Integrin nanoclusters can bridge thin matrix fibres to form cell–matrix adhesions

Abstract

Integrin-mediated cell–matrix adhesions are key to sensing the geometry and rigidity of extracellular environments and influence vital cellular processes. Invivo, the extracellular matrix is composed of fibrous arrays. To understand the fibre geometries that are required for adhesion formation, we patterned nanolines of various line widths and arrangements in single, crossing or paired arrays with the integrin-binding peptide Arg-Gly-Asp. Single thin lines (width ≤30 nm) did not support cell spreading or formation of focal adhesions, despite the presence of a high density of Arg-Gly-Asp, but wide lines (>40 nm) did. Using super-resolution microscopy, we observed stable, dense integrin clusters formed on parallel (within 110 nm) or crossing thin lines (mimicking a matrix mesh) similar to those on continuous substrates. These dense clusters bridged the line pairs by recruiting activated but unliganded integrins, as verified by integrin mutants unable to bind ligands that coclustered with ligand-bound integrins when present in an active extended conformation. Thus, in a fibrous extracellular matrix mesh, stable integrin nanoclusters bridge between thin (≤30 nm) matrix fibres and bring about downstream consequences of cell motility and growth.

Attachment of cells to the extracellular matrix (ECM) in fibrous in vivo environments is imperative for cell functions that regulate growth, differentiation and disease^1,2. Adhesions are mediated by transmembrane heterodimeric integrin receptors that bind ECM fibres such as collagen, laminin or fibronectin³. Unlike large focal adhesions on glass, cell-matrix adhesions are punctate in vivo^4,5, possibly due to the thin ECM fibres (~5–20 nm) present there^6,7. The formation of cell–matrix adhesions on fibrous substrates is poorly understood. We sought to understand both, the minimal fibrous ligand geometry that supports adhesion formation and the role of ligand density versus geometry in adhesion formation.

A minimal ligand density of four to seven nanodots spaced within ~110 nm and functionalized with the Arg-Gly-Asp (RGD) peptide will support cell spreading^8–11, except on very soft substrates¹², possibly through the assembly of clusters of adhesion proteins^13,14. The mechanism of how minimal dot arrays support adhesion formation is not understood. There are two major possibilities: single ligand-bound integrins can activate downstream signals that are additive in small regions, or coclusters of both ligand-bound and unbound integrins function as an adhesion signalling platform. The latter offers advantages, because coclusters increase the avidity of ligand–integrin binding and can amplify the response to low ligand density in a threshold-based manner^15–17. At the cytoplasm–membrane interface, nanoclusters can integrate multiple weak cytoskeletal interactions to form strong adhesions over a wide range of extracellular traction forces¹⁸. Thus, we hypothesize that coclustering of liganded and unliganded integrins aids adhesion formation.

Nascent adhesions^19,20 of ~110 nm (refs. ^21–23) catalyse adhesion maturation based on mechanotransduction from the matrix^24–26. Since mature adhesions are composed of diffraction-limited clusters of integrins^13,14 and support high tension, there is concern about which matrix geometries can support adhesion formation. For example, adhesions do not form on rigid substrates when RGD ligands are separated by >110 nm (ref. ⁸), wherein a ~110-nm modular integrin nanocluster cannot link the RGD dots. Furthermore in adhesions, chimeric integrins lacking ligand binding appear to associate with adhesions^27–29 and crosslinked integrin tails could aid in transmitting tension from the cytoskeleton to the matrix^30,31. However, unliganded integrins in the clusters could be ablated from the clusters by cytoskeletal forces in certain ligand geometries. Particularly in the case of matrix fibres, the geometry of fibre arrays that will support adhesions is not understood.

To probe the geometric organization of ECM fibres required for the formation of cell–matrix adhesions, we nanopatterned lines of titanium (Ti) or gold-palladium (AuPd) mimicking one-dimensional (1D, single line width ≤30 nm) or two-dimensional (2D) geometries²² to alter single geometric parameters individually.. Photo-activated localization microscopy (PALM), revealed integrin nanoclusters containing activated integrins only on 2D substrate geometries where they spanned individual lines. Talin served as an important integrin crosslinker²². Short-lived, sparse aggregates formed on 1D geometries that did not support cell spreading. Thus, fibres are needed to cover a 2D area (>40 nm) to support the force-dependent maturation of adhesions.

Results

Adhesion clusters require 2D ligand geometry for stability.

To determine the geometries of single matrix fibres or fibre networks that support adhesion formation, we used electron-beam lithography to nanopattern Ti or AuPd nanolines with various line widths in single, crossing or paired arrays (Figs. 1a,b and 2a–c). Since the integrin head was ~8 nm (refs. ^32–34), 10-nm lines functioned as linear arrays of single integrin binding sites, mimicking small fibres in vivo (5–20 nm). Because integrin nanoclusters were ~110 nm in size²³, single lines spaced >110 nm should not have been bridged by single clusters and were considered as 1D patterns. However, single lines in paired (within 110 nm) or crossing arrays were considered the minimal 2D fibre arrangement. In addition, wide lines (>30 nm) accommodated multiple integrins in the line width and were thus 2D. Ti was selectively functionalized with cyclic RGD using phosphoinositol 3,5-bisphosphate and biotin-neutravidin chemistry (Fig. 1b), and AuPd was selectively functionalized with cyclic RGD using thiol and biotin-neutravidin chemistry (Fig. 2c)^35,36. In all cases, the glass substrate was passivated using 1,2-dioleoylsn-glycero-3-phosphocholine (DOPC) lipid bi-layers to prevent non-specific ligand binding^37,38 (Supplementary Fig. 1a). To assay the formation of cell–matrix adhesions, either mouse embryonic fibroblasts (MEFs) or human foreskin fibroblasts (HFFs) were plated on patterned substrates for 30 min. On the same coverslips, cells bound to regions of continuous RGD (positive control) and formed larger adhesions (Supplementary Fig. 1b,c). Our studies focused on early cell spreading when β₃ formed adhesions whereas β₁ was recruited only to adhesions at later time points (Supplementary Fig. 1d).

Fig. 1 |. Cell–matrix adhesions form across closely spaced nanofibre mimetic substrates and trigger cell spreading.

a, Scanning electron micrograph (SEM) showing Ti-patterned nanolines in single, crossing or paired arrays. Nanopatterns are systematically varied and named by the geometric parameters of features (from left to right) in a unit. In single lines, the first number is line width and the second is edge-to-edge spacing. In cross-lines, two single-line arrays of 10–490 nm are crossed at an angle of 25° to each other. In line pairs, the first and third numbers are line width, the second is internal edge-to-edge spacing between the two lines in a pair, the fourth is edge-to-edge spacing between adjecent line pairs. b, Super-resolution confocal image of RGD-functionalized substrates of the features shown in a. Inset boxes show enlarged views of nanolines regions depicted in a. c–f, Super-resolution confocal images of MEFs spreading on RGD-functionalized nanopatterns: paxillin (c), actin (phalloidin, d), merged images (e) and enlarged views of c (f). Dashed lines denote cell edges. g, Cell area and h, total focal adhesion (FA) area normalized to total cell area for various nanopatterns. In all box plots of interquartile distance, lines represent the median and the whiskers are standard deviation (n ≥ 7 cells from three independent experiments, one-sided analysis of variance test, P < 0.00001). Sample sizes are summarized in Supplementary Table 1.

Fig. 2 |. The protein FAK is phosphorylated to a greater extent on 2D than on 1D patterns.

a, SEM showing a single feature of AuPd patterned substrates: hexagonal dot pattern (10-nm dots separated by 40-nm centre-to-centre spacing), single line (10-nm width) and line pair (10-nm lines separated by 80-nm centre-to-centre spacing). b, Pseudo-colour SEM of features before lift-off during the patterning process. Inset boxes show enlarged regions depicted in a. c, Super-resolution confocal image of RGD-functionalized substrates. Inset boxes show enlarged regions depicted in b. d, Integrated fluorescence intensity (arbitrary units, a.u.) over 25 μm²; n ≥ 15 independent measurements by t-test). e, Cell spreading area of HFF (n ≥ 17 by t-test). f, Pseudo-colour image of HFF spread on RGD-functionalized nanopatterns for 30 min and immunolabelled with pFAK. Dashed lines denote cell edges. g, Number of immunolabelled pFAK clusters per μm² of the lamellipodial region, recorded 5 μm from the cell edge with cells spreading on various RGD-functionalized substrates (n ≥ 9 cells by t-test). h, Pseudo-colour images of HFF: pFAK immunolabelling (green) and RGD functionalized nanopattern (magenta), and merged image of hexagonal dot pattern. i, Normalized fluorescence intensity of RGD and pFAK through a linear region of interest. j, Pseudo-colour images of HFF on single line and line pair patterns. k, Correlation coefficient of pFAK and RGD nanopatterns (n ≥ 7 cells, P values acquired by t-test). In f,h,j, the double-ended arrows represent the orientation of the fibre mimetic substrate, dashed lines denote the cell edges. In all box plots of interquartile distance, the line represents the median and the whiskers represent s.d. The sample sizes are summarized in Supplementary Table 1.

Single 10-nm lines (1D geometry) separated by either 500 or 250 nm (centre-to-centre; approximately five- or 2.5-fold larger than single-integrin nanoclusters) (Fig. 1c–h) did not support cell spreading, actin stress fibre formation or paxillin recruitment, similar to single lines of 20 or 30 nm separated by 500 nm (Supplementary Fig. 2a–c). However, with line pairs internally separated by 50 or 80 nm (centre-to-centre distance between pairs was 500 nm, and hence ligand density was equivalent to single lines separated by 250 nm), cells spread significantly, formed focal adhesions as observed by paxillin (Fig. 1c–h and Supplementary Fig. 2c) or talin recruitment (as below, Supplementary Fig. 9), and ventral stress fibres of actin (Fig. 1d,e). Magnified images show adhesions formed along the direction of the lines (Fig. 1f). With single 10-nm lines separated by 160 nm, or wide lines (width 40 nm), cells spread to a certain degree and very few small adhesions formed (Fig. 1c–h and Supplementary Fig. 2c). To test whether force was required to form adhesions on parallel lines, we treated MEFs spread on nanopatterned substrates with the Rho kinase inhibitor, Y27632, for 30 min; far fewer and significantly smaller adhesions remained (Supplementary Fig. 3a–c), indicating that force was required to maintain the adhesions. Taken together, this indicated that ligands presented in single lines <30 nm were not sufficient for adhesion formation but that multiple fibres within 110 nm could support tension-dependent, cell–matrix adhesions.

To mimic a fibrous mesh, we nanopatterned two 10-nm line arrays separated by 500 nm that crossed each other at an angle of 25° (Fig. 1a,b). This heterogeneous geometry had the unique feature that it mimicked the ligand density of single 10-nm lines separated by 250 nm; however, at the crossing points, the lines formed a 2D ligand array. Although single lines separated by 500 or 250 nm did not support the formation of adhesions, crossed lines with 2D geometry supported adhesion formation, actin stress fibre assembly and cell spreading (Fig. 1c–h and Supplementary Fig. 2c).

Next, to explore the effects of local ligand geometry versus global ligand density on adhesion formation, we used representative ligand arrays of 1D (single lines), 2D (line pairs internally separated by 80 nm) and hexagonal dot patterns (spaced by 40 nm) that enabled cell spreading^8,9 (positive control) (Fig. 2a–c). The integrated ligand density for single lines was higher than that for hexagonal dot patterns (Fig. 2d and Supplementary Fig. 4a,b). HFFs spread on hexagonal dot or line pair substrates to an extent similar to MEFs, and considerably more than on single lines (Fig. 2e). To test for downstream signalling on all substrates, we immunolabelled phosphorylated FAK (pFAK, Y397) nanoclusters (Fig. 2f,g and Supplementary Fig. 5a) and imaged these with super-resolution confocal microscopy (a twofold increase in resolution enabled us to resolve individual adhesion clusters). On line pairs and cross-lines, phosphoFAK cluster density in the lamellipodium was comparable to that in control cells spreading on continuous glass substrates or hexagonal dot patterns (Fig. 2f,g and Supplementary Fig. 5a). These pFAK nanoclusters overlapped with the ligand (Fig. 2h–k). Far fewer clusters formed on the single lines (10 nm, Fig. 2f,g or 20 nm, Supplementary Fig. 5b,c), indicating that pFAK signalling was significantly reduced. We noted the alignment of adhesions on the path of actin flow from the cell edge and orthogonal to the line pairs, emphasizing that a spacing of 500 nm did not hinder the formation of mature adhesions (Fig. 2j and Supplementary Fig. 5d). Thus, FAK was activated preferentially on the 2D substrates despite higher ligand density on the 1D lines compared to the hexagonal dot patterns.

To compensate for possible lower integrin activation on single lines, we treated the MEFs with manganese chloride (2 mM MnCl₂), a potent activator of α_vβ₃ integrin³⁹, and spread them on single lines and line pairs for comparison. This was not sufficient to form integrin or pFAK clusters on single lines (Supplementary Fig. 6a,b). Although manganese activated integrin and in the presence of excess ligand on single lines, cells did not form adhesions.

To determine whether reduced focal adhesion and FAK signalling on single lines were correlated to reduced integrin cluster size, we measured β₃-photo-activated-green fluorescent protein (PAGFP)⁴⁰ clustering using PALM microscopy. As red fluorescence was quenched by the metals, PAGFP was selected. MEFs expressing β₃-PAGFP were spread on nanopatterned substrates for 30 min and fixed. Control experiments were repeated to render the interpretation comparable to previous results22. On polylysine-coated glass (absence of ligand, negative control), only very few, sparse, random aggregates were observed (consistent with free diffusion of unbound integrin⁴¹, full width at half-maximum (FWHM) 73 ± 17 nm, mean ± s.d.; Fig. 3a,b and Supplementary Fig. 7f). These sparse, random aggregates served as a control to set the lower threshold for defining a biological cluster on substrates presenting ligands. Many large, dense clusters formed on the two positive controls, hexagonal dot patterns (114 ± 11 nm,) and continuous RGD (108 ± 12 nm; Fig. 3a,b), and were similar to those on the line pairs (112 ± 14 nm; Fig. 3a,b and Supplementary Fig. 7f). On single lines, fewer smaller clusters were observed (93 ± 18 nm).

Fig. 3 |. Single-fibre mimetics do not support robust integrin cluster formation.

a, PALM image of MEFs expressing β3-PAGFP spreading on various substrates. Dashed lines denote cell edges and inset boxes show enlarged regions. Double-ended arrows represent the orientation of the fibre mimetics—one for single line and two for line pair. Scale bar, 5 μm (enlarged, 1 μm). b, Size of FWHM of clusters. c, Relative numbers of molecules within the cluster on various substrates (n ≥ 873 clusters from four or more cells, by t-test). In all box plots of interquartile distance, lines represent the median, dots represent the mean and whiskers are s.d. The sample sizes are summarized in Supplementary Table 1.

We next used PAGFP to count the relative number of β₃ integrins per nanocluster (to account for endogenous unlabelled β₃ and error due to incomplete folding of PAGFP). Nanoclusters on hexagonal dot patterns and the continuous RGD-coated glass contained a significantly higher number of β₃ (35 ± 13 and 35 ± 10 molecules, respectively) relative to the aggregates on polylysine (14 ± 9) (Fig. 3c). On line pairs, the relative number of molecules in the clusters (34 ± 21) was similar to positive controls (Fig. 3a,c) whereas on single lines, the few evident clusters had significantly fewer integrins (19 ± 11). Thus, even though the density of ligands on 10-nm lines was greater than on hexagonal dot patterns, the clusters on 10-nm lines were significantly smaller and sparse.

Adhesions form less frequently and are transient on single lines.

To determine whether adhesion assembly, persistence or both was decreased on single lines, we tracked paxillin-mApple-tagged adhesions on both line geometries. Paxillin localized to adhesion clusters on both single lines and line pairs, but far fewer clusters formed on single lines (Fig. 4a, Supplementary Fig. 7a and Supplementary Video 1) than on line pairs (Fig. 4c, Supplementary Fig. 7a and Supplementary Video 2). Time-lapse imaging revealed that aggregates on single lines were transient (<100 s; Fig. 4e,i) but, on line pairs, most clusters persisted for longer than 300 s, the total time of imaging (Fig. 4g,i). Thus, more adhesions formed on line pairs and they persisted much longer than on single lines.

Fig. 4 |. Line pairs support more stable adhesion nanoclusters than single lines.

a–d, MEFs spreading on single lines or line pairs, with paxillin-mApple (a,c) and β3-PAGFP (b,d) photoswitched in several horizontal regions of interest. e–h, Kymographs of a line across adhesion clusters depicting an equivalent image. From the top to the first line, each pixel represents 1-s intervals for 120 s; from the first line to the second, each pixel represents 5-s intervals for 120 s; and from the second line to the bottom, each pixel represents 10-s intervals for 60 s. Scale bars, 10 μm (a–d); 5 μm (e–h). i,j, Frequency plots showing the persistencee of clusters marked with paxillin-mApple (i) or β3-PAGFP (j), on single lines (magenta) or line pairs (green); n ≥ 111 clusters from four or more cells. The sample sizes are summarized in Supplementary Table 1.

Live cell imaging of β₃-GFP (Supplementary Fig. 7b,c) showed diffusive integrin GFP on single lines (Supplementary Fig. 7b and Supplementary Video 3), whereas on line pairs, stable, large adhesions assembled (Supplementary Fig. 7c and Supplementary Video 3), as visualized in the kymographs (Supplementary Fig. 7d,e). To further investigate integrin cluster lifetimes, we recorded photo-activated β₃-PAGFP along the substrate lines (Fig. 4b,d). Kymographs revealed only small numbers of dynamic β₃ integrins on single lines (average lifetime ~30 s) but stable adhesion clusters on line pairs (average lifetime >300s) (see Fig. 4f,h,j). Taken together, these data show that transient integrin clusters formed on single lines whereas stable clusters formed on line pairs.

Unliganded integrins assemble in clusters with ligand-bound integrins.

Since the integrin extracellular domain was ~20 nm in length⁴², single integrins could not span both hexagonal dot patterns and line pairs. Thus, unliganded integrins appeared to bridge the nanodots or line pairs as no evidence of inhomogeneity was observed in the integrin distribution in PALM images. To test whether unliganded integrins could assemble in clusters, RGD-functionalized supported lipid bi-layers were used (see Methods), where adhesion clusters were previously observed to be long-lived and two-colour PALM imaging of integrin and ligand was possible^22,43. Clustering of unliganded integrins was tested with the integrin point mutant β₃-D119Y, which does not bind the RGD ligand (Fig. 5a), and mutant β₃N305T in a constitutively activated form. Both were labelled with mEos2. Furthermore, we developed a third β₃ double-mutant, β₃D119YN305T, which was activated but unable to bind the ligand. To test ligand binding of these mutants we expressed them in Chinese hamster ovary (CHO) cells that express very low levels of endogenous β₃ integrin that is sufficient for cell binding but does not support cell spreading or fomation of stress fiber anchored large focal adhesions⁴⁴ (Supplementary Fig. 9a,b). On expression of wild-type (WT) β₃ or the β₃N305T integrin, the CHO cells spread to a larger area and organized actin stress fibres (Supplementary Fig. 8a,b) but neither β₃-D119Y- nor β₃D119YN305T-expressing cells spread or formed large adhesions (Supplementary Fig. 8a,b). Thus, double-mutant D119YN305T phenocopied the single D119Y mutant, providing genetic evidence that it could not bind ligand.

Fig. 5 |. Activation is sufficient and ligand binding is not necessary.

a, Mutants of β3 integrin in the extracellular domain: D119Y, which cannot bind the RGD ligand; N305T, which is constitutively active; and double-mutant D119YN305T, respectively. b, Schematic of cells forming adhesions on RGD-functionalized supported lipid bi-layers or RGD-functionalized nanoline pairs, respectively. c, MEF cells expressing mutant β3 and endogenous β3 spreading on an RGD-functionalized supported lipid bi-layer. Two-colour PALM image of mutant β3 mEos2 and RGD labelling the endogenous integrin clusters. d, Correlation coefficient of β3 mutant and endogenous β3–RGD clusters (n > 10 cells, t-test). In the diamond plots of interquartile distance, the lines represent the median, the dots the mean and the whiskers the s.d. e, Enlarged two-colour PALM image of a clustered region of D119YN305T. f, Fluorescence intensity plot of the line region of interest in e. g, MEFs expressing β3D119YN305T spread on 10–70-10–410 nm line pairs: PAGFP (green), RGD (magenta) and merged image, respectively. h, Line plot across cluster region showing β3-PAGFP (green) and the line pair (magenta). In c,g, dashed lines denote cell edges and inset boxes show enlarged regions.

β₃ mutants were expressed in MEFs with endogenous β₃, and the cells were spread on RGD-functionalized lipid bi-layers for 15 min. In the mutants that did not bind ligand, the ligand clusters formed by the endogenous integrins only. This allowed us to assay whether non-ligand binding integrin coclustered with ligand bound integrin. The non-ligand-binding β₃-D119Y integrin did not colocalize with the endogenous integrin nanoclusters (Pearson’s correlation coefficient (PCC) = 0.2; Fig. 5 c,d); however, the constitutively activated, ligand-binding N305T integrin colocalized with the ligand clusters (PCC = 0.7; Fig. 5c,d). This also confirmed that the fluorescent tag did not cause exclusion from the adhesion clusters. However, the β₃ D119YN305T-mEos2 double-mutant colocalized with endogenous adhesion clusters (PCC = 0.8; Fig. 5c,d). Furthermore, this was often present around the RGD, implying that unliganded integrins were recruited to clusters formed by ligand-bound integrins (Fig. 5e,f).

These findings indicated that unliganded, but activated, integrins were recruited to line pairs, particularly between them. Two-colour PALM imaging of β₃ D119YN305T-PAGFP, simultaneously with DyLight 650-tagged Ti line pairs (see Methods), revealed that β₃D119YN305T-PAGFP was indeed present in nanoclusters bridging the lines (Fig. 5g,h). Thus we postulated that, on the 1D substrates, unliganded integrins at the edges of the nanoclusters were removed by the actomyosin machinery⁴⁵, destabilizing the clusters (Fig. 6), unless integrins were anchored in two dimensions.

Fig. 6 |. Proposed model for assembly of adhesion nanoclusters.

a,b, On single lines (a), even in the presence of ligands, only a few sparse, transient, adhesion nanoclusters assembled and could not support cell spreading, suggesting that single, thin ECM fibres are not sufficient for cell spreading; on line pairs (b), stable, dense adhesion nanoclusters assembled, composed of ligand-bound and unliganded integrins. These adhesion clusters assembled into larger focal adhesions, implying that integrin clustering and 2D ligand geometry are both required for the formation of nanoclusters of integrins. Cell cross-sections are depicted.

To determine whether the cytoplasmic adhesion protein talin affected cluster size, we used talin-depleted MEFs (talin1^–/– with talin 2 knockdown) and expressed GFP-tagged talin 1 full-length, talin rod domain or the IBS2 mutant in the rod domain (Supplementary Fig. 9). Talin-depleted MEFs, and those expressing the IBS2 rod mutant only, spread to a lesser extent and had very few adhesion clusters compared to full-length talin or talin rod. Thus, we suggest that the IBS2 domain in the talin 1 rod has an important role in crosslinking integrin nanoclusters.

Discussion

These results demonstrate that isolated linear fibres (width ≤30 nm) are unable to support matrix adhesion formation despite ligand density being adequate. However, on 2D substrates of RGD ligands³ where two 10-nm lines (either parallel or crossing) are brought within 110 nm, adhesion nanoclusters assemble similar to continuous RGD-coated glass. These different RGD patterns show that ligand geometry is an important factor in modular adhesion cluster formation, and in subsequent adhesion development on rigid substrates, where a few liganded integrins in a 2D configuration can recruit unliganded integrins to form dense nanoclusters, the modular units of adhesion (Fig. 6)²². Another plausible hypothesis is that two small clusters of half the suggested size could form on each line of line pairs, in close proximity, thereby creating clusters of ~110 nm (ref. ²²). Importantly, single small matrix fibres (≤30 nm) do not stabilize modular adhesion nanoclusters and therefore are unable to support adhesion formation and downstream signalling.

The adhesion proteins FAK, paxillin, talin and β₃ integrin associated as nanoclusters in adhesions on various matrix geometries, suggesting that they contribute to the fundamental units of cell–matrix adhesion and could be required for signalling²¹. In adhesions, the closely spaced but discrete integrin nanoclusters could catalyse the formation of multiple adhesion protein complexes by producing high local concentrations of signalling components while still enabling free access to inactivating proteins—and allowing for the large and varied number of protein–protein interactions needed for signalling via adhesions^46–49. The integrin nanoclusters could be stabilized by a 2D cytoplasmic adhesion protein complex including talin. From measurements of talin length in cells, the molecules are typically in a trigonal configuration spanning >110 nm (Fig. 6)⁵⁰, and talin mutations can change the cluster diameter²². Forcing two integrins that bind to a talin dimer to be close to each other could result in collapse of the talin dimer⁵¹. However, integrins separated by greater distances and bound to the talin rod will maintain talin in an elongated configuration⁵¹ scaffolding the integrin nanoclusters, which is concordant with our results.

During the initial stages of nanocluster formation, there is a rapid sequence of steps that will determine whether the nascent adhesion matures into a focal adhesion (or a podosome) or will disassemble. The transient nature of the integrin aggregates on single lines indicates that these are not normal adhesive nanoclusters (dense and long-lived). One possible explanation for the low level of adhesion formation and the rapid turnover is that linear arrays of liganded integrins are unable to support forces on associated unliganded integrins. Large focal adhesions form when cells are plated on micro-nanopatterned substrates with ligand dots separated by ~60 nm, but only transient adhesions that could not support high forces form when ligand-functionalized nanodots were spaced >80 nm apart^45,52. This indicates that stable, ~110-nm nanoclusters can assemble when ligands are spaced at 40–60 nm but not at ~90 nm (refs. ^8,9,52), since more than four integrins can bind ligands within ~110 nm. Thus, we suggest that unliganded integrins at the edges of nanoclusters could be pulled away by the actomyosin machinery⁴⁵, destabilizing the nanoclusters (Fig. 6) unless they are anchored in two dimensions.

Most matrix proteins assemble into fibres that range in diameter from 5 nm to several micrometres (collagen fibres). Based on our findings, there must be dense arrays of parallel small fibres such that these are spaced within ~110 nm or have overlapped for adhesion formation. One-dimensional stress on fibre arrays should increase fibre density, resulting in an increase in adhesion contacts. On 2D surfaces such as basement membranes, a critical local density of ligands is needed, and this can be as few as four ligands in a <110-nm quadrilateral pattern⁹. Such modular adhesions based on single integrin clusters could be the basis of the small adhesions observed in fibrous in vivo environments when fibres are overlapping. Since only about 10% of the integrins appear to be ligand bound^8,9 for adhesion maintenance, the forces on individual small fibres may be very high and this could aid in matrix remodelling. Fibrosis or excess deposition of large collagen fibres would decrease the force per fibre and compromise the cell’s ability to remodel collagen. Logically, the detailed aspects of 2D integrin clusters influence many cell–matrix interactions and those interactions are critical for determining cell viability and function.

Methods

Nano-array fabrication.

Nanoline and nanodot arrays were fabricated on standard glass coverslips (24 × 40 mm², No. 1.5, VWR). A bi-layer of poly(methyl methacrylate, either molecular weight (MW) 35 K and 495 K or MW 495 K and 950 K, Microchem) was coated with a thin conductive layer (either Aquasave by spin-coating or 10-nm Au by sputtering), and then patterned by electron-beam lithography (NanoBeam nB4, or JEOL JBX-9300FS). After exposure, the conductive layer was removed (Aquasave by deionized water rinsing or Au by wet etching). The samples were developed in methyl isobutyl ketone/isopropyl alcohol (IPA) 3/1 at 4 °C with ultrasonication for 1 min. For AuPd samples, AuPd alloy (60/40) with a Ti adhesion layer was deposited by electron-beam evaporation. For Ti samples, only Ti was evaporated. After the lift-off process in solution (using either Remover PG or Remover 1165) overnight, the samples were rinsed in IPA and blow-dried.

Nanopattern functionalization.

AuPd nano-array functionalization.

The AuPd samples were cleaned in 1.5-h-aged piranha solution (H₂SO₄/H₂O₂ 3/1) for 3 min, then in oxygen plasma for 5 min. The cleaned samples were immediately immersed in a freshly prepared 1-mM mixture of HS-C₁₁-EG₆-biotin and HS-C₁₁-EG₃-OH (ProChimia Surfaces) solution in anhydrous ethanol for ~18 h. The thiolated samples were mounted on a channel slide (Ibidi sticky-Slide VI 0.4) and immediately filled with small unilamellar vesicles (SUVs) of DOPC. After 10 min, supported lipid bi-layers formed on the surface and were washed with PBS and blocked with 1% casein for 1 h at room temperature. Finally, the nano-arrays were functionalized by Dylight 650-Neutravidin (10 mg ml^–1 in PBS for 30 min at room temperature), and then by biotin RGD (10 mg ml^–1 in PBS for 30 min at room temperature).

Ti nano-array functionalization.

The Ti samples were cleaned in isopropanol and then treated with oxygen plasma for 5 min, this process oxidizing the Ti into TiO₂. Samples were mounted on a channel slide and incubated with 200 μg ml^–1 of 1-oleoyl-2-[6-biotinyl(aminohexanoyl)]-sn-glycero-3-phosphoinositol-3,5-bisphosphate (ammonium salt) (biotin PI(3,5)P₂) diluted in deionized water. After 10 min, the TiO₂ arrays were functionalized with biotin. Samples were washed with water and immediately filled with SUVs of DOPC to passivate the substrate. The subsequent protocol is identical to AuPd functionalization.

The advantages of using Ti are twofold. First, TiO₂ reduces the quenching issue caused by AuPd and is therefore beneficial for super-resolution imaging; and second, a one-step lipid functionalization is easier, faster and more robust than the thiolation of AuPd, especially in a humid environment.

Supported lipid bi-layer assembly and functionalization.

Lyophilized lipids were purchased from Avanti lipids; SUVs were made. Cyclic RGD was purchased from Peptide international (No. PCI-3895-PI). NeutrAvidin Protein, Dylight 650 was purchased from Thermofisher Scientific (No. 84607). To assemble RGD-functionalized supported lipid bi-layers, DOPC was doped with 1,2-dipalmitoylsn-glycero-3-phosphoethanolamine-N-cap biotinyl (16/0 Biotinyl Cap PE) on clean coverslips (No.1.5). These were functionalized with biotin RGD using Dylight 650-Neutravidin as a linker between biotin on the lipid and RGD⁵³.

Glass functionalization.

All substrates were freshly prepared on coverslips cleaned by acid wash. Fibronectin-coated substrates were prepared by incubation of 10 μg ml^–1 fibronectin for 1.5 h at 37 °C. Continuously RGD-coated substrates were functionalized using unlabelled Neutravidin (1 mg ml^–1) for 2 h at room temperature, washed with PBS and incubated with biotin RGD at room temperature for 1 h. Poly-l-lysine (PLL)-coated substrates were prepared by incubation of PLL (1 mg ml^–1) on clean coverslips at 370 °C for 2 h.

Cell culture.

Mouse embryonic fibroblasts, HFFs, CHO cells (expressing a very low level of β₃ integrin⁴⁴) and talin1^–/– MEFs^22,54 were used for these experiments. They were cultured in standard Dulbecco’s modified Eagle’s medium (DMEM) with 10% foetal bovine serum and 5% CO₂. Plasmids were transfected by the Neon transfection system using the recommended protocol. Experiments were carried out 42–48 h post transfection to allow for β₃ expression on the plasma membrane. Thirty minutes before the experiments, cells were resuspended using trypsin and allowed to recover in Ringer’s solution. Cells were allowed to spread on the substrates in Ringer’s solution.

Plasmids and antibodies.

β₃-mEos2, β₃D119Y-mEos2 and β₃N305T-mEos2 were a gift from G. Giannone. β₃ D119YN305T-mEos2 was generated using point mutagenesis in β₃ D119Y-mEos2 using a Q5 site-directed mutagenesis kit. Paxillin-mApple and β₃-PAGFP were obtained from the Michael W. Davidson group. PAGFP containing plasmids was co-expressed with paxillin-mApple to identify cells expressing plasmids and to focus the adhesions using total internal reflection microscopy (TIRF). All β₃ mutants were cloned in PAGFP. GFP-tagged talin1 full-length, talin rod and talin rod IBS2* (mutant in talin-rod GFP—K2085D/K2089D, followed by deletion of amino acids 2,086–2,311) are described in ref. ²². Phosphorylated FAK antibody was purchased from Abcam (No. ab81298), paxillin antibody was purchased from BD Biosciences (No. 610051) and β₁ 9EG antibody was purchased from BD Biosciences (No. 553715). Alexa 488-, Alexa 546- and Alexa 647-labelled secondary mouse, rat and rabbit antibodies were purchased from Invitrogen. Phalloidin Alexa 405 was purchased from Thermofisher (No. A30104).

Immunofluorescence.

Cells were fixed with freshly prepared 4% formaldehyde for 10 min and permeabilized with 0.5% Triton X-100 for 15 min at 37 °C, followed by blocking with 1% BSA for 1 h at room temperature. They were then incubated with primary antibody for 4 h, or overnight at 4 °C, and with secondary antibody for 2 h at room temperature.

Treatment with MnCl₂ and Y27632.

For manganese treatment, MnCl₂ was used at a concentration of 0.5 mM. MEFs were treated for 30 min while spreading on the patterned substrates, with no pretreatment. Cells were treated with 30 μm Y27632 for 30 min during spreading, after attachment for 15 min on the substrates.

Microscopy.

The structured illumination confocal microscope comrises (1) a Nikon Ti-E Sapphire fully motorized body; (2) a Spinning Disk engine from Yokogawa W1 with a single, 70-μm pinhole disk, quad dichroic and dual camera port; (3) an electron-multiplying charge-coupled device (EMCCD) (Princeton Instruments, ProEM HS 1024BX3 megapixels with 30-MHz cascade and eXcelon3 coating; (4) a structured illumination re-scanner that provides optical deconvolution, the Roper Scientific France Live SR module; and (5) a back-illuminated scalable complementary metal-oxide semiconductor camera (Photometrics Prime 95B), with 1.44 megapixels and 100- and 200-MHz readout channels. This provides an optical resolution of ~120 nm, which is ideal to resolve modular integrin clusters that cannot be resolved using a standard confocal microscope. This microscope also has a spot photo-activation module, in which a line of the same thickness as a single airy unit is photo-activated using a 405-nm laser line.

PALM data were collected on a TIRF-iLas2 system from Roper, mounted on an Olympus X81 inverted-microscopy body and evolve512 EMCCD camera (Photometry) with a ×100/1.49 numerical aperture UApoN objective lens. Cells expressing PAGFP-tagged constructs were photo-activated using a 405-nm laser, and the photo-activated fluorescence signal was excited with a 488-nm laser. Samples were simultaneously illuminated by both lasers. The power of 405-nm laser was increased during the experiment, to maintain the number of stochastically activated molecules fairly constant and well separated during acquisition. The software used was Metamorph (Molecular Devices) in streaming mode at 20 frames s^–1 and 50-nm exposure time. Fluorescent beads of 100 nm (Spherotech, No. FP-0257–2) added to the sample were used to correct for x–y drift during long-term acquisition. PALM images were reconstructed using ground-state depletion software⁵⁵. PAGFP was imaged in PBS buffer.

Data analyses.

Cell area and focal adhesion area and number were measured from thresholded images that were immunolabelled with paxillin and actin on Ti substrates. In confocal images, focal adhesions were measured using thresholded images, where the minimum size threshold of five pixels was used for identification of adhesions, which were measured in the lamellipodial regions. To measure clusters within the focal adhesions, super-resolution confocal images were used. The numbers of clusters assayed were plotted per unit area of the cell. Clusters were measured in the outer area 5 μm from the cell edge. Clusters were identified in a thresholded image, and sizes >0.1 μm² were counted.

Kymographs were generated and analysed using ImageJ. These were plotted, after bleach correction, along a region of interest perpendicular to the nanoline patterns.

In PALM images, nanoclusters were identified and measured using previously described local maxima methods⁸. Each point in the PALM image was assigned a diameter of 5 nm. The diameter of the nanocluster was estimated as full-width, half-maxima of the identified nanoclusters.

Statistics and reproducibility.

All experiments were repeated between three and five times. Each experiment contained all the nanopatterned substrates on the same sample (coverslip), and hence all other conditions were similar and the only variable was substrate geometry. Data were not pooled from technical replicatations. Box plots show interquartile distance ± s.d. (whiskers), with solid boxes representing individual data points. Analysis of variance statistics were run on the entire population and, for comparison of two subsets, two-tailed Student’s t-test was used and P values noted on the graph.

Use of direct stochastical optical reconstruction microscopy on Ti nanolines.

It is well known that the fluorescence from a molecule in close proximity to a metallic nanostructure is affected by radiative and non-radiative energy transfer, which leads to fluorescent enhancement or quenching depending on the material and geometric parameters^56–58. Au has a strong plasmonic effect, which can cause severe problems in fluorescence microscopy. As a result we used AuPd, which provides attenuated surface plasmon resonance. For the size range of interest, the quenching effect on AuPd nanoparticles is weaker than on pure Au but still significant, as we observed in a previous study⁵⁹. Similarly, the fluorescence signals quench rapidly on AuPd nanolines, making super-resolution imaging extremely inefficient in the red and infrared regions of the spectrum. In this regard, we also used Ti, which was oxidized in oxygen plasma before functionalization. The oxidized Ti nanopatterns were more dielectric and therefore provided more stable fluorescence. Super-resolution imaging worked well, although 80-nm-spaced lines were not resolved separately. This is probably because the quenching effect was not fully eliminated, similar to fluorescence absorption observed for TiO₂ nanoparticles functionalized with enediol ligands^55,60 (see Methods). The use of different functionalization chemistry or nanostructured materials (for example, polymer) could help further improve the imaging resolution in future work.

Although it is still challenging to directly resolve 10-nm-wide nanolines using super-resolution microscopy, it is evident that the ligands are well confined in the nanopatterned region. With our state-of-art protocol, the defects in the bi-layer and non-specific adsorption of ligands are minimal. Regions containing no nanopatterns showed no cell binding (Supplementary Fig. 1a). In previous work, we were able to precisely control ligand occupancy (down to the single molecule on a nanodot)⁵⁹, which scales linearly with pattern density³⁸. This indicates that there is no significant edge effect around the nanostructures backfilled with the lipid bi-layer (accumulation of ligands due to defective bi-layer coverage). In this work, the fluorescence intensity of line pairs is double of that of single lines (Fig. 2d), which shows that the bi-layer covers the small gap between lines in a pair with no extra defective accumulations. Therefore, the fact that unliganded integrins colocalized with line pairs indicates that these bridge the gap between two thin fibres (although not resolved separately).