Cell–extracellular matrix dynamics

Integrins comprise a transmembrane receptor family consisting of 18α and 8β heterodimeric protein subunits that specify the ECM ligand and binding site, respectively, forming 24 different pairs [7, 8]. For example, β1 integrin binds preferentially to the arginine–glycine–aspartic acid (R–G–D) residues of ECM molecules, α1 and α2 subunits promote binding to collagens, and α5 specifies fibronectin binding. Blocking the ligand-binding site (via RGD blocking peptides or antibodies) inhibits ECM binding in a concentration-dependent manner and blocks cell adhesion and migration. The integrin extracellular domain binds ECM ligands (figure 2(A)). The β cytoplasmic tail region interacts only indirectly with actin. Several proteins within the focal adhesion, namely the talins, the tensins, and the kindlins, bind to the cytoplasmic tail of the β subunit and provide a physical link to other proteins, including actin. While there are numerous features of integrin functions, two important and possibly interrelated concepts are key to their role in mechanobiology: integrin activation and integrins acting as catch bonds. Because other proteins of the mechanotransduction machinery are also known catch bonds, we will discuss this concept later in this review.

Zoom In Zoom Out Reset image size

Integrin activation is associated with changes in the conformation state of both the extracellular and intracellular domains of the integrin heterodimer [9]. Activated integrins have an open or extended extracellular domain conformation for binding ligand, accompanied by separation of the cytoplasmic tails of the α and β subunits, which is thought to promote integrin clustering, focal adhesion assembly, and signal transduction (figure 2) [8]. These conformational changes increase ligand-binding affinity and the force-bearing capability of an individual monomer, which together can alter integrin off rates and adhesion lifetime. While activation most often refers to extension of the β subunit, α5 integrin has an activation state associated with the binding of the synergy site to fibronectin that contributes to α5β1 load capacity [10].

Activation can also be induced by the intracellular binding of either kindlin or talin, known as inside-out activation. Their binding to the cytoplasmic domains is not solely responsible for activation but is thought to accelerate the process [8, 9]. Force may also be involved: once integrins extend/open and bind to ligand, the resulting talin/kindlin binding and subsequent actin association for force production stabilizes the ligand association. Because of this increased ECM binding affinity, integrin activation is likely a key regulatory point for mechanosensing.

Focal adhesions act as mechanosensory organs, sensing and adapting to changes in tension, e.g., intracellular contractility or stiffness of the ECM. The three main types of adhesion are (1) nascent adhesions, (2) focal adhesions, and (3) fibrillar adhesions, with the latter being involved in fibronectin fibrillogenesis (discussed elsewhere in this review: table 1 and figure 1). Current consensus is that focal adhesions initially form as force-independent nascent adhesions consisting of integrins, paxillin, vinculin, talin, FAK, Src, GIT, βPix and others (table 1) [11]. Nascent adhesions have short lifespans of seconds to 2 min. To mature to focal adhesions, they require the application of force. Such maturation involves increased numbers of proteins involved in actin polymerization, and its hallmark is growth of the adhesion to a size greater than 1 μm². Adhesion growth also involves a local increase in the density of integrin cell–ECM receptors to reinforce the added forces from the cytoskeleton. Talin is integral to this maturation process, where actin binding helps to unfold this lengthy structural protein, exposing up to 4 and 11 binding sites for actin and vinculin, respectively [12]. This force-dependent unfolding is thought to structurally reinforce the integrin-talin complex [8]. Recent evidence suggests that two lesser-known adhesion components, KANK and CDK1, may bind directly to talin and regulate its ability to activate integrins and its mechanosensitivity, respectively [13, 14]. Without full-length talin, force transmission is weak, so talin is likely the key mechanosensitive protein required for focal adhesion growth and maturation [15].

While force is integral for adhesion growth and stabilization—treatment of cells with contractility inhibitors such as Rho kinase inhibitor Y-27632 and the myosin II ATPase inhibitor blebbistatin inhibit growth, and only nascent adhesions are formed [16]—force is also responsible for focal adhesion disassembly; likely important is the rate of force loading to individual integrins [17, 18]. Currently, the two-spring model has gained traction as the primary model associated with mechanisms of force-dependent cell–ECM interactions; it has recently been strengthened both theoretically and experimentally [15, 17, 19, 20].

The two-spring model describes two Hookean springs in series, one representing the ECM, the other the cytoskeleton, with dynamic cell adhesion complexes (integrins) between them (figure 2(B)) [17, 20]. The cellular components of this model display variable on rate (K_on) and off rates (K_off), where K_on and K_off are highly dependent on integrin avidity and contractility, respectively, and regulate adhesion assembly and disassembly mechanisms. In essence, the model predicts that the softer of the two springs defines the threshold at which an adhesion reaches steady state (K_on and K_off are equal). For the adhesion to reach steady state, both springs must exert equal tension within the system: too little cytoskeletal force results in adhesions that do not stabilize on a rigid surface, and too much cytoskeletal force leads to adhesion disassembly and rapid protein dynamics on a soft surface. Because the soft spring sets the steady-state threshold for stability, it can vary—hence the system is meant to be dynamic and responsive to both internal and external changes. In other words, it is 'mechanoresponsive'. While initially modeled for 2D systems, there is evidence of a similar mechanism in 3D environments [19].

The molecular clutch hypothesis originally hypothesized by Mitchison and Kirschner [21] is involved in traction force propagation—the transmission of actomyosin contractility to the underlying/surrounding matrix through cellular adhesions—that is integral to migration. Application of traction force is posited to slow actin flow at the focal adhesion sites formed at the lamellipodia-lamella border, so that actin polymerization can proceed anteriorly to exceed the retrograde flow and thereby push against the cell membrane and extend the leading edge (figures 1(D) and 3) [1]. Hence, the general purpose of this 'clutch' is to promote migration. The hierarchical transmission of actin flow exists between the different core structural and signaling proteins within focal adhesions. Translocation of the actin-binding proteins talin, vinculin, and α-actinin correlate closely with actin flow, moving rapidly, whereas other signaling proteins (paxillin, FAK) and integrins move relatively slowly with respect to actin dynamics [22]. Chan and Odde [23] demonstrated that this 'motor-clutch' varies depending on substrate compliance and is associated with cellular traction forces through which cytoskeletal forces are transmitted to the substrate: stiff substrates lead to frictional slippage with few 'clutches' being engaged at any time, leading to high actin flow rates and low traction. Softer substrates show a load-fail regime with slower actin flow rates, higher tractions and slower speeds as cells grip-and-release repetitively, retracting newly formed adhesions behind the leading edge and reducing migration. These mechanisms are likely due to the system's rate constants, where the rate of tension development leads to an inverse relationship between actin flow and traction force (i.e., rapid tension development leads to lower traction force). Hence, clutch engagement is higher on softer surfaces. While this model does not explain how stiffer substrates promote higher traction force, others have demonstrated that the load-and-fail model does likely occur with oscillations of traction force in fibroblasts on 2D surfaces [24] and the pulling and retraction behavior of adhesions in soft collagen gels [19]. As for the effect on overall cell migration and cell persistence, it is difficult to assess events that vary greatly in time (seconds versus hours) and physical scale (micro versus macro). However, rapidly moving cells such as fish keratocytes, immune cells and Dictyostelium discoideum show lower traction and have relatively low actin flow rates [25].

Zoom In Zoom Out Reset image size

A key step in the regulation of this clutch machinery is the unfolding of the mechano-sensitive protein talin [15]. Without talin, little to no traction force is generated. Increased traction force likely occurs when additional vinculin and actin binding sites are exposed after talin unfolding. In addition, the force relationship is biphasic, such that a threshold level of force must be reached—otherwise integrins release, and adhesions do not mature and strengthen. This system can be influenced or regulated by not only substrate stiffness but by the relative amount of ECM ligand, which is also known to alter cell migration rates [26]. The type of ECM also affects the transmission of cytoskeletal force depending on the affinity of its bound integrin [27]. This affinity affects how much the molecular clutch will 'slip' with altered dynamics as shown after changing the ECM [27] or after changing the integrin that associates with a particular ECM [18].

Cell adhesion dynamics and the molecular clutch depend on catch bonds. Although most non-covalent protein–protein interactions are considered to be slip bonds, where increased pulling force decreases bond lifetime, integrins and several other proteins of the mechanotransduction machinery can form catch bonds. With catch bonds, increasing tension between two proteins increases bond lifetime. A simple example of a catch bond mechanism is the Chinese finger trap puzzle, where stronger pulling in opposite directions increases the tension within the trap. Numerous catch bonds exist in biology [28]. Recently, a simple tweezer-like design has simulated catch bond kinetics for a number of proteins including integrins and actin [29]. However, increasing tension beyond the loading capacity of a catch bond results in catastrophic rupture, e.g., bond failure of individual integrins [17]. Force versus bond-lifetime curves are biphasic, suggesting an optimal range for their biological function. There are likely at least four catch bonds involved in cell–ECM interactions: integrins, vinculin (now hypothesized to be a 'one-way' catch bond), actin, and myosin II [30, 31]. Interestingly, many processes, including focal adhesion formation and many aspects of cell migration, are biphasic and are likely due to the presence of these force-sensitive bonds.

Single-molecule experiments have shown that β1 integrin activation will affect catch bond strength or affinity as well as the absolute force the bond can withstand, indicating an interrelationship between these important functions. Integrin-ECM bond dynamics change as force increases in three regimes: (1) an initial low load rate (slow building of force) leads to spontaneous bond detachment before a substantial load occurs. (2) As the load rate increases, single bond forces increase and reduce the K_off rate below the K_on rate, leading to reinforcement by the addition of other integrins to increase their local density (integrin clustering). (3) If the loading continues to increase, K_off can elevate above K_on, resulting in the release of individual bonds before others can form, thereby decreasing force transmission [18]. Hence, integrins functioning as catch bonds in the molecular clutch can explain the growth, stabilization, and disassembly of focal adhesions.

While most studies use two-dimensional linear elastic or Hookean hydrogels coated with an adsorbed or conjugated ECM, the majority of biological in vivo 3D ECMs are nonlinear elastic in nature and can undergo strain-stiffening or strain-softening, where the gels either stiffen or soften with added strain. A recent intriguing concept applicable to both 2D and 3D ECMs is that stress relaxation of an ECM can alter adhesion dynamics. As a steady-state level of strain is applied to a material, the relative amount of stress decreases over time, which can be a fast or slow mechanism. Focal adhesions can respond to either fast or slow stress relaxation of hydrogels [2]. For example, higher traction forces can be sustained by fast gels and are associated with filopodia and not lamellipodia, resulting in altered cell migration. While the dynamics of cell–ECM adhesions under stress relaxation remains to be characterized, it is clear that the rheological properties of ECM can dictate the dynamics of cell–ECM adhesions.

Planar 2D ECMs have been a mainstay for biophysical analyses of cell–ECM interactions for many decades due to their ease of generation and repeatability. However, 3D matrices, such as collagen type I, laminin, and fibrin hydrogels, and cell-derived matrices (3D CDM) have emerged as the next relevant in vitro step toward directly studying cell–ECM dynamics in vivo. Other synthetic gels and biomaterials including electrospun fiber made of dextran methacrylate or polycaprolactone have also been used [42, 43], but for brevity we will not discuss these here. Biological hydrogels differ from 3D CDMs in several ways. Hydrogels are: (1) polymerizable, (2) sensitive to gelation conditions, (3) often composed of a single matrix protein, and (4) nonlinear elastic. In contrast, 3D CDMs are dependent on the cells that generate them (e.g., fibroblasts) and are composed of a developmental matrix with fibronectin, collagen I, and proteoglycans, and they are primarily linear elastic. In 3D laminin gels, some cancer cells alter their intracellular mechanics to mimic the stiffness of the surrounding ECM [44], suggesting that cells respond a 3D ECM similarly to a 2D pliable substrate.

Although thousands of publications have used 3D matrices (synthetic or biological) to characterize cell migration and adhesion components, relatively few have studied the dynamics of adhesions in these highly pliable environments, mainly due to the difficulties of imaging 3D volumes at depths of ⩾50 μm. Adhesion lifetimes are substantially longer in 3D CDMs than 2D focal adhesions [16, 19]. Adhesions in 3D collagen gels show a variable lifetime and zyxin turnover rate that is highly dependent on the local stiffness of ECM fibrils—stiffer fibrils promote longer lifetimes, while softer fibrils lead to adhesion disassembly through retraction [19]. In fibrin gels, cell adhesions undergo retrograde movement away from the leading edge with the rate of movement dependent on the protein, e.g., α-actinin moves faster than paxillin [45]. Together, these findings suggest that a similar molecular clutch mechanism is involved in 3D adhesion formation, stability, and disassembly.

Cells in 3D environments apply traction forces and strains to the ECM. Initially, forces were quantified using well-defined linear-elastic polyethylene glycol hydrogels containing both adhesive RGD and degradable proteolytic sites [46]. More recently, nonlinear elastic collagen and fibrin gels have demonstrated cell–ECM strain dynamics that are often cell type-dependent. MDA-MB-231 cells, a commonly used metastatic human breast cancer cell line, produce low ECM strains in 3D collagen. Differing reports describe (1) nearly equal-and-opposite strains [47], (2) high posterior strains [48], or high anterior strains [49]. These discrepancies may be due to different phases of the migration cycle or migratory persistence of individual cells. Overall, however, these cells cannot sustain constant ECM strain for extended periods. In contrast, mesenchymal fibroblasts generate anisotropic ECM strains nearly four-fold higher. These strains are two-fold higher anterior than posterior and are maintained during directional migration. Matrix prestrain in front of migratory cells is also exhibited by mesenchymal fibrosarcoma and melanoma cells [49, 50]. Formation and maintenance of this prestrain depends on levels of activated integrins and myosin IIA—both key regulators of adhesion dynamics.

The intracellular site of contractility in mesenchymal cells, based on 3D imaging of adhesion and cytoskeletal components, is at the anterior of the cell between leading edge and cell body, producing an adjacent pinch-like contraction of the matrix [45, 49]. We recently demonstrated that this anterior contraction is associated temporally with leading edge protrusion to establish a 3D migration cycle. We hypothesize that increasing contractility behind the leading edge stabilizes 3D adhesions and enhances molecular clutch engagement, permitting actin polymerization at the leading edge to exceed the slowing retrograde flow of actin to promote protrusion. Thus, concepts associated with 2D cell–ECM interactions also contribute to mechanosensing in more complex 3D microenvironments.

The majority of cancer-related deaths (∼90%) are caused by metastatic disease rather than the primary tumor [61]. For cancer cells to metastasize, they must first break through a basement membrane barrier and then migrate through the stromal ECM, both having pores narrower than the cells. Thus, to transmigrate and invade, cancer cells must navigate through both the basal/basement membrane and stromal matrices, through either localized proteolysis and/or physical forces (figure 5(C)).

The basement membrane is a thin, sheet-like network of proteins, composed of laminin, collagen IV, perlecan, nidogen, and proteoglycans. Laminin directly binds to cell surface receptors such as β1 integrin and dystroglycan, and it self-assembles into a dense sheet. Collagen IV then polymerizes to form a second covalently crosslinked network [62, 63]. Basement membrane is a nanoporous structure that restricts the diffusion of large molecules while being permeable to small molecules. The sizes of basement membrane pores vary depending on the tissue type, with average pore sizes ∼10–100 nm in diameter in different tissues [64, 65]. During the transition from carcinoma in situ (local cancer) to invasive carcinoma, cancer cells must enlarge basement membrane pores and migrate in the fibrillar ECM toward the circulatory or the lymphatic system. With the nucleus being the largest organelle of the cell (∼10 microns), successful invasion requires cells to widen substantially the nanometer-sized pores of the basement membrane to break through this barrier.

Once cancer cells have breached the basement membrane, they need to traverse tissues that often contain a fibrillar collagen matrix with pores that are roughly 2–10 μm in diameter in vivo [66]. CAFs can remodel ECM independent of proteolysis, through widening of pre-existing breaks or proteolytic perforations in the basement membrane [67]. Invasion of cancer cells does not depend on the stiffness of the matrix but instead on cellular contractility [67]. Interestingly, CAFs are also shown to act as leader cells, leaving 'microtracks' behind them for cancer cells to follow during invasion [53]. In contrast, another type of stromal cell in the tumor microenvironment, tumor-associated myoepithelial cells, can act as an additional barrier between the BM and cancer cells. In a 3D organoid model, myoepithelial cells act as a passive physical barrier to epithelial cells and can recapture escaping cells [68]. In addition to these barriers, a cell's geometric shape [69, 70] and the geometry of the surrounding ECM [71] can also affect the orientation of cell's leading edge and thus the direction of its migration, as demonstrated using micropatterning techniques [69–72].

Classically, cancer cell invasion has been thought to require the proteolytic degradation of both the basement membrane and stromal ECM. Both membrane-bound and secreted proteases have been implicated in basement membrane degradation and tumor cell invasion with matrix metalloproteinases (MMPs) being essential for each process [64, 73]. MMPs are zinc-dependent endopeptidases categorized into groups according to their substrate specificity; some are secreted, and others are membrane-bound (MT-MMPs) [74]. The two main post-transcriptional regulations of MMP activity are activation of the precursor form of the MMP and inhibition of the active MMP by tissue inhibitors of metalloproteinases (TIMPs) [75]. MMPs are synthesized as an inactive pro-enzyme through the interaction of a cystine-bound motif at the pro-peptide domain with a zinc ion at the catalytic site. MMPs are activated extracellularly by the removal of the pro-peptide domain [75, 76]. MMP expression is correlated with increased cancer invasion, metastasis, advanced tumor stage, and higher mortality. The expression of three membrane bound MMPs, MT1-MMP (MMP-14), MMP15 and MMP16, is especially important in tumor cell invasion of basement membrane matrix [77].

4.2.1. Invadopodia

Although initial invasion of cancer cells through the basement membrane barrier was thought to depend on proteolytic degradation through MMPs, emerging evidence has implicated other mechanisms in breaching of the basement membrane for cell invasion. Multiple clinical trials using MMP inhibitors have failed to reduce mortality [62, 64, 82]. This failure of anti-MMP clinical trials might have been due to insufficient or inadequately targeted drug concentrations due to side effects. Nevertheless, these findings raised the possibility that cells may be able to breach the basement membrane barrier through mechanisms independent of proteases during cancer progression. Additionally, although tumor or other cells can proteolytically remodel the stromal ECM, they may also require physical force to mechanically reorganize this matrix [64].

During cancer cell invasion, stromal cells in the tumor microenvironment—CAFs, macrophages, and myoepithelial cells—have been documented to exert force and remodel the basement membrane, opening pre-existing pores in the basement membrane, or leaving micro-tracks for cells to follow for cancer cell invasion [67, 83]. What remains unclear is whether force-driven breaching by the invading tumor cells themselves is equally important in cancer cell invasion and transmigration. In a Caenorhabditis elegans model of invasion, anchor cells use force generated by an actin network via the arp2/3 complex to deform and displace the basement membrane, resulting in accumulation of laterally displaced basement membrane in a ring around perforations in the absence of proteolysis [84]. Interestingly, invadopodia in C. elegans cells devoid of MMPs increase in size five-fold compared to normal invadopodia and are enriched in Arp2/3, ATP, and mitochondria [84]. Moreover, during salivary gland morphogenesis, epithelial cells require myosin II contractility in addition to proteases to perforate and remodel the basement membrane [85]. Although a recent paper demonstrated very similar MMP-induced perforations in the basement membrane required for early mouse embryonic development, the role of myosin II contractility remains to be explored [86]. These findings in normal developing embryos underscore the importance of ECM remodeling for both normal and malignant cells.