Recruitment of α_νβ₃ integrin to α₅β₁ integrin-induced clusters enables focal adhesion maturation and cell spreading

A layer of endothelial cells (ECs) constitutes the endothelium that lines the entire circulatory system, adhering to the underlying vessels or heart tissue. Normal vascular development, homeostasis and remodeling necessitates continuous and extensive crosstalk between ECs and their extracellular matrix (ECM) (Eming and Hubbell, 2011; Hahn and Schwartz, 2009). Under physiological conditions, ECs attach to a basement membrane, rich in collagen IV, laminin and proteoglycans, while injury and inflammation bring about changes in ECM composition. The presence of fibronectin (FN) in the basement membrane is a key element during the dynamic processes accompanying vascular remodeling (Hahn and Schwartz, 2009; Zhou et al., 2008). Indeed, mice lacking FN die before birth due to impaired vascular development (George et al., 1993). The presence of FN in the endothelium has also been correlated with tumorigenesis (Wang et al., 2016) and the progression of atherosclerosis (Feaver et al., 2010; Rohwedder et al., 2012).

ECs are able to both secrete FN and bind circulating plasma FN, which they can then reorganize into an extracellular fibrillar network. In its fibrillar, bioactive form, FN delivers essential mechanical and chemical polarity cues that allow endothelial cells to align correctly during vascular tubulogenesis (Hielscher et al., 2016; Zovein et al., 2010). Endothelial cell adhesion to FN is predominantly mediated by α₅β₁ and α_vβ₃ integrins, members of the integrin family of transmembrane heterodimeric adhesion receptors (Plow et al., 2014). The phenotype of mice lacking the major FN-binding α₅β₁ integrin (also known as the fibronectin receptor) is less severe compared to that of FN-knockout mice (Yang et al., 1993), presumably due to the ability of α_v integrins to compensate for α₅β₁ integrin loss (Yang and Hynes, 1996). By contrast, endothelial cell-specific deletion of α₅ had no obvious effects on developmental vasculogenesis, while a double knockdown of both α₅ and α_v integrins in ECs led to some developmental defects, which were significantly different compared to total α₅ integrin- or fibronectin-null embryos (van der Flier et al., 2010). These findings call for a clearer understanding of the individual contribution and cooperation of these integrins on EC physiology.

Integrins cluster in focal adhesions (FAs) at points of contact with the ECM, providing the essential physical and chemical link necessary for cell survival, proliferation and migration (Geiger et al., 2009). Recent studies have verified FA formation under ECs in vivo with characteristics corresponding to their location in the circulatory system (van Geemen et al., 2014). While FN is mostly absent from healthy EC basal membranes, it is involved in the processes of new vessel formation (Zhou et al., 2008) and atherosclerosis (Feaver et al., 2010). α₅β₁ and α_vβ₃ integrins both bind the RGD amino acid sequence at the FN type III₁₀ repeat; α₅ integrin additionally interacts with a synergy site located in the adjacent FN type III₉ repeat, making α₅β₁ integrin specific for FN, compared to other RGD-containing ECM proteins (Aota et al., 1994; Mould et al., 1997). Importantly, following their initial clustering, only α₅β₁ is able to relocate from FAs in a contractility-dependent manner to fibrillar adhesions, and mediate FN fibrillogenesis (Pankov et al., 2000; Zamir et al., 2000; Zhong et al., 1998).

A substantial degree of cooperation and crosstalk between α₅β₁ and α_vβ₃ integrins has emerged over the years. Both integrins are required for efficient mechanosensing (Schiller et al., 2013) and directional migration (Missirlis et al., 2016). However, beyond their overlapping roles and sharing FN as a common ligand, α₅β₁ and α_vβ₃ integrins exhibit distinct intracellular binding partners and signaling (Danen et al., 2002; Morgan et al., 2009). These differences manifest in altered cytoskeleton organization, force transmission and traction orientation as a function of the integrin type engaged (Balcioglu et al., 2015). In particular, studies have shown that high matrix forces are primarily supported by clustered α₅β₁ integrins, while less stable links to α_vβ₃ integrins initiate mechanotransduction (Lin et al., 2013; Rahmouni et al., 2013; Roca-Cusachs et al., 2009). The synergy between α₅β₁ and α_vβ₃ integrins thus enables the cell to sense and adapt to its mechanical microenvironment (Schiller et al., 2013), a trait that is relevant when tissues go through extensive physical remodeling during embryonic development, angiogenesis and cancer progression.

One aspect of integrin-mediated adhesion to FN expected to influence cell behavior is the temporal control of engagement for each integrin. Integrins transfer between intracellular compartments and the plasma membrane in a process known as integrin trafficking (Paul et al., 2015); interestingly, each heterodimer appears to dictate the recycling pathways of the other (Moreno-Layseca et al., 2019). Initial binding to FN seems to be mediated by α₅β₁ integrins, since blocking this integrin interferes with cell adhesion (Dejana et al., 1988; Schwartz and Denninghoff, 1994), and α₅β₁, but not α_vβ₃ integrins, initially recognize and bind FN (Huveneers et al., 2008; Sun et al., 2005). Following initial binding of α₅β₁ integrins to FN, Strohmeyer et al. (2017) found that force loading of these bonds leads to rapid α_vβ₃ recruitment and adhesion reinforcement; however, these data are in apparent contrast with an atomic force microscopy (AFM)-based study from the same group suggesting that α_vβ₃ out-competes α₅β₁ for initial binding to FN and is responsible for priming α₅β₁ for subsequent engagement (Bharadwaj et al., 2017). It is conceivable that the discrepancy arises from experimental details, or rapid switching of integrins; nevertheless, it is clear that more work is required to clarify this issue.

Overall, despite significant progress, a clear mechanistic picture of how α₅β₁ and α_vβ₃ integrins mediate distinct functions by binding the same ECM ligand is lacking. Such information is critical when designing biomaterials or drug delivery systems that target specific integrins. Many of the aforementioned studies were based on genetic manipulations, which are potentially associated with side-effects, adaptation through compensatory mechanisms or deregulation of integrin trafficking. Moreover, cells were typically presented with ECM ligands that do not enable specific and selective integrin engagement. In this study, we combined the use of engineered substrates and non-peptidic ligands (Mas-Moruno et al., 2016) to drive integrin-selective adhesion and reveal the contribution of each integrin on the same cell population. The α_vβ₃- and α₅β₁-selective integrin antagonists (peptidomimetics) we employed exhibit very potent integrin-binding affinities (IC₅₀ values in the sub- to nanomolar range) towards the respective integrin, combined with high selectivity of at least two orders of magnitude lower affinity towards the other integrin, or α_vβ₅ integrin (Table S1) (Kapp et al., 2017). These antagonists have previously been used to examine the role of these integrins on cell adhesion and migration (Guasch et al., 2015; Missirlis et al., 2016; Rechenmacher et al., 2012; Schaufler et al., 2016). In particular, the α₅β₁ integrin peptidomimetics are unique in this respect, since selective α₅β₁ ligands are still lacking, despite efforts to modulate integrin affinity through mutations of FN fragments (Martino et al., 2009). The specificity and selectivity of substrates functionalized with α_vβ₃- and α₅β₁-selective peptidomimetics have been validated by showing that only cells expressing the corresponding integrin adhere to the substrates (Rechenmacher et al., 2012). Despite the high ligand specificity, FAs of ECs formed on α₅β₁ integrin-selective substrates stained positive for α_vβ₃ integrins, but not vice versa. Hence, we set out to investigate the apparent α_vβ₃ integrin recruitment to α₅β₁ integrin-based adhesions on ECs, describe its functional effects and reveal the underlying mechanism.

The use of nano-patterned gold substrates permits the selective presentation of ligands to cells, with additional control over ligand density and type (Arnold et al., 2004; Guasch et al., 2015). The immobilized gold nanoparticles have a size (5–10 nm) comparable to that of integrins, and the space between nanoparticles is passivated with a poly(ethylene glycol) (PEG) layer. As a result, cells can interact exclusively with immobilized ligands using one integrin per nanoparticle owing to steric properties. The threshold in ligand density for efficient cell adhesion appears to differ between cell types (Arnold et al., 2004; Muth et al., 2013); therefore, we initially examined EC spreading on quasi-hexagonal, nano-patterned substrates as a function of inter-particle spacing to determine the minimal ligand density required for efficient cell spreading (Fig. 1A; Fig. S1). Human umbilical vein ECs spread efficiently on substrates functionalized with a cyclic-RGD peptide (c-RGDfK) (Haubner et al., 1996) only when the inter-particle distance was ≤50 nm, corresponding to a threshold density of ∼400 particles/μm² (Fig. 1B,C; Fig. S1). This value is higher compared to what has previously been reported for a fibroblast cell line (Arnold et al., 2004), but lower compared to that reported for hematopoietic cells (Muth et al., 2013), suggesting that the threshold distance between integrins for promoting efficient cell spreading is indeed cell-type specific. Of note, cells were unable to adhere to the non-patterned, PEG-coated area of the substrates, even in presence of serum, validating the specificity of the interactions with the functionalized ligands (Fig. S1). For ECs, maximal cell adhesion and projected cell area were observed on the shorter spacing examined (Fig. 1D,E), while the cell aspect ratio did not largely depend on ligand density (Fig. 1F). The high aspect ratio on substrates with inter-particle spacing of 70 nm most likely reflects high cell protrusive activity as the cell explores for permissive adhesion sites, and not effective cell polarization (Fig. 1F). Similar trends in cell adhesion and spreading as a function of ligand density were observed for ECs on nano-patterned substrates functionalized with the selective α₅β₁ or α_vβ₃ integrin antagonists (Fig. 1D–F). Despite the slightly lower efficiency of adhesion observed for ECs on integrin-selective substrates compared to c-RGDfK for the 30 nm inter-particle spacing (Fig. 1D), cell spreading and morphology were similar (Fig. 1E,F). Taking into account these results, we selected nano-patterned substrates with inter-particle spacing of 30 nm to investigate EC adhesion as a function of specific integrin engagement.

We next examined the assembly of focal adhesions (FAs) on ECs at the initial stages of cell spreading (3 h post-seeding), when contributions from cell-secreted matrix and substrate remodeling are minimal. ECs on nano-patterned substrates with an inter-particle distance of 30 nm exhibited differences in FA morphology and distribution as a function of the presented ligands, and hence of the corresponding integrins used by ECs to adhere (Fig. 2A). Substrates presenting a c-RGDfK peptide, which primarily binds α_vβ₃ integrins (García et al., 2002; Petrie et al., 2006), or functionalized with α_vβ₃ integrin antagonists, promoted the formation of larger and more peripheral FAs, compared to substrates functionalized with α₅β₁ integrin-selective antagonists, on which ECs assembled smaller but more elongated adhesions, as well as central adhesions (Fig. 2A–C). On control FN-coated glass coverslips, ECs assembled a high number of FAs and fibrillar adhesions, as expected. The differences in EC adhesion cluster assembly as a function of integrin specificity are consistent with previous reports (Danen et al., 2002; Missirlis et al., 2016; Schiller et al., 2013) and indicate that ECs can distinguish the presented ligands on the integrin-selective substrates.

We next examined the effects of selective integrin engagement on integrin localization using the peptidomimetic integrin antagonists (Table S1). ECs express both α₅β₁ and α_vβ₃ integrins, among others (Fig. S2A). The integrin selectivity of substrates was verified by blocking studies: EC adhesion on α₅β₁ integrin-selective substrates was inhibited by soluble α₅β₁ integrin, but not α_vβ₃ integrin, peptidomimetics and vice versa for α_vβ₃ integrin-selective substrates (Fig. S2B). On control FN-coated glass substrates, EC adhesion was reduced by 15% and 60% following blocking of α_vβ₃ and α₅β₁ integrin, respectively, indicating that α₅β₁ is the principal receptor for initial FN binding (Fig. S2C), confirming previous reports on ECs (Dejana et al., 1988; Schwartz and Denninghoff, 1994) and fibroblasts (Missirlis et al., 2016). Blocking both integrins abolished EC adhesion on FN, confirming that α_vβ₃ and α₅β₁ integrins are the principal FN-binding receptors of ECs (Fig. S2C). Accordingly, α_vβ₃ and α₅β₁ integrins co-localized in FAs of ECs spreading on FN (Fig. S2D).

On α₅β₁ integrin-selective substrates, immunostaining of non-transformed ECs with an antibody against the α₅ integrin subunit (clone MAB11) revealed clustering of α₅β₁ integrins as expected (note that α₅ pairs exclusively with β₁) (Fig. 3A). Surprisingly, integrin α_vβ₃ co-clustered in the same adhesions, indicating their recruitment, despite the absence of presented α_vβ₃ ligands (Fig. 3A). This result aligns with preliminary microscopic observations on transformed osteosarcoma cells showing α_vβ₃ integrins co-clustering with α₅β₁ integrins on α₅β₁-selective substrates (Guasch et al., 2015; Schaufler et al., 2016). Immunostaining with monoclonal antibodies against the activated form of integrin α₅β₁ (clones SNAKA51 and 9EG7) and α_vβ₃ (clone Ab62) demonstrated that integrin α_vβ₃ was present in α₅β₁-based adhesion in its active form (Fig. S2E,F). The finding that α_vβ₃ is in its extended (active) form further argues against the possibility that it binds a ligand (e.g. Thy1) anchored on the plasma membrane in a bent conformation (Fiore et al., 2015). The co-clustering of integrins was not observed on α_vβ₃ integrin-selective substrates, on which α₅β₁ integrin was not enriched in FAs abundant in α_vβ₃ integrins (Fig. 3A).

In order to validate the microscopy observations and quantify recruitment, we defined the enrichment ratio as the ratio of antibody staining intensity inside adhesion clusters to the intensity in a surrounding region of interest (for details on enrichment ratio calculation see Fig. S3). Adhesion clusters were determined from the antibody staining of the integrin corresponding to the presented ligand (e.g. adhesion clusters on α₅β₁-selective substrates were outlined from α₅ integrin staining). An enrichment ratio value of 1.0 corresponds to similar staining inside and outside adhesion clusters, and therefore absence of integrin recruitment, while higher values correspond to integrin enrichment within clusters. The enrichment ratio of integrins to their corresponding ligands was high, between 1.5 and 2.0, as expected (Fig. 3B, solid green and orange boxes). The enrichment ratio for α_vβ₃ integrin on α₅β₁-selective substrates was similar to that for α₅β₁ integrins, confirming the observed recruitment (Fig. 3B, hatched green box). By contrast, the enrichment ratio for α₅β₁ integrins on α_vβ₃ integrin-selective substrates was significantly lower than that for α_vβ₃ integrins, and close to the value of 1.0 (Fig. 3B, hatched orange box).

The integrin-specific recruitment was additionally detected on a second, non-transformed cell type, namely on primary human dermal fibroblasts (pHDFs) (Fig. S2G). Moreover, analogous results showing α_vβ₃ integrin recruitment to FAs of ECs assembled on α₅β₁-selective substrates, but not vice versa, was observed on PEG-based, non-patterned, integrin-selective substrates (Fig. S2H,I), thus excluding a potential artifact of nano-patterned ligand presentation. Overall, our quantitative data demonstrate the recruitment of α_vβ₃ integrins to FAs, which are assembled following initial substrate engagement through α₅β₁ integrins.

Previous work implicating acto-myosin contractility in the regulation of α_v and β₁ integrin crosstalk (Schiller et al., 2013) led us to the hypothesis that RhoA-mediated intracellular signaling is responsible for the observed recruitment of α_vβ₃ integrin to α₅β₁ integrin-based FAs. Thus, we examined how modulation of actomyosin contractility affected the aforementioned integrin recruitment, by culturing ECs on PEG-based, nano-patterned hydrogels (Fig. S4A) (Aydin et al., 2010). Hydrogel Young's moduli were controlled in the physiologically relevant range of 1–40 kPa, similar to those reported for arterial vessels (Klein et al., 2009), by changing the concentration of the precursor PEG di-acrylate (Fig. S4B). As expected, ECs responded to substrate elasticity by altering their spread area, FA assembly and actin stress fiber organization (Fig. S4C–E). Importantly, ECs responded to substrate stiffness on both integrin-selective ligands, with a similar increase of cell area (Fig. S4D), and a more pronounced increase in FA area on α_vβ₃-selective substrates with increasing stiffness (Fig. S4E). In addition, an enhancement in nuclear localization of the transcriptional coactivator with a PDZ-binding motif (TAZ) on the stiffer hydrogels was noted, further validating the mechanosensing ability of ECs on the nano-patterned hydrogels (Fig. S4F,G).

Having validated the suitability of integrin-selective, nano-patterned hydrogels for EC mechanosensitivity studies, we examined the distribution of α₅β₁ and α_vβ₃ integrins as a function of substrate elasticity. On α₅β₁-selective 10 kPa and 40 kPa hydrogels, α₅β₁ and α_vβ₃ integrins clustered in FAs (Fig. 4A), similar to the observations on the rigid (50–100 GPa) nano-patterned glass substrates (Fig. 3). The softer 1 kPa hydrogels did not support FA formation, and α₅β₁ and α_vβ₃ integrins co-localized at the cell periphery, independently of the presented ligand (Fig. 4A). Accordingly, direct inhibition of myosin II using blebbistatin resulted in disassembly of FAs on rigid, nano-patterned glass substrates and a similar co-localization of both integrins at the periphery of ECs, independent of the presented integrin-selective ligand, and despite their more spread morphology (Fig. S5). The observed integrin distribution on soft hydrogels and blebbistatin-treated cells is reminiscent of reports on activated integrin localization at the cell edge during active cell protrusion (Galbraith et al., 2007; Kiosses et al., 2001). Quantification of integrin recruitment to FAs using the enrichment ratio parameter confirmed the microscopy observations, and did not reveal changes in the extent of recruitment of α_vβ₃-to-α₅β₁ integrin-based adhesions for the 10 kPa and 40 kPa substrates (Fig. 4B).

Overall, the above results suggest that α_vβ₃ integrin recruitment to α₅β₁ integrin-based FAs occurs under varied mechanical properties of the substrate, as long as these support FA formation. In the absence of FAs due to reduced contractility, there is no integrin segregation, with both integrins seemingly probing the substrate at cell edges.

The positive staining of clustered β₃ integrins with the β₃ integrin-specific LIBS2 antibody (LIBS, ligand-induced binding site) (Du et al., 1993; Wierzbicka-Patynowski et al., 1999) in α₅β₁ integrin-based FAs suggested that α_vβ₃ integrins are in an extended (active) conformation and hence more likely to bind an extracellular ligand. We investigated the possibility that such a hypothetical ligand accumulates at FAs and precedes α_vβ₃ integrin recruitment. Incubation of ECs with soluble α_vβ₃ antagonists in order to block α_vβ₃ integrin engagement did not alter adhesion efficiency on α₅β₁ integrin-selective substrates (Fig. 5A,B; Fig. S2B), but resulted in dramatic changes of cell morphology: cells exhibited an irregular, non-cohesive shape with multiple, long protrusions (Fig. 5A). Similar results were obtained using a blocking antibody against α_v integrins (Fig. S6A). Image analysis revealed a significant reduction in EC spread area (Fig. 5C) and convexity (a measure of shape cohesion) upon α_vβ₃ blocking (Fig. 5D). The loss of cohesion upon α_vβ₃ integrin blocking on α₅β₁ integrin-selective substrates was reminiscent of the phenotype observed when myosin II activity is lost or inhibited (Rossier et al., 2010). Consistent with the reported localization of α_vβ₃ integrin at areas of high traction forces (Schiller et al., 2013) and its suggested role in mediating rigidity sensing (Roca-Cusachs et al., 2009), blocking of α_vβ₃ integrin could render the cells unable to sustain cell-generated tractions solely through α₅β₁ integrins and lead to loss of actin bridges along the cell edge. In line with such a scenario, α_vβ₃ integrin blocking impaired assembly of bona fide FAs, suggesting that α_vβ₃ integrin accumulation at initial adhesion clusters of α₅β₁ integrin is necessary for their maturation (Fig. 5E). Similar results were obtained for fibroblasts (Fig. S6B).

By contrast, blocking of α_vβ₃ integrins had no effect on morphology of ECs spreading on FN-coated glass, even though α_vβ₃ integrin blocking prevented its clustering at FAs (Fig. 5E). We propose that the synergy site adjacent to the RGD sequence present in the central cell-binding domain of FN, which is missing on the reductionist, nano-patterned α₅β₁ integrin-selective substrates, is at least partially responsible for the distinct cell behavior between these substrates. Indeed, α₅β₁ integrin–FN bond strength is decreased when the synergy site is mutated, leading to a dramatic reduction of cell adhesion strength (Friedland et al., 2009; Sun et al., 2005). Moreover, a recent report showed that the synergy site can compensate for loss of α_vβ₃ integrins, and vice versa (Benito-Jardón et al., 2017), supporting our finding that selective engagement of α₅β₁ integrin to the RGD peptidomimetics does not promote cell spreading when α_vβ₃ integrins are blocked.

If α_vβ₃ integrins are recruited to FAs on α₅β₁ integrin-selective substrates by binding an independently accumulated extracellular ligand, then recruitment should occur subsequent to, and not concurrent with, initial α₅β₁ integrin-induced formation of adhesion clusters. We therefore examined the kinetics of α_vβ₃ integrin recruitment to α₅β₁ integrin-based adhesion clusters during early EC spreading. α_vβ₃ integrin co-clustered with β₁ integrins as early as 15 min following cell seeding to α₅β₁ integrin-selective substrates; after 30–60 min, mature FAs clearly exhibited integrin co-localization (Fig. 6A). The enrichment ratio for each integrin, calculated in this case based on staining of the early FA protein paxillin, showed a monotonic increase for α_vβ₃ integrins during the first hour after seeding, but instead reached a plateau for β₁ integrins within 30 min. Moreover, the fraction of adhesion clusters that did not exhibit integrin enrichment (enrichment ratio ≤1) was higher for α_vβ₃ integrin at early time points (Fig. 6B).

Monitoring adhesion cluster formation during cell spreading did not allow the assessment of earlier time points (<15 min after seeding) since only a very small fraction of ECs firmly adhered on substrates and resisted washing. Therefore, we opted for a blebbistatin wash-out experiment: blebbistatin-treated ECs are well spread but lack FAs, which rapidly assemble upon drug removal (Shutova et al., 2012). At the earliest stages of assembly (5 min after wash-out), some adhesion clusters exhibited α₅β₁ integrin staining with absent, or faint α_vβ₃ staining; over time, α_vβ₃ integrins accumulated to maturing FAs (Fig. 6C). These observations are reflected in the calculated enrichment ratio values, which show a significant increase for both integrins between 5 min and 10 min after blebbistatin removal (Fig. 6D). Of note, the difference in absolute enrichment ratio values between the cell spreading and blebbistatin wash-out experiments most likely results from differences in integrin localization and availability on the cell membrane during the FA assembly process. Overall, our data indicate that α_vβ₃ integrins are recruited rapidly following the assembly of optically-resolved adhesion clusters by α₅β₁ integrins.

Our results so far are consistent with a hypothetical model where initial α₅β₁ integrin adhesion clusters recruit an unidentified protein, which then provides binding sites for α_vβ₃ integrin recruitment, and leads to FA maturation and efficient cell spreading on α₅β₁ integrin-selective substrates. We next aimed to identify the hypothetical α_vβ₃ integrin binding ligand. During fibroblast spreading on FN, adhesion clusters are initially rich in α₅β₁ integrins and gradually recruit α_vβ₃ integrins (Zamir et al., 2000), which then allows centripetal α₅β₁ integrin translocation and fibrillar adhesion formation (Pankov et al., 2000). Since the α₅β₁ integrin-selective substrates do not theoretically provide ligands for α_vβ₃ integrin, we reasoned that cellular FN secreted by ECs at adhesion sites could potentially serve this role.

Cellular FN, containing the alternatively spliced EDA domain (EDA FN), was not present in FAs at early time points (Fig. 7A; Fig. S7A). At later time points, EDA FN could be detected in the formation of fibrillar adhesions, and was sometimes present in the proximal part of mature FAs (Fig. 7A). Accordingly, when FN synthesis was blocked using the total protein synthesis inhibitor cycloheximide, ECs were still able to spread and formed enlarged, peripheral FAs where α_vβ₃ and α₅β₁ integrins co-localized (Fig. 7B; Fig. S7A). Interestingly, cycloheximide treatment inhibited fibrillar adhesions, suggesting that cellular FN production is required for their formation (Fig. 7B; Fig. S7A). Analogous results were obtained for fibroblasts, with cycloheximide treatment blocking the translocation of α₅β₁ integrin from FAs to fibrillar adhesions, and resulting in more pronounced integrin co-localization compared to control conditions (Fig. S7B). Taken together, our data indicate that the accumulation of α_vβ₃ integrins at newly assembled FAs on α₅β₁ integrin-selective substrates is not dependent on binding cell-secreted FN.

Our experiments with ECs described thus far were performed in complete medium, which contains 2% serum among other components. Therefore, we next examined the possibility that α_vβ₃ integrin is recruited to α₅β₁ integrin-based FAs after recruitment of soluble FN that is pulled from the culture medium to FAs, and serves as the ligand for α_vβ₃ integrin. Previous studies have demonstrated that α₅β₁ integrins, but not α_vβ₃ integrins, even when activated, can bind soluble FN and incorporate it in a fibrillar matrix (Huveneers et al., 2008), and that α₅β₁ integrins are responsible for FN transport at the leading edge of migrating cells (Sung et al., 2011). However, we found that ECs cultured in medium containing FN-depleted serum were still able to spread on α₅β₁ integrin-selective substrates, assemble FAs and recruit α_vβ₃ integrins (Fig. 7B). Moreover, soluble fluorescent FN (FFN) added to the culture medium during incubation of ECs on α₅β₁ integrin-selective substrates was absent from peripheral FAs and accumulated in fibrils assembled under the cells over time, suggesting that it is not the ligand that recruits α_vβ₃ integrin (Fig. 7C; Fig. S7D). Accordingly, immunolabeling against plasma FN did not stain FN on newly formed peripheral FAs at the ventral side of cells (Fig. S7C). Therefore, our data exclude serum FN as the soluble factor responsible for recruiting α_vβ₃ integrins to α₅β₁ integrin-based FAs.

We next performed experiments in the absence of media supplements and serum in order to test whether some soluble factor is required for α_vβ₃ integrin recruitment to α₅β₁ integrin-based FAs. Under these conditions, ECs were unable to efficiently spread on α₅β₁ integrin-selective substrates and exhibited a phenotype similar to that of ECs with α_vβ₃ integrin blocked (Fig. 7B). Nevertheless, a small fraction of ECs was still able to assemble sparse, thin and elongated FAs, which were positive for β₁, but not β₃ integrins (Fig. 7B; Fig. S8A). FA assembly in this subset of ECs presumably occurred due to lower cell contractility under serum-free conditions, which imposes reduced loads on the engaged α₅β₁ integrins. Accordingly, blocking of α_vβ₃ integrins under serum-free conditions did not abolish FAs, even though ECs were still unable to spread adherently over the substrate (Fig. S8A). Addition of FN to the medium did not elicit observable changes in FA or cell morphology, further supporting our conclusion that soluble FN is not responsible for α_vβ₃ integrin recruitment to α₅β₁ integrin-based FAs (Fig. S8A). Since complete medium contains supplements and growth factors besides serum, we tested cell spreading behavior of ECs cultured in media containing combinations of the different commercial components. ECs were able to spread and assemble mature FAs only when serum was present (Fig. S8B).

Another serum factor, present at high concentrations (300–600 μg/ml) in serum, which could serve as the ligand for α_vβ₃ integrins is vitronectin (VN). While α₅β₁ integrin cannot directly interact with VN, and hence recruit it to FAs, it is possible that it does so indirectly through complex formation. For example, the urokinase plasminogen activator receptor (uPAR, also known as PLAUR) can both associate with α₅β₁ integrin and bind VN (Ferraris et al., 2014; Wei et al., 2005, 1996). We tested VN localization on ECs spreading on α₅β₁ integrin-selective substrates in supplemented media using immunofluorescence (Fig. 7D) or fluorescently labeled VN (Fig. 7E) and observed no evidence of localization at FAs. Moreover, addition of soluble VN under serum-free conditions was not able to rescue FA maturation (Fig. S7E). Thus, our data indicate that VN is not responsible for α_vβ₃ integrin recruitment to α₅β₁ integrin-based FAs.

Serum could also exert its effects through inside-out activation of α_vβ₃ integrins. A recent report described the rapid reinforcement of adhesions initiated by α₅β₁ integrin binding to FN, through rapid α_vβ₃ integrin recruitment and focal adhesion kinase (FAK)-mediated signaling (Strohmeyer et al., 2017). Given that FAK activation preferentially occurs through β₁ compared to β₃ integrins (Costa et al., 2013; Shibue and Weinberg, 2009), we hypothesized that α_vβ₃ integrin recruitment to α₅β₁ integrin-based FAs might be triggered by FAK activation. Pharmacological inhibition of FAK activation using the specific inhibitor PF-573228 (Slack-Davis et al., 2007) did not impair FA assembly or β₁ integrin enrichment in FAs, but attenuated the enrichment of α_vβ₃ integrins as evidenced by optical microscopy, image analysis and quantification of enrichment ratio values (Fig. 8). Thus, in spite of lower α_vβ₃ integrin recruitment, FAs matured to a similar extent in FAK-inhibited ECs as in control cells, suggesting that even low amounts of α_vβ₃ integrin are sufficient for FA maturation. Notwithstanding the inherent limitations of inhibitor specificity, these data indicate a role for FAK-mediated signaling during α_vβ₃ integrin recruitment to adhesion clusters.

Our study highlights the utility of a well-defined, nano-patterned platform and of highly selective peptidomimetics to decouple individual integrin contributions and reveal an intriguing integrin crosstalk process at the early stages of adhesion cluster formation, namely the recruitment of α_vβ₃ integrins to α₅β₁ integrin-based FAs. Initial α₅β₁ integrin binding triggers the formation of adhesion clusters, which in the presence of serum rapidly, yet successively, recruit α_vβ₃ integrins, leading to FA maturation and efficient cell spreading. A similar process occurs with FN, which is initially bound to cells via α₅β₁ integrins; however, FN is able to bind both α_vβ₃ and α₅β₁ integrins in contrast to the substrates presented in our study. In the case for FN binding, an ‘integrin switch’ mechanism has been proposed, wherein α_vβ₃ integrins replace α₅β₁ integrins for binding to FN (Cao et al., 2017). It is therefore tempting to speculate that the processes responsible for recruiting α_vβ₃ integrins upon α₅β₁ integrin engagement are ‘hardwired’ in cell physiology and serve the purpose of stabilizing FAs, while enabling the translocation of α₅β₁ and consequent assembly of FN fibrils by the cells.

As opposed to adhesion on FN, cells initially adhering exclusively through α₅β₁ integrin on α₅β₁ integrin-selective substrates require the presence of serum to recruit α_vβ₃ integrins. When α_vβ₃ integrins were blocked with soluble ligands, cells were unable to spread cohesively, highlighting the functional role of this integrin in FA maturation on these nano-patterned substrates. We suggest two possible, not mutually exclusive, scenarios as the most likely to explain our results.

First, a yet-to-be identified factor in serum could accumulate at adhesion clusters as they mature, and then serve as a ligand for α_vβ₃ integrins. This hypothesis is consistent with our results from the α_vβ₃ integrin-blocking experiment (Fig. 5) and studies demonstrating that α_vβ₃ integrins promoted firm binding and retention in FAs, mediated by their extracellular part, while α₅β₁ integrins were more mobile (Rossier et al., 2012; Tsunoyama et al., 2018). In this work, we have excluded some obvious candidates for this hypothetical soluble serum factor, such as FN and VN; further work is required to test whether accumulation of a soluble α_vβ₃ integrin ligand from serum, directly or through complex formation, drives α_vβ₃ integrin recruitment to maturing FAs.

A second possibility we put forward is that soluble serum factors promote α_vβ₃ integrin activation through intracellular signaling mechanisms, following initial engagement of α₅β₁ integrin. Support for this hypothesis comes from a previous study implicating inside-out activation for α_vβ₃ integrin recruitment to FN (Strohmeyer et al., 2017), which is consistent with our preliminary findings that FAK inhibition attenuates α_vβ₃ integrin recruitment to FAs (Fig. 8). Importantly, a recent study demonstrated that a constitutively active α_vβ₃ integrin mutant that cannot bind ligands is nevertheless robustly recruited to adhesion clusters (Changede et al., 2019). In our study, it is unlikely that α_vβ₃ integrin remains unligated at FAs, since competition with a soluble ligand impairs adhesion formation (Fig. 5), but its accumulation to integrin clusters might be triggered by its inside-out activation. At these adhesion clusters, besides the possibility of binding a ligand recruited from serum as discussed above, it is conceivable that α₅β₁ integrin-mediated, inside-out activation of α_vβ₃ integrins and its high local concentration due to recruitment combine to eventually enable α_vβ₃ integrin binding to the α₅β₁ integrin-selective ligand. Indeed, the affinity of other integrin family members (α₅β₁, α₄β₁) towards the ligand increases hundreds- to thousands-fold upon integrin conformational changes occurring during their activation (Li and Springer, 2018; Li et al., 2017). Of note, however, is that α_vβ₃ integrin in absence of α₅β₁ integrin-mediated binding does not support cell adhesion on α₅β₁ integrin-selective substrates (Fig. S2B) (Rechenmacher et al., 2012). Future studies utilizing β₃ integrin mutants that cannot bind RGD ligands (e.g. D119Y) or that constitutively activate the integrin (e.g. N305T) (Cluzel et al., 2005), or their combination, as subsequently described in double mutants (Changede et al., 2019), should shed more light on the mechanism of β₃ integrin recruitment.

Notably, the type of crosstalk between α₅β₁ and α_vβ₃ integrins we have described is not bi-directional and any proposed mechanism must be consistent with the fact that α_vβ₃ integrin-based FAs do not recruit α₅β₁ integrin FAs, and are able to sustain FA assembly on their own. Intriguingly, a recent study (Kalappurakkal et al., 2019) showed that formation of plasma membrane nanodomains occurred exclusively on FN-coated substrates, via β₁ but not α_v integrins, supporting our findings, and raising the possibility that α_vβ₃ integrin recruitment is associated with changes in the local lipid microenvironment. Such accumulation of α_vβ₃ integrins could occur through diffusion at the plasma membrane (Rossier et al., 2012) or directed exocytosis at FAs (Huet-Calderwood et al., 2017).

Examined under the intense interest for selective integrin targeting to limit or induce angiogenesis depending on the context (e.g. tumor angiogenesis versus tissue engineering scaffold angiogenesis) (Mas-Moruno et al., 2010), our results point to the need to consider such crosstalk in intervention design. The initial steps of cell anchoring to a substrate, whether it be their ECM or an engineered biomaterial or scaffold, can dictate long-term cell behavior, even in the presence of the ensuing active remodeling of the microenvironment. Viewed in this respect, our data indicate that blocking α_vβ₃ integrins could have a destabilizing effect on EC adhesion and argue for the need to move towards a more global, mechanistic understanding of how specific integrins regulate EC adhesion.

A list of commercially available chemicals and antibodies used in this study can be found in Tables S2 and S3. A cyclic RGDfK peptide (c-RGDfK) functionalized with a thiol group, and integrin-selective, peptidomimetic ligands against α₅β₁ and α_νβ₃ integrins were synthesized as previously described (Kapp et al., 2017; Rechenmacher et al., 2012) and their structures are presented in Table S1. Fluorescent FN was prepared using plasma FN (Life Technologies, 33010018) and an Alexa Fluor 488 (Thermo Fisher Scientific, A10235) or Alexa Fluor 568 labeling kit (Thermo Fisher Scientific, A10238). Fluorescent vitronectin was prepared using human serum vitronectin (Life Technologies, PHE0011) and an Alexa Fluor 568 labeling kit (Thermo Fisher Scientific, A10238). Poly(ethylene glycol) diacrylate with molecular weight of 10,000 was synthesized as previously described (Aydin et al., 2010). FN-depleted serum was prepared as previously described (Missirlis et al., 2017). The absence of lower molecular weight FN fragments in FFN solution or FN-depleted serum was confirmed using western blot analysis (Fig. S7F–I). Polystyrene-block-poly(2-vinylpiridine) (PS-b-P2VP) block copolymers of different molecular weights were used as per the manufacturer's instructions (Polymer Source, Canada).

Nano-patterned gold particles on glass coverslips were prepared using block copolymer micellar lithography (BCML) as previously described (Glass et al., 2003). In brief, PS-b-P2VP were dissolved in toluene and assembled into uniform inverse micelles. HAuCl₄ was added to the micellar suspension and partitioned in the micelle cores after magnetic stirring for >1 day. A monolayer of micelles was then deposited on piranha solution-treated [3:1 (concentrated H₂SO₄):(30% H₂O₂)] glass coverslips either by dip-coating or spin-coating of the micellar suspension. Reduction of the gold and removal of the polymer was carried out by means of hydrogen plasma treatment (0.4 mbar H₂, 600 W, 45 min) with a P210 microwave plasma system (PVA TePla Americana, Corona, USA). The area between the nano-patterned gold particles was passivated using triethoxysilane PEG [molecular weight (MW) of 2000] as previously described (Blümmel et al., 2007). Substrates were kept in a nitrogen atmosphere and used within 1 day following passivation.

For the preparation of nano-patterned hydrogels we adapted a previously published procedure from our group (Aydin et al., 2010). Nano-patterned glass coverslips were prepared as described above, except that (1) a milder plasma treatment was used [0.4 mbar, W7 gas (93% Ar, 7% H₂), 200 W, 45 min] in order to allow for transfer to the hydrogel surface, and (2) the passivation step was omitted. Nano-patterned glass substrates were treated with N,N′-bis (acryloyl)cystamine solution (2 mM in ethanol) for 1 h and washed three times in ethanol to provide anchoring to the hydrogel and ensure the transfer of the pattern. The functionalized nano-patterns served as the top surface of a ‘sandwich’ setup for hydrogel formation. APTES [(3-aminopropyl)triethoxysilane] functionalized glass coverslips served as the bottom surface to ensure coupling to the hydrogel. The middle layer contained the hydrogel precursor consisting of 100 μl of PEG-DA in ultrapure water, 0.35 μl APS (10% w/v) and 0.05 µl TEMED per mg PEG-DA. The distance between the two coverslips was fixed at 1 mm using custom-made Teflon holders and the cross-linking reaction was left to proceed for 30 min. The entire ‘sandwich’ assembly was then placed in ultrapure and sterile water at 4°C for 48 h to allow for swelling of the hydrogels and detachment of the upper coverslip.

The gold nanoparticles immobilized on the nano-patterned substrates were functionalized using thiol-gold chemistry. Thiol-modified c-RGDfK or integrin-selective antagonists presenting a thiol group were incubated with substrates at a concentration of 25–100 µM in MilliQ water for 1–2 h at room temperature. Substrates were thoroughly rinsed with MilliQ water and then with PBS to remove unbound ligands. Functionalized substrates were used for cell experiments immediately after washing. Glass coverslips coated with 10 μg/ml FN overnight at 4°C, washed and blocked with 1% BSA served as controls.

For non-patterned substrates, glass coverslips were initially coated with silane–PEG–maleimide. 20×20 mm coverslips were activated with O₂ plasma (100 W, 0.4 mbar) for 5 min and then were immediately immersed in anhydrous toluene under inert (N₂) atmosphere, in which a small amount of silane–PEG–maleimide was dissolved and a few drops of triethylamine added. The solution was heated at 80°C and left overnight. Coverslips were then washed with ethyl acetate and methanol, and dried with a gentle N₂ flow. Thiol-containing, integrin-selective antagonists at a concentration of 25 μM in PBS were incubated with surfaces for 1 h, washed with PBS three times and used for cell experiments within 2 h.

Nano-patterned structure dimensions were analyzed by scanning electron micrographs recorded on a field emission electron microscope (Zeiss Leo 1530, Zeiss SMT, Oberhocken, Germany). The inter-particle distances and the order of the hexagonal patterns were determined by analyzing the micrographs with a custom-made ImageJ plugin (available upon request). The successful transfer of the gold nano-pattern from glass to the respective hydrogel was validated by cryo-scanning electron microscopy (SEM) imaging using a Zeiss Ultra 55 field emission electron microscope. Cryo-SEM was performed under low temperature conditions (T_op=−130±5°C) and low acceleration voltages of 1.0–1.5 kV due to the low conductivity of samples.

Nano-patterned hydrogels were mechanically characterized by indentation measurements using a Nano-Wizard III atomic force microscope (JPK Instruments AG, Germany). Cantilevers with a spherical, borosilicate glass probe (sQube) of 5 µm in diameter were used. Cantilever spring constants were determined using the thermal noise calibration method and ranged between 0.45 N/m and 0.60 N/m. Young's moduli were derived by fitting force–distance curves with a Hertz model using the free software AtomicJ (version 1.7.2) (Hermanowicz et al., 2014) and assuming a Poisson ratio of 0.5. Indentation measurements at three different locations per sample, with two samples per experiment and three independent experiments were analyzed.

Primary human umbilical vein endothelial cells (HUVEC), pooled from different donors, were purchased from Promocell (C-12203) and cultured according to the instructions provided as sub-confluent monolayers at 37°C and 5% CO₂. HUVEC were cultured in supplemented endothelial cell growth medium (Promocell, C-22010), which consists of endothelial cell basal medium (Promocell, C-22110), supplemented with 2% fetal calf serum, 0.4% endothelial cell growth supplement, 0.1 ng/ml epidermal growth factor, 1 ng/ml basic fibroblast growth factor, 90 μg/ml heparin and 1 μg/ml hydrocortisone, and which we refer to as complete medium. Cells were detached using the Detach kit (Promocell, C-41200). HUVEC were used until passage 8.

Primary human dermal fibroblasts (pHDF) were purchased from ATCC (PCS-201-010) and cultured as sub-confluent monolayers at 37°C and 5% CO₂, in Dulbecco's modified Eagle's medium (DMEM; Life Technologies, 10938), supplemented with 10% fetal bovine serum (FBS) and 1% penicillin/streptomycin. pHDF were used until passage 12. HUVEC and pHDF cultures were routinely checked for the absence of mycoplasma.

Surface integrin expression of HUVEC was analyzed by flow cytometry. HUVEC (5×10⁵ cells) were suspended using trypsin and left to recover in serum-free medium for 10 min at 37°C and 5% CO₂, prior to incubation with monoclonal antibodies (100 μl at a concentration of 10 μg/ml) against α₅β₁ (clone JBS5) or α_νβ₃ (LM609) for 30 min at 4°C. Cells were washed twice with serum-free medium and incubated for another 30 min with Alexa-Fluor 674-conjugated anti-mouse secondary antibody (100 μl; 5 μg/ml). Cells incubated only with the secondary antibody served as controls. After washing twice with serum-free medium, cells were analyzed on a LSRFortessa flow cytometer (BD Biosciences).

The efficiency and specificity of integrin-selective substrates for cell adhesion was assessed using integrin blocking experiments. HUVEC (6.7×10³ cells) were incubated in absence or in presence of integrin antagonists for 15 min on ice, prior to seeding on α₅β₁- or α_νβ₃-selective substrates in 6-well plates. After 3 h, ECs were rinsed three times with PBS to remove non-adherent cells and fixed using PFA. The number of cells was manually counted at five random spots over the substrate and cell density was expressed as percentage adhesion relative to the seeding density.

The efficiency of HUVEC adhesion on FN-coated glass was measured assessed 30 min after seeding. Briefly, suspended cells were kept under ice for 10 min, in presence or absence of integrin-selective antagonists, before being seeded in 96-well plates coated with 1 μg/ml FN. After 30 min, wells were washed twice with ice-cold PBS, the solution was aspirated and plates were placed at −80°C overnight. Relative cell numbers were quantified using the Cyquant cell proliferation assay kit (Life Technologies).

Cells were seeded on integrin-selective substrates for 30 min and then incubated with 25 μM blebbistatin (in DMSO) for another 1 h in order to inhibit myosin II. For FAK inhibition experiments, cells were incubated 10 min prior to and during cell spreading with 10 μM PF-573228. DMSO-treated cells served as controls in both cases. For the blebbistatin wash-out experiment, cells were washed once with PBS, fresh medium added and cells were fixed after predefined time points. For the α_vβ₃ integrin blocking experiment, cells were incubated with the α_vβ₃ ligand for 15 min under ice before cell seeding. For serum-free experiments, non-supplemented endothelium cell basal medium (Promocell, C-22210) was used. To block endogenous protein synthesis, cells were incubated with 50 μg/ml cycloheximide 1 h before seeding and also during spreading on substrates.

HUVEC or pHDF cells (1×10⁴ cells/cm²) cultured on substrates under indicated conditions and for predetermined time periods were washed with PBS and fixed with 4% paraformaldehyde (PFA) for 15 min at room temperature. Membranes were permeabilized with Triton X-100 (0.1%) followed by blocking with BSA (3% in PBS). Primary antibodies (in 1% BSA) were incubated for 1–2 h at room temperature or overnight at 4°C, and Alexa Fluor-labeled secondary antibodies were incubated for 1 h at room temperature. DAPI and phalloidin were used to stain nuclei and filamentous actin (F-actin), respectively. Epifluorescence microscopy images were acquired using a Delta Vision (DV) system (Applied Precision Inc.) on an Olympus IX inverted microscope, equipped with a cooled CCD camera and a 60×/1.4 NA oil-immersion objective (Olympus) or a 20×/0.3 NA air objective (Olympus). Confocal microscopy images were acquired on a Zeiss LSM 880 laser scanning confocal microscope using a 63×/1.4 NA oil-immersion objective (Zeiss) or a 20×/0.8 NA objective (Zeiss).

Cell projected area, aspect ratios and convexity values (convex hull perimeter divided by cell perimeter) were determined through image analysis of phalloidin-stained cells using the ImageJ software. Focal adhesion morphology analysis was performed using a custom-made ImageJ plugin (Missirlis et al., 2016). Briefly, immunofluorescence images were first background-corrected using a rolling ball filter (diameter of 25 pixels) and then smoothed using a Gauss kernel with a 5-pixel radius and standard deviation of 1 pixel. Adhesions were identified as bright pixels after applying automatic threshold using Otsu's method. An area threshold of 0.4 μm² was set to exclude small adhesions and noise. Bright spots were localized and used as binary masks for calculating sum intensities from the original images.

Statistical analyses were performed using Prism (GraphPad Inc.). Experimental data were analyzed using either unpaired t-tests or one-way ANOVA with Tukey's post-hoc analysis, unless otherwise noted. The middle line in box plots indicates the median, the box indicates the interquartile range, the whiskers the 5th and 95th percentiles and the cross the mean. Column plots represent mean±s.e.m. *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001; n.s., not significant.

We thank Joachim P. Spatz at the MPI for Medical Research, Heidelberg, for useful discussions, for hosting this project and providing free access to the facilities in his lab.