (Baneyx et al., 2001; Smith et al., 2007) and is controlled by cell contractility (Baneyx et al., 2002).

Collagen I fibrils can also be stretched with forces of 1–4 nN, which extrapolates to a force of 4–90 pN per collagen monomer (Graham et al., 2004). The role of force in collagen function is unclear, but many of the key binding sites in collagen, including the GFOGER sequence that can bind to β1 integrin, are cryptic and are only fully exposed upon proteolysis, partial denaturation or force application (Orgel et al., 2011; Perumal et al., 2008). Interestingly, collagens also possess several copies of the RGD sequence (Davis, 1992), which has different integrin specificity and is characteristic of other ECM molecules such as fibronectin (Hynes, 2002). In collagen, the RGD site is only exposed upon denaturation (which can be caused by force), which could change integrin specificity and trigger integrin-mediated wound response (Davis, 1992). The coiled-coil sequences in the fibrinogen α-helices also unfold under forces of ∼100 pN (Table 1), and this molecular unfolding seems to be the main factor determining the elasticity of fibrin fibers in blood clots (Brown et al., 2009). Fibrinogen unfolding can also activate its integrin receptor, macrophage-1 antigen (MAC1 or integrin αMβ2) and promote inflammation, suggesting that there is a possible mechanosensing mechanism (Deng et al., 2011). The relevance of mechanical stimuli for the function of other ECM molecules, including laminins and vitronectin is essentially unexplored. However, the list of cryptic binding sites in these molecules that could be exposed by force is extensive (Davis et al., 2000), opening many intriguing possibilities.

Integrins are heterodimeric transmembrane proteins composed of an α- and β-subunit, and are major molecular links between cells and the ECM. Different combinations of subunits lead to 24 different types of integrins in humans and determine their specificity for ECM molecules (Hynes, 2002). Whereas most of the integrin molecule is located extracellularly, to bind to the ECM, a short cytoplasmic tail binds to adaptor proteins, as discussed below. Although force does not unfold integrins, it can affect them through other mechanisms. First, applied forces can disrupt integrin–ECM bonds. Such rupture forces have been studied extensively for the bonds between fibronectin and α5β1 integrin, and fibrinogen and αIIBβ3 integrin (see Table 1). Given the variety of experimental approaches used, the narrow range of reported unbinding forces (50–100 pN) is remarkable. Integrin–collagen interactions have been far less studied, but the binding strength between the GFOGER binding site in collagen and α2β1 integrin has been reported to be 160 pN (Niland et al., 2011), suggesting that they have a higher resistance to force than integrin–fibronectin adhesions. When rupture forces are measured in live cells containing many integrin–ECM bonds, the calculated force per bond is much lower (below 10 or even 1 pN) than that obtained from single-molecule measurements (Friedland et al., 2009; Roca-Cusachs et al., 2009; Walter et al., 2006). This suggests that in integrin clusters, only a small fraction of molecules withstand forces (Moore et al., 2010).

The bond between fibronectin and α5β1 integrin shows catch bond behavior (i.e. a bond that strengthens when force is applied; see Box 2) in which maximum lifetimes occur at forces of 10–30 pN (Kong et al., 2009). Such catch bond behavior has not been observed for αIIβ3 (Litvinov et al., 2011) or αvβ3 integrin. This could explain the much higher resistance to force of fibronectin–α5β1-integrin bonds compared with those of fibronectin–αvβ3-integrin bonds (Roca-Cusachs et al., 2009) although the latter are required for force sensing. Thus, although stable α5β1 catch bonds can regulate adhesion, more fragile αvβ3 integrin bonds might constitute mechanosensors that undergo frequent unbinding and rebinding events under force, triggering downstream events (Arias-Salgado et al., 2005; von Wichert et al., 2003). In parallel to possible catch bond behaviors, integrins also undergo conformational changes (activation) that prime them for binding to the ECM (Luo et al., 2007; Moser et al., 2009). Activation can also change their affinity for cytoplasmic partners such as α-actinin (Sampath et al., 1998) and catalyze downstream events including tyrosine phosphorylation of different substrates (Legate et al., 2009). Structurally, integrin activation involves separation of all domains except the ECM-binding head, and can be triggered by either ECM binding itself (so-called ‘outside-in activation’) (Kim et al., 2003; Takagi et al., 2002) or binding to its cytoplasmic partner talin (Bouaouina et al., 2008; Calderwood et al., 1999), possibly in combination with kindlin (Moser et al., 2008) (so called ‘inside-out activation’). Integrin activation can also be triggered by force, as shown for α5β1 catch bonds (Friedland et al., 2009) and suggested by results of steered molecular dynamics simulations (Puklin-Faucher et al., 2006). Because integrin-binding partners, such as talin, filamin and α-actinin, bind to the cytoplasmic integrin-β tails (but not α-tails) (Otey et al., 1990; Pfaff et al., 1998), force application from the cytoskeleton through these partner molecules could induce the integrin subunit separation necessary for activation (Puklin-Faucher and Sheetz, 2009). Thus, force-induced integrin activation can also be considered to be a mechanotransduction event.

Integrins, which lack actin-binding domains, are indirectly connected to the actin cytoskeleton through different protein complexes (Fig. 1) that form different types of cell–ECM adhesions, including nascent adhesions, focal complexes, focal adhesions, podosomes and others (Dubash et al., 2009). These complexes can contain over a hundred different types of proteins (Zaidel-Bar et al., 2007a), with many binding in a force-dependent manner (Pasapera et al., 2010; Schiller et al., 2011). Here, we will focus on the path that force must follow from integrins to actin and its effect on the molecules involved.

The connection between integrin and actin can be mediated by a single molecule, such as talin. Given that the N-terminal head domain of talin binds to and colocalizes with integrin-β tails, and the C-terminal domain binds to and colocalizes with actin (Critchley, 2009; Kanchanawong et al., 2010), talin is ideally positioned in the force path between integrins and actin. Talin-depleted fibroblasts initially spread normally, but then retract and round up instead of developing mature adhesions (Zhang et al., 2008). This behavior is rescued by full-length talin but not by the integrin-binding and -activating talin head alone, suggesting that the force transmitted between integrins and actin by full-length talin has a key role. Indeed, forces of 12 pN can expose cryptic binding sites in talin molecules and cause vinculin binding (del Rio et al., 2009), which might explain why talin is required for adhesion reinforcement and maturation (Jiang et al., 2003; Roca-Cusachs et al., 2009). In cells, talin can stretch cyclically from its folded length of 50 nm to over 400 nm in a vinculin-dependent manner, and the stretching cycles correlate with the lifetime of vinculin in adhesions (Margadant et al., 2011). Thus, vinculin, which has been reported to withstand forces of up to 2.5 pN (Grashoff et al., 2010), might reinforce the mechanical integrin–actin connection by simultaneously binding talin and actin (Burridge and Mangeat, 1984; Menkel et al., 1994). However, the tension level in vinculin molecules has not been directly related to its recruitment to adhesions, and vinculin-null cells have been reported to have both impaired (Diez et al., 2011) and normal (Grashoff et al., 2010) focal adhesions and contractile forces. Therefore, the exact role of vinculin in force transmission and mechanotransduction remains unclear.

Several other molecules can also simultaneously bind to actin and integrins, including α-actinin (Pavalko et al., 1991; Pavalko and LaRoche, 1993), filamin (Hartwig and Stossel, 1975; Loo et al., 1998; Sharma et al., 1995) and tensin (Calderwood et al., 2003; Lo et al., 1994). α-Actinin is an antiparallel dimer and a key actin crosslinker (Sjöblom et al., 2008) that is involved in adhesion maturation (Choi et al., 2008). The crosslinking role of α-actinin is important in adhesions, because normal adhesion elongation is recovered in α-actinin-depleted cells when actin crosslinking is restored through use of a myosin mutant that constitutively binds actin (Choi et al., 2008). However, the transmission of forces between integrins and actin by α-actinin could also have a relevant role. Indeed, inactivating α-actinin in stress fibers leads to their detachment from focal adhesions (Rajfur et al., 2002), and blocking force transmission with exogenous α-actinin fragments that contain only the actin- or the integrin-binding domain, respectively, can lead to the disassembly of stress fibers and focal adhesions (Pavalko and Burridge, 1991; Triplett and Pavalko, 2006). α-Actinin has cryptic sites for binding to the smooth muscle protein smitin within its spectrin-repeat domains (Chi et al., 2005), although these domains might be too stiff to be stretched physiologically (Table 1). Filamin is composed of an actin-binding domain and 24 immunoglobulin-like domains (Popowicz et al., 2006), and also forms dimers that can orthogonally crosslink actin, thus forming three-dimensional meshworks (Flanagan et al., 2001; Hartwig and Shevlin, 1986). This crosslinking activity is crucial for motility, actin architecture, force generation and rigidity sensing (Byfield et al., 2009; Cunningham et al., 1992; Flanagan et al., 2001; Lynch et al., 2011). However, the binding of filamin to integrins is also important, as filamin competes with talin for integrin binding (Kiema et al., 2006), thereby regulating talin-mediated adhesion (Nieves et al., 2010).

Forces of 40–70 pN can both detach filamin from actin and unfold the immunoglobulin-like domains of filamin (Table 1), which contain binding sites to multiple partners. This suggests that different force-loading rates and directions of force application can lead to either unfolding or unbinding, or to unfolding followed by unbinding as proposed for the actin-associated cytoskeletal proteins spectrin and ankyrin in red blood cells (Krieger et al., 2011). Shear strain applied to actin networks crosslinked with filamin increases integrin binding (Ehrlicher et al., 2011), probably by unfolding the 21st immunoglobulin-like domain and exposing a binding site for integrin-β tails (Chen et al., 2009; Pentikäinen and Ylänne, 2009). However, strain also reduces the binding of filamin-A-associated RhoGAP (FILGAP, also known as Rho GTPase-activating protein 24) (Ehrlicher et al., 2011), possibly by separating the two FILGAP-binding domains present in the filamin dimer. This, therefore, is an elegant mechanism by which a single mechanical stimulus could regulate the binding of two proteins in opposite ways. Finally, tensin also forms dimers that have multiple actin-binding sites and a domain that binds to the NPXY motifs on integrin-β tails (Lo, 2004). In fibroblasts, colocalization of tensin with integrins is much more prominent in mature fibrillar adhesions than in early focal contacts (Zamir et al., 2000), pointing towards a role in adhesion maturation. Tensin localization also requires talin and integrin-linked kinase (ILK) in Drosophila (Torgler et al., 2004), suggesting that there are interactions between the three molecules. Expression of a tensin fragment containing the actin homology 2 region can inhibit adhesion formation and the normal rearward translocation of α5β1 integrins (Pankov et al., 2000), indicating that tensin normally transmits force from the cytoskeleton to integrins and is necessary for adhesion maturation. Such a mechanical function is further confirmed by the fact that the defects observed in tensin-null Drosophila (i.e. wing blisters) only appear after mechanical stimulation (Torgler et al., 2004). However, no information is currently available on the mechanical properties of tensin or its molecular interactions.

Links formed by more than one molecule can also connect integrins to actin. For example, ILK binds to integrin-β tails and connects to actin through an interaction with parvin, forming the ILK–pinch–parvin (IPP) complex (Wickström et al., 2010). This complex regulates cell contractility (Wickström et al., 2010) and mechanosensation in touch neurons (Hobert et al., 1999). Kindlin participates in integrin activation (Moser et al., 2008) and could connect to filamin through migfilin (Tu et al., 2003). Given the high degree of complexity in the protein interaction networks in focal adhesions (Zaidel-Bar et al., 2007a), several additional paths connecting actin to integrins are possible. Force transmitted through these paths could be sensed by mechanosensory proteins, because several proteins change enzyme activity or become phosphorylated in response to force (Moore et al., 2010). Several proteins are also recruited by myosin II contractility, including FAK, zyxin, vinculin, α-actinin and LIM proteins such as pinch, zyxin, paxillin and migfilin (Pasapera et al., 2010; Schiller et al., 2011). However, which of the molecules affected by force are true mechanosensors (i.e. they respond directly to forces) and which simply act downstream, remains unclear. For instance, LIM proteins are likely to bind to exposed sites in mechanonsensing proteins but are not mechanosensors themselves (Schiller et al., 2011). Some insight into this question has been obtained by stretching of membrane-stripped cytoskeletons, which should eliminate many downstream biochemical pathways present in live cells. Cytoskeletal stretching leads to increased cytoskeletal binding of paxillin, focal adhesion kinase (FAK), p130Cas (also known as BCAR1), protein kinase B (Akt), Rap guanine nucleotide exchange factor 1 (known as C3G, RAPGEF1 and GRF2) and Crk isoform II (Sawada and Sheetz, 2002; Tamada et al., 2004). Despite these observations, direct evidence for mechanosensing in individual molecules is scarce. Mechanical stretching of the substrate domain of p130Cas increases its phosphorylation by Src family kinases in vitro, enabling the binding of several proteins (Sawada et al., 2006). p130Cas is thus one of the very few demonstrated mechanosensors, although the molecular links through which it would be stretched in cells are unclear. Force application increases the tyrosine kinase activity of FAK (Tomar and Schlaepfer, 2009), possibly by disrupting an inhibitory interaction between its FERM and kinase domains (Cooper et al., 2003). FAK connects indirectly to integrins and actin, through proteins such as talin (Chen et al., 1995), paxillin (Hildebrand et al., 1995), or Arp3 (Serrels et al., 2007). In summary, integrins can be connected to actin through one or more molecules, but, with the exception of talin, it remains unclear which of them experience and respond to forces in vivo.

The chain of forces goes from matrix to integrins to adaptor proteins bound to actin, and the actin filament network transmits forces exerted by myosin motors. Myosin I detachment from actin filaments is decreased by up to 75-fold under forces greater than 2 pN, in a mechanism that could contribute to mechanosensing in hair cells (Laakso et al., 2008). Similarly, the myosin-II–actin interaction behaves like a catch bond, with an optimal load of 6 pN (Guo and Guilford, 2006). This is also the maximum force that the motor can exert, indicating that the motor is designed to hold for the longest time at maximal force. External forces might, thus, optimize myosin II binding and promote further force generation by the cell. Additionally, myosin light chain kinase (MLCK) binds both myosin and actin (Hatch et al., 2001; Sellers and Pato, 1984; Yang et al., 2006; Ye et al., 1997) and moves rearward with actin during cell spreading (Giannone et al., 2004). MLCK could therefore be stretched between myosin and rearward flowing actin, resulting in the release of MLCK auto-inhibition and in myosin II activation. Mechanoactivation has indeed already been observed for another member of the MLCK family, the kinase titin, which is unfolded by 10 nm under forces of 30 pN (Gautel, 2011; Puchner et al., 2008).

Integrins and the cytoskeleton can also transmit forces without involving an integrin–actin link. For instance, increases in membrane tension that result from actin polymerization can activate contractility and exocytosis (Gauthier et al., 2011), or regulate the activity of stretch-sensitive ion channels (Sbrana et al., 2008). Mechanosensitive ion channels also associate with β1 integrins (Chao et al., 2011), and might be activated by forces transmitted through them (Matthews et al., 2010). Force activation of these channels leads to adhesion reinforcement, and affects migration, contractility and recruitment of vinculin (Matthews et al., 2006; Munevar et al., 2004). Besides ion channels, membrane tension (or ECM deformation) might regulate integrin spacing, thereby stretching integrin crosslinking proteins such as receptor-type tyrosine-protein phosphatase alpha (RPTP-α), which can expose a catalytic cleft leading to the activation of Src family kinases (Jiang et al., 2000; von Wichert et al., 2003).

Besides actin, intermediate filaments and microtubules also have a role in force transmission, which is much less explored. The intermediate filament protein vimentin localizes to adhesion sites, and its depletion reduces cell contractility, adhesion size and migration (Eckes et al., 1998; Tsuruta and Jones, 2003). Furthermore, vimentin regulates cell extensions through its interaction with filamin (Kim et al., 2010), depolymerizes upon myosin II inhibition (Johnson et al., 2007), colocalizes and co-precipitates with α2β1 integrins (Kreis et al., 2005), and is required for proper β1 integrin localization at adhesions (Ivaska et al., 2005). Microtubules are also targeted to early adhesion sites (Kaverina et al., 1998), although their depolymerization leads to formation of stress fibers and focal adhesions (Kaverina et al., 1998; Liu et al., 1998), and increased contractility (Stamenović et al., 2002). This result has led to the hypothesis that microtubules normally carry contractile compressive stresses, but it could also be explained by there being increased myosin contractility upon microtubule depolymerization (Kolodney and Elson, 1995). Thus, although cellular forces are mostly generated by actomyosin contractility (Cai et al., 2006), microtubules and intermediate filaments could transmit mechanical forces in special cases.

In summary, forces affect a complex series of molecular links that connect the ECM to the actin cytoskeleton. These links can operate in parallel, given that many different proteins can simultaneously support forces (Fig. 2A). The resulting collective behavior naturally depends on the rate of force loading and cannot directly be predicted from individual bond properties. At low loading rates, for instance, when a bond breaks, the force redistributes to the remaining bonds, leading to a faster detachment. The average detachment force per bond can then be smaller in a cluster than in an individual bond (Seifert, 2000). At faster loading rates, rupture is too fast and this effect is not important (Sulchek et al., 2005). However, in live cells, maximum force per bond increases sixfold in pentameric versus monomeric fibronectin–α5β1 bonds (Roca-Cusachs et al., 2009), indicating that, beyond physical factors, protein complexes in integrin clusters stabilize mechanical links. Indeed, integrin cluster formation involves integrin interactions with kindlin, talin and phosphatidylinositol (4,5)-bisphosphate [PtdIns(4,5)P₂] (Moser et al., 2009; Saltel et al., 2009), as well as other integrins (Li et al., 2003b). These interactions require a spacing that is smaller than 60 nm because larger distances between integrins in nanopatterned substrates do not sustain cell adhesion (Arnold et al., 2008; Jiang et al., 2003; Schvartzman et al., 2011).

Once formed, actin filaments grow from clusters, and contractile forces between them bring them together (Rossier et al., 2010; Yu et al., 2011; Ghassemi et al., 2012) (Fig. 2B). This requires a link between adhesion sites and polymerizing actin that could be provided by formins. Formins are found on dorsal stress fibers that emanate from focal adhesions and regulate the distribution of focal adhesions at the lamellum–lamellipodium interface, as well as focal adhesion turnover (Gupton et al., 2007; Hotulainen and Lappalainen, 2006). Formins elongate filament barbed ends at rates of up to 27 nm/second for the formin mDia2 or 120 nm/second for mDia1, as measured in vitro (Chesarone and Goode, 2009; Kovar et al., 2006). These rates are well within the range of rearward flows of actin in vivo (see below). Supporting this hypothesis, formin-mediated polymerization of single actin filaments has been measured to exert forces greater than 1 pN (Kovar and Pollard, 2004). Thus, actin polymerization from clustered integrins could generate forces on adhesions, but details of the interactions need to be worked out. What is clear, however, is that the mechanical behavior of adhesion molecules depends not only on their individual properties but also on their density and spacing in clusters, their interactions and the force loading rate.

Although contractile forces pull initial integrin clusters towards each other, once clusters grow and mature, force is applied mainly by actin flowing rearward from the leading edge to the center of cells (Fig. 2C). External stimuli can also apply forces through the ECM, which can be transmitted to distant points across a cell (Hu et al., 2003), activating factors such as Src (Na et al., 2008) or Rac (Poh et al., 2009). However, endogenous rearward flow appears to regulate adhesion maturation and traction force generation. This flow is mainly driven by myosin II contractility because inhibition of myosin decreases the velocity of actin flow by over two thirds (Cai et al., 2006) and leads to unbalanced actin polymerization forces that dramatically increase membrane tension (Gauthier et al., 2011).

In a variety of cell types and depending on the location within the cell, rearward actin flow is in the range of 2 to 600 nm/second (Fisher et al., 1988; Forscher and Smith, 1988; Gardel et al., 2008; Theriot and Mitchison, 1991). Generally, higher flows are associated with weak or nascent adhesions and low traction forces on the substrate, and vice versa (Chan and Odde, 2008; Kress et al., 2007; Zhang et al., 2008). This suggests that high actin speeds lead to high force loading rates that decrease bond lifetimes (Box 2) and therefore the ability to transmit force from actin to adaptor proteins and integrins (Li et al., 2010). This effect of loading rate would no longer be relevant at speeds below 10 nm/second, at which point the relationship between rearward flow and traction force becomes direct rather than inverse (Gardel et al., 2008). Given the transient nature of the mechanical links, actin flow is coupled best to actin-binding proteins, such as talin and particularly α-actinin, which also move rearward at a speed of ∼5 nm/second, less efficiently to FAK, paxillin or vinculin, and even less efficiently (∼1.3 nm/second) to substrate-bound αv integrins (Hu et al., 2007). The efficiency of this transmission also depends on the ligand density in the ECM (Brown et al., 2006), showing that molecular clustering is a determinant of this process. Thus, in mature adhesions, transient stick–slip bonds seem to couple integrins to adaptor proteins and to the actin cytoskeleton (Margadant et al., 2011).

The rearward speeds of proteins, multiplied by their stiffness (i.e. the ratio between applied force and resulting molecular extension), will determine force-loading rates and thus bond lifetimes (Box 2). Protein stiffness increases as molecules are extended, often following a worm-like chain behavior (Marko and Siggia, 1995), but constant stiffness values can be approximated for small extensions (Table 1). In a serial connection between an ECM molecule, an integrin and an adaptor protein, the total stiffness would be dominated by the softer elements (Fig. 2A). For instance, if fibronectin (0.5 pN/nm) is linked to talin (1.5 pN/nm), both elements are soft and would probably extend and possibly unfold, enabling mechanosensing events. If fibronectin is linked to α-actinin (1700 pN/nm), however, only fibronectin would extend, with α-actinin simply transmitting the force. This difference could explain the reported mechanosensitive unfolding of talin (del Rio et al., 2009) but not of α-actinin. Indeed, talin molecules in adhesions continually extend and relax in concert with the concomitant binding and release of vinculin (Margadant et al., 2011). However, the timescales of talin stretch–relax cycles (6–16 seconds) do not match those of traction forces (30–90 seconds) (Ghassemi et al., 2012). This suggests that actin flow might transmit forces through stiff molecules such as α-actinin, while signaling by repeatedly stretching talin, then unbinding from it, and finally rebinding again. At a loading rate of 2.5 pN/second (0.5 pN/nm and retraction speed of 5 nm/second), for instance, a stretch cycle of 16 seconds would build a force of 40 pN, sufficient to unfold talin and trigger mechanotransduction (Table 1). Given the high retraction speeds used in talin unfolding experiments (400 nm/second, Table 1), in vivo unfolding forces would however be much lower. Thus, it remains unclear whether unfolding or unbinding from actin (measured at 2 pN at 60 nm/second, Table 1) would dominate. The interaction between these two events would define the signaling effectiveness of talin stick–slip cycles. This concept of stick–slip stretching of molecules as a primary force signaling mechanism has yet to be tested in various systems, but it could explain how the cell ‘samples’ adhesions over time and generates integrated signals to the nucleus or other parts of the cell.

Downstream signaling from molecular mechanosensing events is mediated by phosphorylation, dephosphorylation, small GTPase signaling and phospholipid signaling, among others (Fig. 3). Whereas the ensuing signaling networks are vast, we will discuss here examples of known signaling cascades that emanate from the specific mechanosensing events discussed above.

The mechanically activated receptor-type tyrosine-protein phosphatase alpha (RPTP-α, encoded by PTPRA) transmits the mechanical signal to Src family kinases (SFKs), which are activated by dephosphorylation (Jiang et al., 2006; Roskoski, 2005; von Wichert et al., 2003; Ylinen, 1996; Zheng et al., 2000). Alternatively, clustering of integrins could also directly activate SFKs (Arias-Salgado et al., 2003; Yu et al., 2011). SFKs then phosphorylate tyrosine residues on Rho GTP exchange factors, such as leukemia-associated Rho GTPase exchange factor (LARG, also known as Rho GEF factor 12) or Tiam1 (Elias et al., 2010; Guilluy et al., 2011), resulting in Rho GTPase activation and cytoskeletal reorganization (Majumdar et al., 1999; Nobes and Hall, 1995; Rottner et al., 1999). During adhesion formation in fibroblasts, RPTP-α also interacts with αvβ3 integrins to activate the SFK Fyn (Kostic and Sheetz, 2006; Su et al., 1999; von Wichert et al., 2003). Although the GTP exchange factors (GEFs) remain unidentified, two Rho GTPases (Rac1 and Cdc42) are activated downstream of RPTP-α, and they are known to regulate actin dynamics and focal adhesion turnover during fibroblast spreading (Arthur and Burridge, 2001; Herrera Abreu et al., 2008; Jones and Katan, 2007; ten Klooster et al., 2006). Rho GTPases can also regulate the region of activity of other Rho GTPases to determine cell morphology (Vega et al., 2011), activate the ARP2/3 complex (Welch, 1999) or directly activate formins. In addition, formins can pass on mechanical signals to the nucleus to induce expression of genes that are under the control of the serum response element (Young and Copeland, 2010). In some, if not most, of these cases, there appears to be crosstalk between different small G proteins (Machacek et al., 2009).

Initially identified as a Src kinase substrate (Schaller et al., 1992; Weiner et al., 1993), the tyrosine kinase FAK is a pivotal signaling molecule for focal adhesion dynamics, and acts as both a direct mechanosensor, as discussed above, and downstream of mechanical signaling events (Schober et al., 2007; Wang et al., 2001). FAK activation is initiated by autophosphorylation at tyrosine 397, which can be induced by integrin clustering or binding of tyrosine-phosphorylated peptides (Chen and Chen, 2006; Sieg et al., 1999; Tomar and Schlaepfer, 2009). FAK can also activate integrins, resulting in adhesion strengthening (Michael et al., 2009). The activated FAK alone – or in complex with Src – can then phosphorylate target proteins or act as scaffold to recruit additional SH2-binding proteins to adhesions. Central to its function as regulator of adhesion dynamics, FAK phosphorylates both GEFs and GAPs to modulate the activity of Rho GTPases (Fig. 3) (see Tomar and Schlaepfer, 2009). FAK also phosphorylates paxillin, thereby enhancing focal adhesion turnover (Zaidel-Bar et al., 2007b), and binding of FAK in a complex with Src to p130Cas results in Cas activation (Fonseca et al., 2004).

Cas family proteins are hyperphosphorylated downstream of FAK and SFKs, in a manner that, at least in the case of p130Cas, is enhanced by mechanical stretching (Cary et al., 1998; Sawada et al., 2006; Tamada et al., 2004). During cell migration, phosphorylation of p130Cas at tyrosine 249 acts as a switch for the formation of a complex with Crk and the Rac1 GEF DOCK180 (also known as DOCK1). This complex regulates Rac1 activity and cytoskeletal dynamics (Brugnera et al., 2002; Gustavsson et al., 2004; Webb et al., 2004). Additionally, p130Cas associates with the GEF AND34 (also known as BCAR3) and regulates the activity of the small GTPases Ral, Rap1 and R-Ras in cell motility (Gotoh et al., 2000; Riggins et al., 2003).

Posttranslational modifications other than phosphorylation are relevant for many cytoskeletal and adhesion proteins, and could be involved in mechanosensing. Acetylation of α-tubulin stabilizes microtubules, which can regulate vesicular trafficking, autophagy or cell motility (Hubbert et al., 2002; Perdiz et al., 2011; Tran et al., 2007). The addition of arginyl from tRNA onto the N-terminus of proteins has been detected in vivo on cytoskeletal proteins including α- and β-actin, myosin and tubulin (Karakozova et al., 2006; Wong et al., 2007). In actin, arginylation changes the interaction dynamics with several binding partners and alters the polymerization properties of the proteins (Karakozova et al., 2006; Saha et al., 2010). Modification of proteins with fatty acid or prenyl residues assists localization to cellular structures and is found on cytoskeletal and adhesion proteins, such as vinculin or several members of the formin family (Han et al., 2009; Harris et al., 2010; Kellie and Wigglesworth, 1987; Resh, 2006).

Nucleocytoplasmic O-GlcNAcylation (i.e. addition of O-linked N-acetylglucosamine) is especially interesting owing to its extensive crosstalk with phosphorylation, as both modifications frequently compete for the same serine or threonine residues (Wang et al., 2007; Wang et al., 2008; Wang et al., 2010). The differences in size and charge between these modifications can alter protein function and stability (Chen et al., 2006; Liang et al., 2006). Several adhesion and actin-binding proteins, such as talin, actin-capping protein, actin-binding LIM proteins (Chalkley et al., 2009; Hagmann et al., 1992; Wang et al., 2010; Zhu et al., 2001) and striated muscle isoforms of actin and myosin (Hédou et al., 2009; Ramirez-Correa et al., 2008) are subjected to O-GlcNAcylation, which might modulate properties of the cytoskeleton, or alter muscle contractility (Cieniewski-Bernard et al., 2009). Increased O-GlcNAcylation is a cellular response to various stress-inducing agents, whereas heat-shock proteins can associate with the glycan to protect the modified proteins (Lefebvre et al., 2001; Zachara et al., 2004). O-GlcNAcylation of adhesion and cytoskeletal proteins could therefore also constitute a protective mechanism against tensile stress. Post-translational modifications such as O-GlcNAcylation can thus be expected to receive further attention, particularly as its deregulation has been correlated with the occurrence of cancer metastasis and invasiveness (Gu et al., 2010; Perdiz et al., 2011).

Despite many recent advances, we still lack a clear picture of the molecular mechanical pathways that transmit and detect forces in adhesions, upstream of biochemical signaling. Simple biophysical considerations regarding molecular stiffness and binding forces (Table 1) can help us to identify probable mechanosensors in soft compared with stiff environments, and separate them from molecules that simply transmit forces. Nevertheless, a systematic analysis of adhesion protein binding and unfolding under physiological pulling rates is required to clarify where and when bonds break and proteins unfold in the network of links that are connected in series from the ECM to actin, and in parallel in integrin clusters. This seems even more pressing after considering that common pulling rates in single-molecule experiments (10²–10⁴ nm/second, Table 1) are much higher than physiological rates (1–1×10² nm/second). Such an analysis might unravel simple but elegant biophysical mechanisms; for instance, regulation of the rearward speed of cytoskeletal components and the loading rates might trigger different mechanotransduction cascades by changing the relative resistance to unbinding and unfolding of different proteins. Differences in loading rates and force levels could also help to explain the different behaviors characteristic of fast events, such as initial adhesion formation, as compared to slower tissue or ECM remodeling. Only when this biophysical picture is complete, will we be able to reliably and mechanistically explain how the biophysical inputs (i.e. forces) control the biological outputs of biochemical signaling and effector pathways that comprise cell behaviors.

The authors thank Simon Moore for helpful discussions.