Local Tension on Talin in Focal Adhesions Correlates with F-Actin Alignment at the Nanometer Scale

Abstract

Cellular force transmission and mechanotransduction are critical in embryogenesis, normal physiology, and many diseases. Talin plays a key role in these processes by linking integrins to force-generating actomyosin. Using the previously characterized FRET-based talin tension sensor, we observed variations of tension both between and within individual focal adhesions in the same cell. Assembling and sliding adhesions showed gradients with higher talin tension toward the cell center, whereas mature, stable adhesions had uniform talin tension. Total talin accumulation was maximal in high-tension regions; by contrast, vinculin intensity was flat or maximal at the adhesion center, and actin intensity was maximal toward the cell center. To investigate mechanism, we combined talin tension imaging with cellular cryotomography to visualize the correlated actin organization at nanometer resolution. Regions of high talin tension had highly aligned linear actin filaments, whereas regions of low tension had less-well-aligned F-actin. These results reveal an orchestrated spatiotemporal relationship between talin tension, actin/vinculin localization, local actin organization, and focal adhesion dynamics.

Introduction

Cells both exert and sense mechanical tension between integrin-dependent adhesions and the extracellular matrix (ECM) (1, 2). Tension from cells mediates morphogenetic movements and ECM assembly, whereas the mechanical properties of the ECM regulate cell contractility, growth, differentiation, and survival (3, 4, 5, 6). As a result, these linkages play key roles in a vast array of biological processes including embryonic development, wound healing, and normal physiology and diseases such as cancer, hypertension, and fibrosis, among others.

Talin is a key mediator of force transmission between integrins and actin filaments (7, 8). Its N-terminal FERM domain binds integrin-β-subunit cytoplasmic domains, whereas its C-terminal rod domain links to F-actin both directly through its two actin binding sites and indirectly through vinculin (9, 10). Talin is thus critical for force transmission between the actin cytoskeleton and the ECM and for cells sensing the mechanical properties of their environment.

We developed a fluorescence sensor of tension across specific proteins and applied this approach to measure tension across talin1 in cultured cells (11). These studies confirmed that talin1 in focal adhesions (FAs) is under tension and revealed several additional unexpected features. Talin in central FAs was generally under less tension than in peripheral FAs. Tension in large FAs required actin-binding site 2 in the center of the rod, whereas actin-binding site 3 at the C-terminus was dispensable. Talin tension was lower in cells on soft substrates, in keeping with the phenomenon of cellular stiffness sensing, in which soft surfaces trigger a reduction in contractility (12).

These results prompted us to examine local talin tension in more detail. Our results first revealed that talin tension can vary not only between FAs but locally within individual FAs related to their dynamic state. To understand these variations at the nanometer scale, we then carried out correlative light and cellular electron cryotomography (cryo-ET) studies on the underlying cytoskeleton. This technique allows imaging of actin filaments at nanometer resolution in conjunction with talin tension in an intact cellular environment without dehydration or contrast-enhancing agents. These experiments revealed a marked correlation between local talin tension and the organization of the actin filaments.

Materials and Methods

Cell culture and transfection

Talin1^−/− cell lines were cultured in Dulbecco’s modified Eagle’s medium/F12 (Gibco, Carlsbad, CA) with 10% fetal bovine serum (Gibco), penicillin-streptomycin (Gibco), β-mercaptoethanol (5 μL in 500 mL; Sigma-Aldrich, St. Louis, MO), and sodium bicarbonate (8 ml of 7.5% (wt/vol) in 500 mL media; Sigma-Aldrich). Cells were plated in a 35-mm dish at 40% confluency overnight before transfecting the sensor using 5 μL of lipofectamine 2000 (Invitrogen, Carlsbad, CA) and 2 μg of sensor plasmid in Opti-MEM (Gibco) according to the manufacturer’s protocol. Transfection medium was replaced with cell culture without antibiotic after 6 h. Transfection efficiency was checked after 36 h, at which time cells were replated for imaging. Fluorescence resonance energy transfer (FRET) imaging was carried out on glass dishes (MATEK Corporation, Ashland, MA) that had been coated with 10 μg/mL fibronectin overnight at 4°C. Blebbistatin (B0560; Sigma-Aldrich) was added to cells plated on dishes during microscopy.

Immunostaining

Cells were once washed with PBS, fixed in 4% paraformaldehyde (JT Baker, Pillipsburg, NJ) in phosphate-buffered saline (PBS) for 20 min, and permeabilized with 0.1% Triton X-100 (American Bio, Natick, MA) in PBS for 20 min at room temperature (RT). Actin was labeled using Alexa-Fluor-647-conjugated phalloidin (1:200 in PBS; Molecular Probes, Eugene, OR). Cells were blocked with 1% bovine serum albumin in PBS for 1 h at RT and incubated with primary antibody-mouse vinculin (1:500 in 1% bovine serum albumin in PBS; Sigma-Aldrich) at 4°C overnight. Cells were washed 3× in PBS and incubated with Alexa-Fluor-647-conjugated secondary antibody diluted in PBS (1:1000; Invitrogen) at RT for 1 h.

Cells for electron microscopy

Carbon-coated holey finder electron microscopy grids (Quantifoil, Jena, Germany) were ultraviolet sterilized for 30 min under a tissue culture hood, coated with fibronectin (4 μg/mL in sterile PBS) overnight at 4°C, and washed 3× with PBS. At 6 h after plating, cells were fixed using 2% formaldehyde (Tousimis, Rockville, MD) in water (Sigma-Aldrich) containing 0.1 M Piperazine-1,4-bis(2-ethanesulfonic acid) (PIPES; Sigma-Aldrich), 1 mM ethylene glycol-bis(β-aminoethyl ether)-N,N,N′,N′-tetraacetic acid (EGTA; Sigma-Aldrich), and 1 mM MgSO₄ (JT Baker) (pH 6.9) at 37°C for 15 min. Grids were washed 3× (1–3 min each) using PHEM buffer: 60 mM PIPES (Sigma-Aldrich), 25 mM Hepes (Acros Organics, Morris, NJ), 2 mM MgSO₄ (JT Baker), 10 mM EGTA (Sigma-Aldrich) at pH 6.9 and 37°C for 15 min. FRET imaging was carried out within 24 h of fixation. These sample grids were kept 4°C immersed in PHEM before manual plunge freezing in liquid-nitrogen-cooled liquefied ethane using a home-designed cryoplunger. Samples were stored in liquid nitrogen until analysis by electron microscopy.

FRET measurements and image analysis

FRET images were taken and analyzed essentially as described (11). Images were taken on an Eclipse Ti microscope (Nikon, Tokyo, Japan) equipped with an Ultraview Vox spinning disk confocal system (PerkinElmer, Waltham, MA) and an electron-multiplying charged-coupled device C9100-50 camera (Hamamatsu, Hamamatsu City, Japan), using a 100×, 1.4 NA (numerical aperture) oil objective, resulting in a pixel size of 70 nm. Images were acquired using Volocity software. Three sequential images with 500-ms exposure times were acquired with the following filter combinations: donor (EGFP) channel with a 488-nm line (excitation) and 527/55 (emission), acceptor (tagRFP) channel with a 561-nm line (excitation) and 615/70 (emission), and the FRET channel with a 488-nm line (excitation) and 615/70 (emission). Donor leakage was determined from EGFP-transfected cells, whereas acceptor cross excitation was obtained from tagRFP-transfected cells. For all the calculations, respective background subtraction, illumination gradient, and pixel shift correction were performed followed by three-point smoothing. The slope of pixel-wise donor or acceptor channel intensity versus FRET channel intensity gives the leakage (x) or the cross-excitation (y) fraction, respectively. FRET map and pixel-wise FRET index for the sensors were determined from

$$FRETindex=\frac{\text{FRET}\text{channel}-\text{x}\times \text{Donor}\text{channel}-\text{y}\times \text{Acceptor}\text{channel}}{\text{Acceptor}\text{channel}}.$$

To get the FA segmentation, the acceptor channel (tagRFP) image was three-point smoothed and the mean background intensity subtracted using a macro written in ImageJ. The image was corrected for the illumination gradient, and then the cytosolic signal from talin-TS/CS transfection was subtracted by using a 50-pixel-wide sliding paraboloid. The background subtracted image was convolved using a 5 × 5 kernel with all elements set to −1 except the central element, which was 24, and then the Otsu algorithm was applied with the fill hole function to get a segmented FA image. Another FA mask was acquired by applying enhance local contrast with block size = 19, histogram = 256, and maximum = 6 and then using the exponential function. The image was contrast enhanced with saturation = 0.35. The image was three-point smoothed, and the Otsu algorithm was applied with the fill hole function to get a second segmented FA image. These two segmented images were added to get the final FA mask, which was used to get the FRET index within FAs. Only FAs with size >0.25 μm² and intensity >50% above the background intensity (i.e., signal/noise >1.5 times at least) but not saturated on the 14-bit camera were considered for FRET calculations.

The FRET index can be approximated to force using the in vitro calibration curve published earlier (13). However, recent work (14) shows that in cellulo, the peptide in the FRET/tension sensor module may not be behaving the same as in the in vitro condition, raising a question about the calibration curve itself. Further, our FRET index values are close to FRET efficiency except for the G factor, which is needed to account for quenching, etc. Hence, we approximated the force value that corresponds to FRET index in the following way. First, raw FRET values were background subtracted using the intermolecular FRET as baseline. These values were then normalized to the control sensor (talin-CS) FRET index. This procedure yields the result that average force in matured FA is ∼6 pN; in small adhesions, talin is almost relaxed ∼1 pN while assembling; and sliding adhesions have a linear average gradient between 1 and 5 pN between the distal and proximal edges.

ImageJ (National Institutes of Health, Bethesda, MD) was used for basic image processing. All analyses of the FRET images were done using custom-written software (MATLAB R2016b; The MathWorks, Natick, MA). Graphs were plotted in Origin (9.1; 64 bit).

Cellular cryotomography

Grids were manually plunge frozen in liquid ethane at liquid nitrogen temperature. Before imaging, the grids were put into AutoGrids, a grid sample carrier, which was then transferred via a Krios NanoCab transfer device (FEI, Hillsboro, OR/Thermo Fisher, Waltham, MA) into the Krios Cryo AutoLoader, the automated sample loader/unloader system of the Titan Krios electron cryo-microscope. Within the Cryo AutoLoader, the AutoGrids were kept at liquid nitrogen temperature at all times. Cryo-ET data was obtained with a Titan Krios (FEI /Thermo Fisher) equipped with an extra-high brightness field emission gun operated at 300 kV. Tilt series (±65°, every 1.5°) were acquired on a back-thinned 4 × 4 k Falcon II detector (FEI/Thermo Fisher) under minimal dose conditions using the Tomo package (FEI/Thermo Fisher) in batch mode at an average dose of ∼100–120 e⁻/Ų for the entire tilt series and defocus of 10–14 μm.

For the first steps in image acquisition after the AutoGrids are transferred into the AutoLoader, grids are screened for usability and quality. Once reviewed, the correlated and control regions were selected and set up for cryotomography data collection. A total of 27 cryotomograms from six cells with correlated FRET data were used for this study. In addition, 64 tomograms from 11 control cells without fluorescence were taken for refining imaging conditions and to serve as control. Using full mapping of the grid followed by midrange magnification tiling, we systematically bridged the gaps in scales and located the cells of interest in the electron microscope using the markings of the finder grids. Midrange-magnification images at pixel sizes of 14 and 28 nm were acquired for each cell to allow efficient correlation with the corresponding FRET images. The magnification of the tomographic reconstructions resulted in a pixel size of 0.47 nm. The fidelity and quality of the data collection were monitored with real-time automatic reconstruction protocols implemented in the pyCoAn package, an extended Python version of the CoAn package (15).

Image reconstruction and alignment

Tilt series were aligned using the IMOD package (16) with a combination of fiducial-based and patch-based approaches. Three-dimensional densities were generated using the simultaneous iterative reconstruction technique as implemented in Tomo3D (17). Reconstructions were binned by a factor of four for further analysis. Low-magnification cryo-transmission elctron microscopy (cryo-TEM) images were aligned with the talin-sensor FRET images using the holes in the carbon (18). The accuracy for the alignment as estimated from the prediction error was 69.8 ± 5.8 nm. The tomogram regions were located and aligned to the low-magnification images using a Z-projection image of the final tomogram and correlation-based template matching. The combined inverse transformation was then applied to the aligned FRET images for overlay with the tomograms and for evaluation of correlated force index.

Analysis of filament alignment

Two-dimensional fast Fourier transforms were calculated using 50-nm-thick slices of the respective tomogram near the ventral membrane from 240-nm- × 240-nm-sized overlapping boxes. The distance between boxes was set to 20 nm to emulate a sliding window. Next, the Fourier amplitudes of the contributing boxes were averaged to enhance the signal. An alignment score was calculated as follows. 1) The intensity at the origin of Fourier space was set to zero; 2) to reduce the effect of random noise in the Fourier amplitude averages, an iterative median filter of three iterations was applied (19); 3) next, the background was flattened using a Gaussian with a kernel size of 53 × 53 Fourier pixels to account for background variation caused by the missing wedge of tomographic data collection; and 4) the resulting images were normalized to a mean of one and a root mean-square deviation of one. The resulting images served as a basis for segmenting the component due to the filament alignment (foreground) from the general background and noise components (background).

The segmentation between foreground and background was based on maximizing the sum of their Renyi entropies of order two $\mathcal{R}$, subject to modifying the threshold T (20, 21), i.e.:

$$\mathcal{R}=-\ln \sum _{g=M}^T{\left(\frac{\mathcal{P}\left(g\right)}{\mathcal{C}\left(T\right)}\right)}^2-\ln \sum _{g=T+1}^G{\left(\frac{\mathcal{P}\left(g\right)}{1-\mathcal{C}\left(T\right)}\right)}^2,$$

with g denoting a gray value at a particular pixel. The first sum is running over all pixels with gray values between the minimum M up to the threshold T, and the second sum is running over all pixels with gray values larger than T up to the maximal gray value G. $\mathcal{P}$ denotes probability and is estimated from the gray-level histogram. (T) denotes the cumulative probability of T:

$$\mathcal{C}\left(T\right)=\sum _{g=M}^T\mathcal{C}\left(g\right),$$

where the sum runs over all gray values g up to T. Conceptually, this strategy attempts to detangle two different signal sources, one coming from the aligned filaments and the other one from background noise. In essence, maximizing the sum of the order-two Renyi entropies bounds the collision probability between these samples. The collision probability is the rate at which the distribution repeats when successive samples are taken. As a final step, an iterative median filter of three iterations was run on the segmentation results to remove spurious peaks.

The alignment score was calculated as 200/N_F, where N_F is the number of foreground pixels. Because there are always two spots, owing to Fourier-space symmetry, this means that a spot size of 100 pixels corresponds to a score of one. More disperse spots (lesser degree of alignment) get lower scores. The local filament alignment scores for the linear regression analysis were calculated using sets of Fourier-amplitude averages from 50 boxes covering the entire fluorescing regions of the tomograms. The force index FL(x) at the center position x of a particular 50-box average was calculated from the intensities IF of the correlated FRET image as FL(x) = 1 − 5IF(x), which maps IF between zero and one. The averaged alignment scores for the regions in Figs. 4, 5, 6, and 7 were calculated by averaging local alignment scores using only boxes covering the respective area of interest (only blue or red areas). All FRET/cryo-ET alignments and analyses at the electron microscopy scales were performed within pyCoAn.

Figure 4.

High talin tension region and actin organization. (A–C) Correlated imaging of talin tension FRET and cryo-transmission elctron microscopy (cryo-TEM) in FAs. The color bar for the FRET index is the same as in Fig 1. (D) The sliding-window Fourier amplitude average calculated in the blue high-tension region shows two well-defined spots with an average alignment score of 0.97, demonstrating high filament alignment. (E) Segmentation of individual actin filaments. Scale bars in (A) 5 μm, (B) 2 μm, and (C) 300 nm; (E) the diameter of a single actin filament is 9 nm.

Figure 5.

Low talin tension and actin organization. (A–C) Correlative imaging of a region with low talin tension from the same cell as Fig. 4. The color scale for the FRET index is the same as in Fig. 1. (D) The sliding-window Fourier amplitude average calculated in the red region shows diffuse spots with an average alignment score of 0.19. (E) Segmentation of individual actin filaments. Scale bar in (A) 5 μm, (B) 2 μm, and (C) 300 nm; (E) the diameter of a single actin filament is 9 nm.

Figure 6.

Talin tension and actin filament alignment. Regions with different talin tension and corresponding sliding-window Fourier amplitude patterns. The FRET index color scale is the same as in Fig. 1. Averaged alignment scores are indicated in the top right corner of the respective Fourier amplitude averages. The top row (A–C) shows regions with high talin tension. The lower row (D–F) shows regions with low talin tension. Regions with low talin tension in filopodial protrusions (e.g., (F)) also are less well aligned, though are more aligned than regions with low talin tension outside filopodia (e.g., (E) and Fig. 5). Scale bar in main panels (A–F) 300 nm; top right panels 2 μm.

Figure 7.

Talin tension correlates with local actin filament organization. Regression analysis of filament alignment score calculated from cryotomogram local Fourier-amplitude averages plotted against force index (arbitrary units for which 0 corresponds to minimal tension (FRET index 0.2) and 1 corresponds to maximal tension (FRET index 0.0)). See Materials and Methods for details. A strong positive correlation between the two variables is indicated by the R-value of 0.70 with a p-value < 0.0001 and sample size N = 180. To see this figure in color, go online.

Results

Spatial and temporal variation within single FAs

We first examined talin1^−/− cells reconstituted with transiently expressed talin tension sensor (TS) or the control sensor (CS) in which the FRET module is linked to the C-terminus and cannot bear tension. Cells were plated on fibronectin-coated glass dishes and imaged as we did previously (11). The FRET index, in which FRET efficiency is corrected for total intensity, provides a measure of FRET per molecule, with lower FRET indicating higher tension. FRET index maps revealed that many FAs showed substantial internal variations (Fig. 1, A and B). Although the line profile of the FRET index (averaged over three-pixel-wide lines with pixel size = 70 nm) along the length of individual FAs does not show any distinct patterns in talin-CS-transfected talin1^−/− cells (Fig. S1 A), it revealed three distinct patterns in talin-TS (Figs. 1, C–E and S1; detailed criteria in figure legend). In group (i) (77 out of 178 adhesions), FRET was high at the cell periphery and decreased toward the cell center, indicating higher tension toward the center (Figs. 1 C and S1 B). Group (ii) (51 out of 178 adhesions) had high FRET/low tension that was relatively constant along their length (Figs. 1 D and S1 C). Group (iii) (50 out of 178 adhesions) had low FRET/high tension that was constant (Figs. 1 E and S1 D). These three groups have different FA lengths and very distinct dynamics as described below and characterized in Fig. S1, E–G (statistics of slope, FRET index, and FA length for multiple FAs from different cells for these three groups of talin-TS and talin-CS).

Figure 1.

Spatial variation in talin tension within single FAs. (A) Representative pseudocolor images of FRET index for talin-TS within FAs of live cells plated on fibronectin-coated glass coverslips are shown. The white line shows the cell boundary. (B) A zoomed-in image showing the differential FRET index within individual FAs located near the cell periphery. The color bar shows index for high (0.2) and low (0.0) FRET values. (C–E) Representative line profiles of the FRET index within single FAs. 0 μm is near the cell periphery. The black line is the linear fit to the FRET profile. (C) A group (i) FA (marked as 1 in (B)) at the cell front. (D) A group (ii) FA (marked as 2 in (B)) is shown. (E) A group (iii) FA (marked as 3 in (B)). (F) The fraction of FAs (N = 178) from different cells with lamellipodial extensions (N = 5 cells) classified as group (i) (green), (ii) (red), and (iii) (blue). Group (i) FAs were defined as a decrease in FRET index over FA length (average slope, defined by FRET index change over FA length, is −0.045 per micron); groups (ii) and (iii) have no gradient in the FRET index. Group (ii) FAs were defined as FRET index >0.1, whereas group (iii) had a FRET index between 0.05 and 0.1.

To investigate the origins of these different patterns of tension and morphology, we carried out time-lapse imaging. This analysis showed that group (i) FAs were usually of intermediate size (1–3 μm in length) and were either growing FAs near the cell leading edge (Figs. 2, A–C and S2, A and B; Video S1) or were near the retracting edge and showed “sliding” behavior (Figs. 2, D–F and S2, C and D). Additionally, Video S2 shows an example of a mature adhesion (group (iii) FA) transitioning to sliding adhesion (group (i)), which is due to disassembly at the distal edge and assembly at the proximal edge, i.e., treadmilling (22, 23). Group (ii) FAs were smaller than group (i) (<1–2 μm in length) and near the leading edge. A time-lapse series showed that the majority of these adhesions were transitioning from small focal complexes to larger FAs (Fig. 2, A–C; Video S1, initial time points). Group (iii) adhesions were larger (length >2 μm, up to 9 μm) and relatively static over the period of imaging, indicative of mature FAs (Figs. 2, G–I and S2, E and F; Video S3). Fig. 1 F shows the relative fractions of these categories. Similar time-lapse imaging and plots for talin-CS-transfected cells showed no specific variation in FRET index profiles (Fig. S3; Video S4).

Figure 2.

Talin tension and talin intensity in assembling, sliding, and stable FAs. Time-lapse images of assembling (A–C), sliding (D–F), and stable FAs (G–I) from cells as in Fig 1. (A), (D), and (G) show inverted talin (tagRFP) intensity with time point indicated at the top of each image. (B), (E), and (H) show pseudocolor maps of the FRET index. (C), (F), and (I) show line profiles of talin intensity and FRET index for a typical (C) assembling, (F) sliding, and (I) stable FA (marked as ^∗ in (A), (D), and (G)).

Video S1. Video of Assembling/Growing FAs at the Leading Edge of a Talin1^−/− Cell Transfected with Talin-TS Showing Inverted Talin Intensity, Corresponding FRET Index, and the Plot Line Profile for Talin Intensity/FRET Index for Each Time Point

Scalebar and time are indicated in each frame.

Video S2. Video of Sliding FAs in a Talin1^−/− Transfected with Talin-TS Showing Inverted Talin Intensity, Corresponding FRET Index, and the Plot Line Profile for Talin Intensity/FRET Index for Each Time Point

Scalebar and time are indicated in each frame.

Video S3. Video of Mature Stable FAs at the Edge of a Talin1^−/− Cell Transfected with Talin-TS Showing Inverted Talin Intensity, Corresponding FRET Index, and the Plot Line Profile for Talin Intensity/FRET Index for Each Time Point

Scalebar and time are indicated in each frame.

Video S4. Video of FAs at the Edge of a Talin1^−/− Cell Transfected with Talin-CS Showing Inverted Talin Intensity, Corresponding FRET Index, and the Plot Line Profile for Talin Intensity/FRET Index for Each Time Point

Scalebar and time are indicated in each frame.

Interestingly, in group (i) and (ii) adhesions, the peak of total talin intensity was usually toward the proximal edge of the FA (toward the cell center), which was also the region of higher talin tension (Figs. 2, A–F and S2, A–D; Videos S1 and S2). By contrast, talin tension in mature, stable adhesions was uniformly high, with talin intensity again highest in the FA center. As total force is the product of force per molecule and the number of talin molecules that carry the load, total traction force carried by talin would appear to be higher toward the proximal region of these FAs. We also examined the total levels of F-actin, which initiates the force, and vinculin, which, by bridging talin and actin, contributes to the tension on talin. These experiments used cells that were fixed with paraformaldehyde, which preserves FRET levels (11). Labeling F-actin with Alexa-647-conjugated phalloidin in fixed talin-TS cells showed peak intensity slightly shifted toward the cell center relative to talin intensity (average shift 8.1%; Fig. 3, A–C; quantified in Figs. 3, D–F, S4, and S4 inset). Vinculin staining intensity was flat or peaked in the center of the FA (Fig. 3, G–I, quantified in Fig. 3, J–L). Vinculin engagement/localization therefore does not closely correlate with the differential force (FRET index) on talin within the FA (a plot of mean FRET index versus mean vinculin intensity is shown in Fig. S5). Thus, although vinculin is known to contribute to talin tension (11), there must be other factors that contribute to tension and mediate these spatial characteristics. Short (15 min) treatment of cells with 10 μM blebbistatin to inhibit actomyosin contractility leads to reduced force on talin (increased FRET index) with uniform FRET profiles along the FA length (Fig. S6, A and B).

Figure 3.

Talin tension and actin/vinculin intensity within single FA. Representative inverted-intensity images of tagRFP talin (A and G), Alexa-647-conjugated phalloidin (B), and vinculin (H). FAs for which line profiles are plotted in (D)–(F) and (J)–(L) are indicated by asterisks. (C and I) A pseudocolor map of FRET index for talin-TS within FAs of the same cell shown in (A), (B), (G), and (H). Color bars show FRET index. (D–F and J–L) Representative line profiles of talin intensity (solid black line), actin/vinculin intensity (dotted gray line), and FRET index (colored solid line; color represents the group that the FA belongs to) within single FAs, shown in images (A) or (G).

Correlative imaging of talin tension and the underlying ultrastructure at the nanometer scale

As force on talin is mediated by F-actin (1), we next investigated the correlations between local actin organization and tension on talin. To determine cytoskeletal structure at nanometer resolution, we combined FRET imaging with cellular cryo-ET. Because of the requirements for cryo-ET, we focused on adhesions at the cell periphery, where cells are thin enough for high-resolution analysis. Cells plated on fibronectin-coated electron microscopy grids showed cell and FA morphology and talin FRET maps that were indistinguishable from cells on fibronectin-coated glass (Fig. S7).

To allow direct comparison of FRET images with cryo-ET structures without interference from cell movement and to maintain FA structural integrity, cells were rapidly fixed before FRET imaging. Fixed specimens were vitrified immediately after light imaging to prevent structural collapse or shrinkage associated with dehydration, and vitrified grids were imaged in the Titan Krios electron cryo-microscope. Midrange magnification tiling was used to locate regions of interest for detailed data acquisition and analysis. These images were subject to marker-free alignment (18) to match the FRET images with the cryotomograms with ∼100 nm precision. After this correlation step, cryo-ET images were acquired. This imaging modality enables the analysis of biological samples in three dimensions and in their hydrated state at nanometer resolution. The three-dimensional volumetric reconstructions (cryotomograms) of the imaged regions of the cell peripheries span windows of 2 μm width and were reconstructed in three dimensions with a pixel size of 0.47 nm. This magnification is sufficient to distinguish between the various types of cytoskeletal filaments as well as vesicles and large macromolecular complexes such as ribosomes and proteasomes (24). Twenty-seven cryotomograms from six different cells that were fully correlated with talin-tension data were collected and analyzed.

Alignment of actin filaments at the nanometer scale as a function of force on talin

The dominant features in the reconstructions of FAs are actin filaments, identifiable by their line-like appearance and characteristic 7–9 nm thickness. Actin filaments were organized into well-aligned parallel bundles in regions with high tension on talin (Fig. 4, A–C). By contrast, in regions of lower talin tension, actin filaments were much less aligned (Fig. 5, A–C). To quantitatively evaluate this feature, we employed a sliding-window Fourier transform method that reveals the degree and direction of alignment (Fig. S8). In these images, the contrast and shape of the bright areas indicate the degree of alignment, with bright focused spots indicating high alignment. These plots reveal much higher alignment in areas of low FRET/high tension on talin (Figs. 4 D, 5 D, and 6). Segmentation of the actin filaments in these regions also revealed higher alignment (Figs. 4 E and 5 E). There was no evidence for a difference in filament length in the two tension regimes.

Actin alignment was quantified by a procedure that is, to our knowledge, novel, combining averaging of sliding-window Fourier amplitude images and entropy-based segmentation to assess alignment in areas of arbitrary shape. The quantification indicated that actin filaments in regions with high talin tension were consistently well aligned, whereas actin filaments in regions with low tension had significantly lower alignment. Linear regression over many images showed that talin tension correlated with local filament alignment (Fig. 7, with R = 0.70). Even in filopodial protrusions, in which actin filaments are generally thought to be well aligned, actin filaments in FAs with low talin tension were on average less well aligned than actin filaments in FAs with high talin tension near the cell edge (Fig. 6, D and F) outside of filopodia.

Discussion

Integrin-mediated adhesions are the essential organelles that transmit force from intracellular actomyosin to the extracellular matrix to regulate cell and tissue movements and matrix assembly (9, 10). Conversely, altering force through applied strain, changes in cell contractility, or substrate stiffness alters FA assembly, disassembly, and organization (5). These bidirectional mechanisms therefore create a regulatory circuit that mediates homeostasis. Although light imaging and biochemical studies have yielded a basic understanding of FA structure and biology, many questions remain about their nanoscale organization and force sensitivity.

This study began with the observation that tension on talin varied locally in a subset of FAs. This behavior was then found to correlate with adhesion dynamics: both growing FAs at the leading edge and “sliding” FAs at the trailing edge showed gradients with higher tension/talin molecule toward the cell center. In both cases, high talin tension corresponds to the region of assembly. The regions of high talin tension also had higher talin intensity in both cases. These observations are consistent with the notion that tension on talin can promote FA assembly by recruiting other molecules such as vinculin and strengthening the link to actin (1, 2). By contrast, stable FAs had relatively constant talin tension along their length, though with higher talin intensity toward cell center. These results contrast to some extent with high-resolution traction force measurements that suggested higher tractions either in the center or toward the distal end of the FAs (25, 26). Traction force is calculated by determining bead displacements in soft, elastic materials underneath FAs, whereas the results presented here report force per molecule of talin in cells plated on glass coverslips. Several factors likely account for these differences. FAs are known to behave differently on rigid versus compliant substrates (27, 28). Additional linkers such as α-actinin, filamin, and parvins also transmit force between actin and integrins and could account for the difference. In contrast to measurements of force per molecule from talin FRET, traction depends as well on the number of molecules that transmit force. Indeed, inverse correlations between vinculin density and tension as reported by FRET index have been noted (29, 30). We also observed higher talin intensity in central and proximal regions of FAs compared to the distal region (Fig. 2; Videos 1, 2, and 3). As reported earlier, the relationship between actin speed and traction force is biphasic, with a decrease in force at very high speeds (31).

Results from vinculin FRET sensors showed higher tension per molecule toward the cell edge together with highest intensity around the FA center (29, 30). These results reinforce the notion that although vinculin contributes to force in talin, it cannot account for the complete behavior or regulation. Recent studies report that talin rod domain helix bundles in their folded states bind to a number of ligands, including RhoGAPs such as DLC1 (32) and proteins linked to FA turnover such as KANK1 and 2 (33, 34). Tension on talin may therefore result in complex signaling outputs that control local FA organization. Future modeling of force by FAs within cells using molecular clutch models will need to consider these complexities.

Analysis of the cytoskeletal organization in the same cells by high-resolution cryo-ET revealed that areas of high talin tension contained well-aligned actin filaments, whereas areas of lower talin tension had actin filaments of similar abundance and length but less well aligned. These data prompt the critical question: are the changes in actin organization upstream versus downstream of the changes in tension? In other words, does the F-actin alignment increase when tension is high, or do well-aligned actin filaments exert higher tension? The likely answer is both, based on the well-established cooperativity between force and FA formation. Myosin filaments are both the main source of tension across FAs and potent actin bundlers, apart from force generation. Further, myosin-dependent force generation increases under load both via mechanisms intrinsic to the myosin motor itself and via effects on coupling of upstream elements such as Rho and ROCK. Myosin is thus likely at the center of the positive feedback loop that mediates both actin bundling and tension, and accounts for the high correlation between these variables.

Taken together, the major conclusion from this study is the identification of regions of FAs with high talin tension and well-aligned actin filaments as the principal sites of assembly or growth, whereas regions of low talin tension and less-well-aligned actin filaments correlate with disassembly. Adhesion assembly thus correlates more strongly with force per talin molecule than with total traction stress, which was reported to be maximal toward the center or distal end (25, 26). Bundled F-actomyosin has been proposed to serve as a scaffold that favors FA assembly, apart from mechanical tension (35). Additionally, tension on talin can open binding sites for vinculin and perhaps other molecules that may favor adhesion assembly (7). The results are therefore consistent with the current view that tension on talin and actin-dependent scaffolds work in concert to promote adhesion assembly.

These results raise a number of questions. Given that stable FAs can have high talin tension and well-aligned actin, what variables distinguish stable, high-tension regions from those with net assembly? What variables determine local talin and vinculin density? Which other cytoskeletal linkages mediate force transmission? What are the regulatory mechanisms that mediate tension-induced actin alignment versus actomyosin-mediated FA assembly? Locally determined rates of talin and vinculin recruitment and dissociation are likely to underlie some of these phenomena and thus should be an important direction for future work.

Author Contributions

D.H., N.V., and M.A.S. conceived, directed, and funded the project. A.K. performed the FRET imaging and data analysis, prepared cells for EM tomography and wrote the first draft. K.L.A. developed and supervised the cell culture protocols for CLEM and cryotomography, worked with DH on setting all cryopreparations, and performed CLEM imaging. M.F.S. acquired all the cryotomography data. N.V. carried out three-dimensional reconstructions and correlation between FRET and EM data and devised and carried out the analysis of the EM data. A.K., D.H., N.V., and M.A.S. revised the manuscript.

Editor: Alexander Dunn.