Ultrafast single-molecule imaging reveals focal adhesion nano-architecture and molecular dynamics

Abstract

Using our newly developed ultrafast camera described in the companion paper, we reduced the data acquisition periods required for photoactivation/photoconversion localization microscopy (PALM, using mEos3.2) and direct stochastic reconstruction microscopy (dSTORM, using HMSiR) by a factor of ≈30 compared with standard methods, for much greater view-fields, with localization precisions of 29 and 19 nm, respectively, thus opening up previously inaccessible spatiotemporal scales to cell biology research. Simultaneous two-color PALM-dSTORM and PALM-ultrafast (10 kHz) single fluorescent-molecule imaging-tracking has been realized. They revealed the dynamic nanoorganization of the focal adhesion (FA), leading to the compartmentalized archipelago FA model, consisting of FA-protein islands with broad diversities in size (13-100 nm; mean island diameter ≈30 nm), protein copy numbers, compositions, and stoichiometries, which dot the partitioned fluid membrane (74-nm compartments in the FA vs. 109-nm compartments outside the FA). Integrins are recruited to these islands by hop diffusion. The FA-protein islands form loose ≈320 nm clusters and function as units for recruiting FA proteins.

Figure 1.

The objectives of this study are twofold: (1) to establish the optimal conditions for ultrafast PALM of mEos3.2, ultrafast dSTORM of HMSiR, and their simultaneous imaging and (2) to apply these methods and ultrafast SFMI (its development is described in the companion paper) to elucidate the FA architecture and protein dynamics in the FA. The second purpose, together with the previous results, is summarized in the figure. (A and B) Both the apical (dorsal) PM (A, left) and basal (ventral) PM (A, right) are compartmentalized in a nearly identical manner by actin-based membrane-skeleton meshes (fences; brown mesh in B) and rows of transmembrane-protein pickets anchored to and aligned along the actin fence (blue molecules in B), which induce the hop diffusion of virtually all membrane molecules in both the apical and basal PM (B). See the companion paper for these results. (C) Three fundamental questions about the FA molecular organization addressed here are as follows: (1) the characteristics of the FA-protein clusters/oligomers/islands, (2) the higher-order organizations of the FA-protein clusters/oligomers/islands, and (3) the possibility that the fluid membrane part in the FA is compartmentalized, like the bulk basal PM. For details, see D. (D) The three models of the molecular organization in the FA proposed previously (a–c) and the two new models (d and e) proposed here. The new models incorporate the formation of loose clusters of FA-protein islands (d) plus the compartmentalization of the fluid membrane part in the FA (e).

Figure 2.

Optimizing the excitation laser power density and the data acquisition frame rate for the ultrafast PALM imaging of mEos3.2 in living cells. (A) The mean number of photons emitted from a mEos3.2 molecule during an on-period is maximized (49 ± 0.54 photons) at a laser power density of 30 µW/µm² at the sample plane (c), providing a single-molecule localization precision of 29 ± 0.22 nm (d) with an on-period of ≈ 2 ms (a) and a mean number of on-periods before photobleaching of 1.4 (e). Caveolin-1-mEos3.2 expressed at very low levels (<<1 caveolin-1-mEos3.2 molecule/caveola) in the basal PM of T24 cells was imaged using live cells. (a) Histograms of individual fluorescent on-periods (with a gap closing of 1 frame) obtained at various laser power densities at the sample (indicated on the right of b). They could be fitted by stretched exponential functions $\phi \left(t\right)={\phi }_0{e}^{-{\left(t/\tau \right)}^{\alpha }}$, where φ₀ is the prefactor, α is the stretching exponent, and τ is the time constant (Morimatsu et al., 2007; mean ± SEM; SEM was determined as a 68.3% confidence limit for the fitting, which is also the same in b; the numbers of spots observed are the same as those indicated in the boxes in b). (b) Distributions of the numbers of detected photons during an on-period. The histograms could be fitted with single exponential decay functions, with the decay constants providing the mean numbers of detected photons during an on-period. n = number of observed spots. (c) Summary plot for the results in b, showing that the laser power density of 30 µW/µm² at the sample plane provides the maximal number of detected photons during an on-period of a single mEos3.2 molecule (49 ± 0.54 photons). (d) The mean localization precision for mEos3.2 in the basal PM was 29 ± 0.22 (SEM) nm under the optimized laser excitation conditions of 30 µW/µm². The localization precision for each on-period of a single mEos3.2 molecule was estimated using the theoretical equation derived by Mortensen et al. (2010), employing an “excess noise” factor (F) of 1.2 determined for the developed camera system (see Fig. S2 of the companion paper). (e) The distribution of the number of on-events (localizations) for a single mEos3.2 molecule (N) at a laser power density of 30 µW/µm². Each detection was found by examining the proximity of the spots recorded at different frames, with a cutoff time of 3 s (Durisic et al., 2014) and a cutoff distance of 82 nm ($\sqrt{2}\times 2\times$ [mean localization precision for mEos3.2 = 29 nm]). The histogram could be fitted well (green curve) with the geometric function $f\left(N\right)=p\bullet {\left(1-p\right)}^{N-1}$ based on the model for a monomeric blinking fluorophore by Hummer et al. (2016), yielding the P value (fluorophore bleaching probability) = 0.72 and the mean number of detections (on events)/molecule (1/p) = 1.4. (B) The data acquisition at 1 kHz is nearly an optimal frame rate for PALM imaging using mEos3.2 as a probe. The figure shows typical consecutive single-frame images of caveolin-1-mEos3.2 molecules in the basal PM acquired every 1 ms (1 kHz) using an observation laser power density of 30 µW/µm². Based on the number of detections (localizations), caveolin-1-mEos3.2 was found to be expressed at 42 ± 5.2 copies/caveola, which only includes the fluorescent mEos3.2 and not the non-fluorescent mEos3.2, but was normalized by the overcounting of 1.4; n = 15). While many spots exist in only a single image frame, some spots appear in two or three images (spot colors are changed every 1 ms). Using these spot images, the SD of the Gaussian spot profile was determined to be 129 ± 1.3 nm (n = 50 caveolin-1-mEos3.2 molecules). This value was used for reconstructing the diffraction-limited images.

Figure 3.

Ultrafast PALM reveals the shape changes, migrations, and formation/disappearance of caveolae in 3 s in live cells. (A and B) Caveolin-1-mEos3.2 expressed in the basal PM of T24 cells at a density of 42 ± 5.2 caveolin-1-mEos3.2 molecules/caveola (n = 15; see the legend of Fig. 2 B) was imaged in both live (A) and fixed cells (B), using identical ultrafast PALM imaging conditions. Data acquisitions were performed at a rate of 1 kHz for 3 s (3,000 frames), and PALM images were reconstructed using the data acquired for every 1 s (= 1,000 frames; b) and every 0.33 s (= 333 frames; c). (a) Diffraction-limited and PALM images of caveolae in 10 × 10-µm² observation areas, using a data acquisition period of 1 s. For the spatial resolution of these images, see Materials and methods. (b) Enlarged images of the purple-square regions in (a) showing time-dependent changes (every 1 s). The images in the middle column (top and middle rows) are the expanded images of the purple-square regions in a, for the data acquisition between 1 and 2 s. The regions surrounded by yellow squares in the middle row are magnified in the bottom row, and the localizations of single mEos3.2 molecules determined within each 1-s period are indicated by red dots. (c) The PALM image reconstruction performed every 0.33 s for the same caveola that is shown in the bottom row in b.

Figure 4.

Ultrafast PALM is capable of imaging a view field as large as 640 × 640 pixels (35.3 × 35.3 µm²), often encompassing almost an entire live cell, at a data acquisition rate of 1 kHz using mEos3.2 (linked to paxillin; a total of 10,000 frames obtained for 10 s), while at the level of a single FA, it reveals an archipelago architecture of paxillin-enriched islands with a 33 nm mean diameter. (A and B) Typical reconstructed PALM (A) and diffraction-limited (B) images of mEos3.2-paxillin on ≈ 2/3 of the entire basal PM of a T24 cell (data acquisition for a view field of 640 × 640 pixels ≈35.3 × 35.3 µm²). PALM image: 3,526 × 3,526 pixels with a pixel size of 10 nm. Diffraction-limited image: 640 × 640 pixels with a pixel size of 55.1 nm. The yellow contours outline the cell peripheries determined in the diffraction-limited image on the right, using Sauvola’s local thresholding method with a local domain radius of 64 pixels, and k and r values of 0.5 and 128, respectively (Sauvola and Pietikäinen, 2000). Throughout this study, the reconstructions of PALM images with a pixel size of 10 nm were performed using the ThunderSTORM plugin for ImageJ (Ovesny et al., 2014) installed in the Fiji package (Schindelin et al., 2012) and Gaussian rendering with a localization precision of 29 nm (Fig. 2 A-d), whereas diffraction-limited images with a pixel size of 55.1 nm were generated by Gaussian rendering with a spot with an SD of 129 nm (Fig. 2 B). (C and D) Enlarged images of the domains enclosed in squares in A and B, respectively. The contours of the FAs in the PALM image shown in red (C) were determined by using the SR-Tesseler software based on Voronoï polygons (Levet et al., 2015), with a thresholding paxillin number density of 1.45. This contour is overlaid on the diffraction-limited image on the right (D). The contours of the FAs determined from the diffraction-limited image using the minimum cross entropy thresholding are shown in cyan (D). Comparison of the two FA contours indicates that the contour determined from the PALM image is slightly smaller than that determined from the diffraction-limited image and is more sensitive to much smaller FAs (or paxillin clusters outside larger FAs). (E) Further enlarged image from the squared domain in the PALM image in C. (F) The Voronoï polygon diagram of the PALM image in E, showing the contours of the FAs (red) and paxillin-enriched islands (dark green), using a thresholding paxillin number density of 1.45 for both contours. (G) Distribution of the paxillin-enriched island diameters obtained by the Voronoï tessellation analysis using the SR-Tesseler software. The mean diameter of the islands determined here was 59 nm, but after the correction for the effect of the 29-nm single mEos3.2-molecule localization precision on the SR-Tesseler segmentation (Fig. S2 C), the actual mean diameter of the islands was estimated to be 33 nm. (H) Distribution of the number of detections (localizations) of mEos3.2-paxillin molecules per detected paxillin island. The additional x-axis scale shows the estimated total (= mEos3.2-conjugated + endogenous) paxillin copy number per island. Although these values obtained in G and H (the mean paxillin island diameter of 33 nm containing a median of 36 copies of paxillin/island) would be quite informative, they need to be interpreted with caution. First, larger islands might be collections of smaller islands that could not be resolved. Second, the islands containing many other FA proteins, but only <6 paxillin copies, would be missed by definition. Third, paxillin-enriched islands with smaller diameters would be missed even though their diameters are ≥13 nm: This was evaluated by Monte Carlo simulations, which showed that ∼15, 29, and 82% of paxillin-enriched islands with diameters of 13, 20, and 30 nm, respectively, were detectable (when the protein density for the simulation was increased by a factor of 2 to 0.04 copies/nm², ≈30 and 57% of the 13- and 20-nm islands, respectively, were detectable). Approximately, 85–95% of the islands with ≥40 nm diameters were detectable (the localization precisions of 19 and 29 nm hardly affected the result). Taking these limitations into account, based on the data shown in G and H, we propose that the paxillin island diameters are mostly in the range of 13∼100 nm and that the paxillin nanoclusters containing <6 paxillin copies might either exist alone or be located in the islands enriched in other FA proteins. Furthermore, smaller FA-protein islands might contain a total of ≥6 copies of various FA proteins, but the copy number of each constituent protein species might be <6.

Figure S1.

The PRNU of the developed camera system scarcely affects the single-molecule localization precision, examined at 10 kHz. Two image intensifier amplifications were employed: 8,100× (A) and 253× (B). (A and B) (a) Images used for evaluating the PRNU effects of the developed camera system on the localization precision of a single Cy3 molecule. Top: 256 × 256–pixel images representing the PRNU of the developed camera system, obtained by averaging images over 40,000 consecutive frames recorded at 10 kHz under uniform illumination, so that the mean pixel intensity counts became 515 ± 34 (A) and 513 ± 32 (B; SD for 256 × 256 pixels), which are approximately half of the maximum intensity count of 10 bits. The uniform illumination was generated by Köhler illumination, using the halogen lamp of the microscope and a 572- to 642-nm bandpass filter (FF01-607/70, Semrock). Bottom: Modulation of the image of a single Cy3 molecule by PRNU, evaluated by calculation. Left: The Cy3 image was approximated by an ideal two-dimensional Gaussian point spread function (PSF) in the 15 × 15–pixel region, based on an experimentally determined SD of 2.2 pixels for 50 Cy3 molecules immobilized on the glass, obtained by the TIR illumination at 79 µW/µm² and a peak intensity count of 511 (half of the maximum intensity count of 10 bits). In the actual imaging experiments, we employed 55.1 nm/pixel: 2.2 pixels = 123 nm. The PSF peak was placed at the center of the 15 × 15–pixel region. Middle: The 15 × 15–pixel yellow regions shown in the top images are magnified. Right: Images on the left and middle were multiplied pixel by pixel and normalized, generating the PRNU-modulated images of a single Cy3 molecule. (b) The effect of PRNU on the single-molecule localization precision is quite limited. Top: Maps of the 2D position deviation from the pixel center (coded on the gray scale). These maps were generated by moving the 15 × 15–pixel image of an ideal Gaussian PSF simulating the Cy3 image (a, bottom left), scanning over the 256 × 256–pixel PRNU images (a, top) pixel-by-pixel, and calculating the 2D position deviation at every position in the scan. Bottom: Distributions of the 2D position deviations, showing that the mean deviations are 0.033 and 0.034 pixels (= 1.8 and 1.9 nm at 55.1 nm/pixel; n = 57,600 pixels; arrowheads) for A and B, respectively, which are comparable to the typical PRNU effect found with the EM-CCD camera (Pertsinidis et al., 2010). Furthermore, the reverse cumulative distributions (shown in red) indicate that 95% of the 2D position deviations are within the range of 0.066 pixels (= 3.6 nm) and 0.068 pixels (= 3.7 nm) for A and B, respectively. These results suggest that the effects of PRNU on the single-molecule localization precision are limited and almost identical at amplifications of 8,100× (A) and 253× (B).

Figure S2.

Determination of the expression level of mEos3.2-paxillin in the clonal T24 cells stably expressing mEos3.2-paxillin used in this study (0.9× the endogenous paxillin; i.e., total paxillin ≈1.9× of the endogenous paxillin), determination of the thresholding density factor for the Voronoï segmentation analysis, and the correction method in the estimate of the FA-protein island diameters for limited single-molecule localization precisions of PALM and dSTORM images (using Voronoï segmentation analysis). (A) The mEos3.2-paxillin–expressing T24 cell clone used in this work expresses mEos3.2-paxillin at ≈0.9× of the endogenous paxillin in non-transfected cells, and thus the total paxillin amount in these cells will be ≈1.9× the endogenous paxillin without transfection, assuming that the endogenous paxillin expression level is unchanged after the mEos3.2-paxillin expression. The figure shows that the number of mEos3.2 on-events (the number of fluorescent spots in the raw PALM image = the number of localizations) detected in the entire view-field was 330,000 ± 38,000 (n = 9 cells). Based on this value, the amount of expressed mEos3.2-paxillin was roughly estimated in the following way. Assuming that the view-fields employed here represent about two-thirds of the entire basal PM (including the FAs) and that 70∼90% of expressed mEos3.2-paxillin molecules are recruited to the basal PM, and since the mean number of on-events per each mEos3.2 molecule = 1.4 (Fig. 2 A-e) and only ≈60% of mEos3.2 is fluorescent (Baldering et al., 2019), the number of expressed mEos3.2-paxillin molecules is estimated to be 655,000–842,000 copies/cell (3,300,00×[3/2]/[0.7–0.9]/1.4/0.60). The copy number of endogenous paxillin expressed in a T24 cell is unknown, but since 920,000 copies of zyxin and 810,000 copies of VASP, which are both essential components of FAs, are expressed in a T24 cell (Tsunoyama et al., 2021 Preprint), it would not be unreasonable to assume that an approximately similar number of endogenous paxillin copies exists in a T24 cell (say, ≈850,000 copies). Therefore, after the expression of mEos3.2-paxillin, the total paxillin might be over-expressed by factors of 1.8–2.0 ([655 k + 850 k]/850 k∼[842 k + 850 k]/850 k; assuming that endogenous paxillin expression was not decreased due to the overexpression of mEos3.2). Therefore, we will employ the copy number at which the expressed mEos3.2-paxillin is 0.9× the level of endogenous paxillin and the total number of paxillin molecules is 1.9× the level of endogenous paxillin in non-transfected T24 cells. (B and C) Determinations of the proper Voronoï polygon thresholding density factor (B) and the effects of single-molecule localization precisions (19 and 29 nm) on the island diameters evaluated by the Voronoï tessellation analysis (Levet et al., 2015) for the PALM (29-nm precision) and dSTORM (19-nm precision) images (the relationship of the evaluated diameters with true diameters; C), using Monte Carlo simulations of the PALM and dSTORM images of the paxillin islands. Details of simulation and analysis: A circular paxillin island with a given diameter between 20 and 120 nm was placed at the center of a square with a side length 10 times the circle diameter (except for the 20-nm diameter circle where a side length of 20 times the diameter was employed). Single molecule localization errors of 0 nm (B) and 19 and 29 nm (C; for dSTORM of HMSiR and PALM of mEos3.2) were employed. The positions of the fluorescent spots in the PALM/STORM raw images outside the circle were randomly placed at a number density of 0.002/nm² (a typical number of on-events [fluorescent spots in the PALM/STORM raw images] inside the FA in the experimental images), with added Gaussian noise to account for the localization errors (0, 19, and 29 nm). The locations of the fluorescent spots in the PALM/STORM raw images inside the circle were generated in the same way, with the exception that the number density was increased to 0.02/nm², 10× greater than that of the outside density, which is typical of the number density found in experimentally observed islands (0.019/nm² for the identified islands with diameters in the range of 30∼100 nm). The average number densities of fluorescent molecules will be [1/1.4]x and [1/2.7]x of these on-event densities (the densities of the fluorescent spots) for mEos3.2 and HMSiR, respectively (Fig. 2 A-e and Fig. S3 A-d), but since the actual blinking number for each molecule will follow the probability distribution described by the geometric function (Hummer et al., 2016), we included this effect in the simulation. 30 images were generated for each condition, and the islands in each image were detected by the Voronoï tessellation analysis with a minimum diameter of 13 nm, and the diameters of the detected circles were determined (see the main text). The diameters evaluated in this manner (mean ± SEM) are plotted against the true diameters used in the simulation. (B) Determination of the optimal Voronoï polygon thresholding density factor based on the best estimate of island diameters, showing that a thresholding density factor of 1.45 provides the most accurate estimation of the island diameters. The figure shows the evaluated diameters plotted against the true diameters (diameters used for simulation) for various Voronoï polygon thresholding density factors. The single-molecule localization error was set at 0 nm for this determination. From these plots, the closeness of the estimated diameter (d_Estimated) to the true diameter (d_True; dashed line represents d_Estimated = d_True) was estimated as the sum of $({\left[d_{\mathrm{E}\mathrm{s}\mathrm{t}\mathrm{i}\mathrm{m}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{d}}-d_{\mathrm{T}\mathrm{r}\mathrm{u}\mathrm{e}}\right]}^2/d_{\mathrm{T}\mathrm{r}\mathrm{u}\mathrm{e}}^2)$ determined every 10 nm in the range of 30–100 nm (orange text). See the table on the right. (C) The relationship of the evaluated island diameters with the true diameters (open circles). The curves represent the best-fit quadratic functions for the plots of open circles in the range of d_True ≥20 nm, which were used to estimate the true mean diameters of the FA-protein islands in Fig. 4 G, Fig. 5 D, and Fig. S4 A and B. The dashed linear lines indicate the ideal case of d_Estimated = d_True. The filled circles show d_Estimated−d_True (mean ± SEM; for both open and filled circles). The differences increase with an increase of the true diameter, but appear to level off from around d_True ≈100 nm.

Figure 5.

Ultrafast simultaneous dSTORM (HMSiR-labeled Halo-paxillin) and PALM (mEos3.2-paxillin) in live mEos3.2-paxillin–rescued MEFs reveal that FAs contain paxillin-enriched islands of ≈32 nm in diameter with a median of ≈30 paxillin copies, which are not homogeneously distributed in the FA, but form island-enriched domains (loose island clusters) with a diameter of ≈300 nm. (A) Typical ultrafast dSTORM image of HMSiR-labeled Halo-paxillin expressed on the basal PM of a live MEF (data acquisition at 1 kHz for 10 s, with a total of 10,000 frames) and its expanded image. (B) The Voronoï diagram of the dSTORM image is shown on the left. The contours of the FA (red) and paxillin-enriched islands (green) are shown (see Fig. S2 B). (C) Left: A typical overlaid image of simultaneously recorded ultrafast live-cell dSTORM of HMSiR-labeled Halo-paxillin (magenta) and PALM of mEos3.2-paxillin (green) in an mEos3.2-paxillin-rescued MEF. Data were acquired at 1 kHz for 10 s, with a total of 10,000 frames for both dSTORM and PALM (three figures on the right). The paxillin-enriched islands detected by the Voronoï segmentation analysis of the image on the left (the images are superimposed here) exhibited only partial colocalizations of the paxillin-enriched islands, probably due to limited copy numbers of paxillin in the islands. (D) Distributions of the diameters of paxillin-enriched islands obtained from dSTORM and PALM images using the tessellation analysis. The mean diameters were 50 and 58 nm, respectively, but after the correction for single-molecule localization precisions (Fig. S2 C), the true mean diameter of the islands was estimated to be 32 nm for both the dSTORM and PALM results. (E) Distributions of the number of detections (localizations) of mEos3.2- and HMSiR-Halo–linked paxillin molecules per detected paxillin-enriched island. The additional x-axis scale shows the estimated endogenous paxillin copy number per island in the parental MEFs. (F and G) Auto-correlation (F) and cross-correlation (G) functions calculated for all of the fluorescent spots localized in the detected paxillin-enriched islands (polygons; calculated for each FA and the mean ± SEM was obtained). The best-fit functions are shown in color. (H) Schematic FA model, proposing that the FA consists of islands of paxillin (and other FA proteins) with a mean diameter of ≈32 nm, which are not homogeneously distributed in the FA, and instead form loose clusters of ≈300 nm in mean diameter. The FA region outside the paxillin-enriched islands is also probably enriched in paxillin monomers and oligomers (shown by green shading and smaller dots, respectively).

Figure S3.

Live-cell dSTORM data acquisition conditions established for HMSiR-labeled FA molecules. (A) The MEF cells used in this study stably express mEos3.2-paxillin at the level of 0.64× of the endogenous paxillin in the parental MEFs. The figure shows the paxillin Western blot membrane and the plots of paxillin band intensities of the parental MEFs and paxillin-null MEFs (Sero et al., 2011) rescued by the stable expression of mEos3.2-paxillin (cloned), showing that mEos3.2-paxillin–rescued paxillin-null cells (mEos3.2-paxillin–rescued MEFs) express 0.64× the endogenous paxillin in the parental MEF cells. For the dSTORM experiments, the mEos3.2-paxillin–rescued MEFs were further transfected with the Halo-paxillin cDNA for transient expression, and the expressed Halo-paxillin was labeled with HMSiR at ≈ 90% efficiency (Morise et al., 2019). dSTORM observations were performed using live cells exhibiting similar levels of the HMSiR-labeled Halo-paxillin signal. These cells were found to express Halo-paxillin at the level of 0.16× the endogenous paxillin in the parental MEF line, based on the following observations and calculations (and thus these cells express a total of 0.8× the endogenous paxillin; i.e., 0.64× for mEos3.2-paxillin and 0.16× for Halo-paxillin). The ratio of detected spot densities of mEos3.2-paxillin vs. those of HMSiR-Halo-paxillin in the FA was ≈1.4:1. Considering that (1) the number of on-events for a single HMSiR molecule is 2.7 (B d) and that for mEos3.2-paxillin is 1.4 (Fig. 2 A-e), and (2) 90% of HMSiR and 60% of mEos3.2 are fluorescent, the copy number ratio of mEos3.2-paxillin vs. Halo-paxillin including non-fluorescent molecules is estimated to be ≈4.1:1 (1.4/1.4/0.6: 1/2.7/0.9 = 1.667:0.411). Therefore, the amount of Halo-paxillin in these MEF cells is 0.64x/4.1 = 0.16x (0.64x is the ratio of mEos3.2-paxillin vs. endogenous paxillin found in parental MEF cells here) of the amount of endogenous paxillin in the parental cell line. (B and C) First, we will give an overall explanation, and the detailed legends will be presented later. For establishing optimal dSTORM data acquisition conditions, we first examined the on-period durations of HMSiR because this will limit the data acquisition frame rate for dSTORM. With an increase of the TIR excitation laser illumination intensity at a wavelength of 660 nm from 2.2 to 43 µW/µm² in the sample plane, the on-period durations gradually decreased and plateaued at 2.1 ms at a laser intensity of 23 µW/µm² (Fig. S3 B-a), showing that further increases of the laser intensity will not improve the data acquisition frame rate. Therefore, we decided to use a camera frame rate of 1 kHz for dSTORM data acquisition, which is the same rate as that employed for PALM data acquisition using mEos3.2. Further increasing of the illumination laser intensity beyond 23 µW/µm² continued to increase the numbers of detected photons during the on-period of a single HMSiR molecule, with a concomitant improvement of the localization precision of a single dye molecule for a single on-event. However, the extent of improvement was quite limited (Fig. S3 B-b and -c), and thus increasing the laser intensity beyond 23 µW/µm² was deemed not worthwhile, due to the increased probability of photo-damage to live cells. Since the illumination by a 561-nm excitation laser intensity at 23 µW/µm² for 1 min had minimal impact on cell viability (Fig. 2, D and E, in the companion paper), and since this laser intensity is about optimal for the 1 kHz data acquisition rate for HMSiR (on-duration of 2.1 ms; Fig. S3 B-a), we chose to use the 660-nm laser intensity of 23 µW/µm² for the dSTORM experiments. At this laser excitation power density, the on-period reached the plateau at 2.1 ms, providing 477 ± 6.4 photons and a single-molecule localization precision of 19 ± 0.15 nm per on-event (Fig. S3 B-b and -c), and the mean number of on-events per HMSiR molecule was 2.7 (Fig. S3 B-d). At a laser intensity of 23 µW/µm², after the illumination for 10 s (10,000 frames at 1 kHz), ≈80% of HMSiR was photobleached (Fig. S3 C), indicating that the data acquisition for 10 s is close to the optimal conditions for the photon usage. In order to ensure single-molecule detection conditions for all experiments while employing a 660-nm laser intensity of 23 µW/µm² for all ultrafast dSTORM data acquisitions, we adjusted the expression levels of Halo-tagged proteins and/or HMSiR labeling efficiencies. (B) The spontaneous blinking characteristics of HMSiR bound to Halo-paxillin located on the basal PM of live MEF cells observed at 1 kHz (1-ms frame time) with 660-nm excitation laser intensities of 2.2, 6.0, 14, 23, and 43 µW/µm². (a) Duration of on-periods. The histograms show the distributions of consecutive fluorescent on-periods (with a gap closing of 1 frame). They could be fitted by stretched exponential functions $\phi \left(t\right)={\phi }_0{e}^{-{\left(t/\tau \right)}^{\alpha }}$, where φ₀ is the prefactor, α is the stretching exponent, and τ is the time constant (Morimatsu et al., 2007; mean ± SEM; SEM was determined as a 68.3% confidence limit for the fitting; the number of on-events observed [n] is given in c). (b) Distributions of the numbers of detected photons from a single molecule during a single on-period. The histograms could be fitted with single exponential decay functions, with the decay constants providing the mean numbers of detected photons during an on-period (the SEM was given as a 68.3% confidence limit for the fitting). (c) Histograms of localization precisions for individual on-events of single molecules, which were estimated from the numbers of detected photons during a single on-period using the theoretical equation derived by (Mortensen et al., 2010) with an “excess noise” factor (F) of 1.2 determined for the developed camera system (see Fig. S2 of the companion paper). (d) The distribution of the number of detections (on events) for a single HMSiR molecule (N) bound to Halo-paxillin and observed in the bottom PM of a chemically fixed MEF at the laser power density of 23 µW/µm². Here, the fixed cell was used to exclude the effect of the continuous paxillin exchange between the FA and the cytoplasm. Each detection was found by examining the proximity of the spots recorded at different frames (with a cutoff time of 10 s, which is a typical dSTORM data acquisition period for the observations at 1 kHz employed for this work) with a cutoff distance of $\sqrt{2}\times 3\times$ (mean localization precision for HMSiR [19 nm]) = 81 nm. The histogram could be fitted well with the geometric function $f\left(N\right)=p\bullet {\left(1-p\right)}^{N-1}$ based on the model for a monomeric blinking fluorophore by Hummer et al. (2016) (magenta curve), providing the P value (fluorophore bleaching probability) = 0.37 and the mean number of detections (on events)/molecule (1/p) = 2.7. (C) Time-dependent reductions in the numbers of fluorescent spots of HMSiR plotted against the elapsed time after starting the continuous illumination by the 660-nm laser at 2.2 and 23 µW/µm² (frame rates of every 4 and 1 ms, respectively). HMSiR bound to the Halo-paxillin located on the MEF’s bottom PM was detected. The plots represent the sum of the spot numbers in five cells for each condition, normalized to 100% at time 0. The reduction was slower in live cells (blue) than in chemically fixed cells (green), indicating that paxillin in the FA is continuously exchanging with that in the cytoplasm in the time scale of a few tens of seconds, consistent with the previous FRAP data (Legerstee et al., 2019). For longer observations (like 60 s; Fig. 7) we employed an excitation laser intensity of 2.2 µW/µm², whereas 23 µW/µm² was used for shorter observations (like 10 s; Figs. 5 and 6). The total number of counted spots: 1,733,927 in 15,000 frames (blue), 2,476,668 in 15,000 frames (green), and 701,702 in 10,000 frames (red).

Figure 6.

Islands of various FA proteins and their loose clusters of ≈320 nm in diameter. (A) Typical overlaid PALM and dSTORM images simultaneously obtained in a live MEF for mEos3.2-paxillin and five other HMSiR-labeled Halo-linked FA molecules (data acquisition at 1 kHz for 10 s). (B) The contours of the FA and the islands of paxillin and other FA proteins were identified by the Voronoï segmentation analysis of the images in the regions enclosed within squares shown in A (expanded and overlaid on the images). (C) Autocorrelation functions obtained from the dSTORM data for the FA proteins (calculated for all of the fluorescent spots localized in the detected islands; see the legend to Fig. 5 F) with their best-fit functions. Integrins β3 and β1 did not exhibit the longer autocorrelation component, whereas they exhibited short-range auto-correlation lengths (2×ξ₁ values of 70 and 66 nm, respectively) slightly greater than those for talin, FAK, vinculin, and paxillin (2×ξ₁ values of 40∼46 nm). However, these slightly greater 2×ξ₁ values are likely spurious, as the mean diameters of the islands enriched in integrins β3 and β1, after localization error correction, are 57 nm in the island diameter distributions (Fig. S4 A), consistent with those of paxillin, talin, FAK, and vinculin (48–62 nm; Fig. S4 A). Therefore, we conclude that the longer short-range autocorrelation lengths of integrins β3 and β1 (2×ξ₁ values of 70 and 66 nm, respectively) are probably due to the slight mixing of the longer correlation component, which could not be resolved. (D) Crosscorrelation functions between the dSTORM spots of Halo-linked FA proteins and the PALM spots of mEos3.2-paxillin in the detected islands (see the legend to Fig. 5 G).

Figure S4.

Diameter distributions of the islands of various FA-proteins in the MEFs, obtained by the tessellation analysis of the dSTORM images, the effect of elongating the spot merging distance in the dSTORM and PALM image processing on the paxillin island diameter distribution (MEF cells), and representative dSTORM image sequences of HMSiR-labeled Halo-paxillin on the basal PM of T24 cells (similar representative image sequences in MEFs are shown in Fig. 7). (A) Diameter distributions of various FA-protein islands. All FA-proteins were Halo-tagged at their N-termini and labeled with HMSiR. The arrowheads indicate the mean values. After the correction for single-molecule localization precisions of 19 nm for dSTORM images of HMSiR (Fig. S3), the true mean diameters of the islands were estimated to be 24–31 nm. (B) The effects of the merging distances (the threshold distances to identify the detected spots as those representing the same molecule) on the characteristics of the paxillin islands. We generally use $\sqrt{2}\bullet 3$ σ and $\sqrt{2}\bullet 2$ σ (σ = single-molecule localization error) for dSTORM (81 nm) and PALM images (82 nm), respectively (bottom row, which are reproduced from Fig. 5 D). In the bottom row, the results using the extended merging distances of $\sqrt{2}\bullet 5$ σ = 134 nm for dSTORM and 205 nm for PALM are shown. (C) Typical dSTORM image sequences of HMSiR-labeled Halo-paxillin on the basal PM of live (top panel) and fixed (bottom panel) T24 cells, shown every 10 s (data acquisition of 2,500 frames) (top rows) and those of the expanded square regions superimposed by the contours of FAs (white) and paxillin islands (green) determined by the Voronoï segmentation analysis (bottom rows). The results obtained in T24 cells are similar to those obtained by using MEFs (Fig. 7 A and B).

Figure 7.

Synchronized paxillin recruitment to loose paxillin-enriched island clusters detected by the live-cell dSTORM. (A and B) Spatially non-homogeneous paxillin recruitment in time found by dSTORM of live MEF cells. Typical dSTORM image sequences of HMSiR-labeled Halo-paxillin on the basal PM of a live MEF (A) and a fixed MEF (B) are shown every 10 s (data acquisition of 2,500 frames at 250 Hz; top rows). The bottom rows show the expanded images of the regions in squares in A superimposed by the contours of FAs (white) and paxillin-enriched islands (green) determined by the Voronoï segmentation analysis. For the circles in the expanded images of the live MEF in A, see the legend of C. (C) The bottom-left image in A (0–10-s expanded image superimposed by the Voronoï diagram) is further expanded. Here, we placed seven 250-nm diameter circles, which are the plausible loose clusters of paxillin-enriched islands. These seven circles are included in the image sequences in the bottom row in A to show the spatially heterogeneous time-dependent changes of the local fluorescent paxillin concentration. Even outside of the detected paxillin-enriched islands, the FA is enriched in paxillin, which might exist in islands enriched in other FA proteins and/or as smaller paxillin oligomers and monomers. (D) Paxillin recruitment might take place synchronously in each loose cluster, suggesting that the loose island cluster functions as the unit for recruiting paxillin. The figure shows the time-dependent changes of the percentages of paxillin localizations in each circle (loose cluster) shown in C, relative to the total paxillin localizations, during each 10-s sequence in the entire FA. The number of each graph refers to each numbered circle in C. (E and F) Recruited single paxillin molecules frequently undergo rapid movements on the FA membrane, which might induce synchronous paxillin recruitment to paxillin-enriched islands within the loose island cluster. Compare the results obtained in live (E) and fixed (F) MEF cells. The trajectories in the figure show those of newly detected paxillin molecules of four frames and longer (for ≥16 ms; data acquisition at 250 Hz) observed during a 2-s dSTORM data acquisition period (500 frames) to obtain the superimposed 0- to 10-s dSTORM images, which are the same as those shown in A and B (images on the extreme left). The trajectory color is changed every 0.5 s in the 2-s image frame sequence.

Figure 8.

TfR undergoes hop diffusion within the FA region but with a smaller median compartment size of 74 nm (vs. 110 nm in the bulk basal PM) and a longer dwell lifetime of 36 ms (vs. 24 ms in the bulk basal PM). (A) Typical ultrafast single fluorescent-molecule trajectories (0.167-ms resolution; 6-kHz frame rate) of TMR-Halo-TfR diffusing outside the FA (top), entering the FA from the bulk PM (middle), and diffusing inside the FA (bottom). In the middle figure, the background of the trajectory is the mGFP-paxillin image, and the yellow line shows the boundary of the FA region determined by binarization, using the minimum cross entropy thresholding. (B) Distributions of RDs (relative deviation from ideal Brownian diffusion; for the definition and explanation, see Fig. 4 B in the companion paper) for the trajectories of TfR diffusing inside (top) and outside (bottom) the FA region in the basal PM (shaded histograms; 54 and 57 molecules, respectively). The open histogram (the same for both inside and outside the FA) represents the RD distribution for simple-Brownian trajectories generated by Monte-Carlo simulations (5,000 trajectories), with the red and blue vertical lines indicating the 2.5 percentiles from both ends of the distribution (RD_min and RD_MAX, respectively). The trajectories exhibiting RD values below RD_min are categorized into the suppressed diffusion mode. For details, see Fig. 4 B of the companion paper. The data for “Outside FA” shown here in B–D are reproduced from Fig. 7, B–D (top), in the companion paper for the direct comparison with the “Inside FA” data. (C) Distributions of the compartment sizes determined by the hop-diffusion fitting of the MSD-∆t plot of each TfR trajectory (see Fig. 4 A, Supplemental theory 2, and related main text in the companion paper), inside (top) and outside (bottom) the FA region in the basal PM. Arrowheads indicate the median values (statistically significant difference with P = 1.7 × 10⁻⁷, using the Brunner–Munzel test). (D) Distributions of the TfR residency times within a compartment, obtained for each molecule in each compartment determined by the TILD analysis (see Fig. S5 and its legend in the companion paper), with the best-fit exponential curves (providing the dwell lifetimes), inside (top) and outside (bottom) the FA region. A statistically significant difference exists between before and after stimulation with P = 4.9 × 10⁻⁴, using the log-rank test. For the derivation of the single exponential dependence of the dwell lifetime distribution, see the subsection “Expected distribution of the residency times: development of the hop diffusion theory” in the legend to Fig. S5 in the companion paper. (E) Schematic model of our proposed FA architecture, based on the observations made in the present study. The FA-protein islands with various molecular stoichiometric compositions (24–32 nm in the mean diameter in MEFs; green hexagons) form loose clusters with a diameter of ≈320 nm in the compartmentalized fluid (actin meshwork schematically shown by the green lattice). Recruitment of paxillin to these islands might not occur randomly at individual islands but instead occur synchronously at islands within the same loose cluster. The fluid membrane in the inter-island channels in the FA is partitioned into 74-nm compartments (110 nm in the bulk PM). These compartment boundaries are probably composed of the actin-based membrane-skeleton mesh, which might be bound and stabilized by various FA proteins as monomers, oligomers, and islands. (F) Distributions of D_MACRO for TfR determined by the hop-diffusion fitting of the MSD-∆t plot for each TfR trajectory obtained at 6 kHz. Arrowheads indicate the median values (statistically significant difference with P = 9.3 × 10⁻⁹, using the Brunner–Munzel test).

Figure S5.

TfR molecules diffuse within the FA, but in certain subregions, TfR is excluded or has limited diffusibility. Typical trajectories of single TMR-labeled Halo-TfR obtained at a frame rate of 60 Hz (durations of 1.0–5.5 s) are overlaid on the simultaneously recorded TIRF image of the mGFP-paxillin (not a super-resolution image). Trajectories are colored to aid discernment where the trajectories are crowded (the color was not changed in a single trajectory). The contours of the FA regions (yellow lines) were determined by binarization, using the minimum cross entropy thresholding. Previously, our observations at slower frame rates (30–250 Hz) revealed that non-FA protein TfR molecules entered the FA region and diffused more or less freely in the fluid membrane region inside the FA (Shibata et al., 2012; Shibata et al., 2013; Tsunoyama et al., 2018). However, the presence of loose clusters of FA-protein islands shown in Figs. 6 and 7 raises the possibility that TfR molecules might not enter these island cluster domains, or if they do enter, they may not undergo free diffusion there. To address this question, we examined the movements of TMR-labeled Halo-TfR molecules located in and near the FAs at a frame rate of 60 Hz, to determine the long-range, long-time movements, such as those occurring over a time frame of 1.0–5.5 s (Fig. S5). The typical trajectories overlaid on the fluorescent paxillin images (not super-resolution) suggest that these trajectories probably represent the smeared-out (indistinct) hop diffusion of TfR, due to the slow rate of observation (60 Hz; see Fig. 8). The superimposed image in Fig. S5 implies that TfR is somewhat excluded from the areas with elevated paxillin concentrations, and when TfR molecules occasionally entered these areas, their diffusion is slowed and confined, although a quantitative analysis was beyond the scope of this study. These results indicate that the molecular diffusion within the FA is spatially quite heterogeneous, which aligns with the presence of loose FA-protein island clusters. Meanwhile, by applying the hop-diffusion fitting to 250-ms-long TfR trajectories recorded at 6 kHz in the FA (Fig. 8, A–D), we obtained the distribution of the macroscopic diffusion coefficients of individual molecules (representing the diffusion rate over several compartments rather than that within a compartment; D_MACRO; Fig. 8 F). It shows that ≈19% of TfR molecules are almost immobile in the FA, with D_MACRO values <0.0063 µm²/s (Fig. 8 F). This result provides clear evidence for the existence of FA subdomains where TfR diffusion is suppressed. This is likely to occur in the FA subdomains where the FA-protein islands, which act as diffusion obstacles, exist at higher number densities, such as the loose clusters of FA-protein islands.

Figure 9.

Integrin β3 molecules are temporarily immobilized at the paxillin-enriched islands and perhaps on FA-protein islands in general. (A) Typical results of single-molecule tracking of SeTau647-ACP-integrin β3 (recorded at a 4-ms resolution; large magenta spots; images shown every 0.1 s or 25 frames) simultaneously performed with the live-cell PALM of mEos3.2-paxillin, visualizing the paxillin-enriched islands (green spots; the same PALM image is used for all of the images). Data were obtained by using T24 cells. The magenta spots representing single integrin β3 molecules appear much larger than the green spots representing the paxillin-enriched islands because the spot size in single-molecule imaging is diffraction-limited (≈250 nm for single SeTau647-ACP-integrin β3 molecule: the single-molecule localization precision was ≈21 nm, obtained from the SDs of 15-frame immobile trajectories; n = 30), whereas the FA islands were imaged by PALM. The FAs we consider here are indicated by the white contours in the first frames, which were determined by applying the minimum cross entropy thresholding to the PALM images. (a) Three fluorescently labeled integrin β3 molecules (marked as α, β, and γ) appeared in the FA region. The trajectory of the α molecule is shown. Time 0 is set at the time when the α molecule started exhibiting immobilization in an FA. The immobilization of this molecule lasted for 0.19 s (orange frames). The signal intensities of the magenta spots at times 0.3 and 0.4 s are lower, probably due to the stochastic fluctuation of emitted photon numbers and blinking (off-periods) shorter than 4 ms. (b) An integrin β3 molecule was immobilized from time 0 till the end of the observation, for 1.1 s (270 frames at 250 Hz). Large majorities (>90%) of integrin β3 molecules in the FA were immobile throughout the observation period of 1.1 s, consistent with the previous results reported by Tsunoyama et al. (2018). (B) Distribution of the pixel intensities of the mEos3.2-paxillin signal in the FA indicated by the white contour in the first frame in A-a, normalized by its median pixel intensity. The thermographic color scale for this FA is superimposed (in the linear scale between 0× and 6× of the median intensity). (C) The thermographic PALM image of mEos3.2-paxillin shown in A-a, using the color scale shown in B. The immobilized area identified in the trajectory of the molecule α, as previously defined (Simson et al., 1995), is shown by a green 104-nm diameter circle along with the yellow trajectory (also see the expanded figure on the right). The paxillin pixel signal intensity at the center of the immobilized position was measured (open arrow in B), and after this was done for 60 integrin β3 immobilization events, the histogram in D bottom was generated. (D) Top: Averaged normalized distribution of mEos3.2-paxillin signal pixel intensities (the distributions like that shown in B for a single FA are averaged over 55 FAs). Bottom: Distribution of the normalized mEos3.2-paxillin signal pixel intensities at the centers of the immobilization circles of single integrin β3 molecules (such as the green circle in C; n = 60 events), indicating that integrin molecules are preferentially immobilized on the paxillin-enriched islands (refer to the main text). (E) Typical ultrafast single fluorescent-molecule trajectories of integrin β3 molecules observed at 6 kHz, diffusing outside the FA region (top), entering the FA region (middle), and diffusing and becoming temporarily immobilized inside the FA zone (bottom). The middle figure displays a typical result of simultaneous observations of ultrafast (6 kHz) single-molecule tracking of SeTau647-Halo-integrin β3 (colored trajectory) and ultrafast live-cell PALM of mEos3.2-paxillin (islands; 1 kHz, 10-s integration), showing the entry of integrin β3 from the bulk basal PM into the FA region. Consistent with the FA model of the archipelago architecture of the FA-protein islands in the compartmentalized fluid, integrin molecules continued to exhibit hop diffusion when they entered the FA region outside the paxillin-enriched islands.