Spatial organization of proteins in metastasizing cells


  • Daniel Rönnlund and Annica K. B. Gad contributed equally.

Correspondence to: Department of Experimental Biomolecular Physics/Applied Physics, KTH-Royal Institute of Technology, AlbaNova University Center, 106 91 Stockholm, Sweden. E-mail:


The ability of tumor cells to invade into the surrounding tissue is linked to defective adhesive and mechanical properties of the cells, which are regulated by cell surface adhesions and the intracellular filamentous cytoskeleton, respectively. With the aim to further reveal the underlying mechanisms and provide new strategies for early cancer diagnostics, we have used ultrahigh resolution stimulated emission depletion (STED) microscopy as a means to identify metastasizing cells, based on their subcellular protein distribution patterns reflecting their specific adhesive and mechanical properties. We have compared the spatial distribution of cell-matrix adhesion sites and the vimentin filamentous systems in a matched pair of primary, normal, and metastatic human fibroblast cells. We found that the metastatic cells showed significantly increased densities and more homogenous distributions of nanoscale adhesion-related particles. Moreover, they showed an increase in the number but reduced sizes of the areas of cell-matrix adhesion complexes. The organization of the vimentin intermediate filaments was also found to be significantly different in the metastasizing cells, showing an increased entanglement and loss of directionality. Image analysis procedures were established, allowing an objective detection and characterization of these features and distinction of metastatic cells from their normal counterparts. In conclusion, our results suggest that STED microscopy provides a novel tool to identify metastasizing cells from a very sparse number of cells, based on the altered spatial distribution of the cell-matrix adhesions and intermediate filaments.

Survival of cancer patients can be dramatically increased if a reliable diagnosis can be achieved when the lesions still are very small, and before the onset of metastasis. However, current diagnostic practises often require sample amounts not possible to collect at a very early stage. Alternatively, sampling of suspect tumor material in amounts sufficient for a reliable diagnosis is not advisable due to risks of severe sampling-induced side-effects, including dissemination of cancer cells. Compared with surgical excision, fine needle aspiration (FNA) cytology and core needle biopsy (CNB) can for several forms of suspect cancer lesions offer less invasive sampling modalities. However, also for FNA and CNB there is a trade-off between sampling related side effects and diagnostic reliability [1-3]. FNA is patient friendly, minimally invasive, rapid, and cost efficient, but yields small amounts of sample with cells taken out of their tissue context. This makes it more difficult to identify and quantify invasiveness, and the FNA technique relies to a large extent on the experience of the operator and cytopathologist. By CNB, tissue samples are obtained, and therefore typically higher diagnostic sensitivity and specificity can be reached. In contrast, due to the larger needle diameters, CNB also display higher rates of complications, including hematoma, infections, and not the least higher risks of cancer cell seeding along the needle tracts. To circumvent the need to prioritize either diagnostic reliability or minimized sampling-related side effects in the choice between FNA and CNB, and for cancer diagnostics in general, it is strongly motivated to develop the use of highly sensitive, high resolution methods by which the necessary information for a specific diagnosis can be extracted on a molecular level from a minimal number of cells. It is also important to consider all available information retrievable from the individual sampled cells. In addition to cellular protein content, and the up- or down-regulation in the expression of particular cancer-related proteins, characterization of highly resolved, spatial distribution patterns of certain proteins within individual, intact sampled cells can provide important additional diagnostic information. To evaluate the potential of tumor- or metastasis-specific differences in the distribution patterns of specific proteins as diagnostic criteria, we analyzed the subcellular distribution patterns of two protein targets controlling the adhesive and mechanical properties of cells, and implicated in the process of metastasis.

The metastasis of tumor cells is a multistep process that includes local tumor cell invasion into the surrounding tissue, entry to as well as exit from the vasculature and the subsequent colonization of distant organs [4]. During tumor cell invasion, the cells alternate between different modes of migration, either they invade collectively or individually by either a mesenchymal or an amoeboid type of movement [5-7]. One common feature of these different types of movement is that they require altered adhesive and/or mechanical properties of the cells. Therefore, the proteins that regulate these cellular properties can be expected to have altered spatial distribution patterns in the metastasis-competent cells. Concerning proteins modulating adhesive properties, cells in general interact with their surrounding fibrous extracellular matrix via cell adhesion molecules, such as integrins, to which intracellular multiprotein clusters localize at the internal side of the plasma membrane. On ligation of the cell adhesion molecules to the extracellular matrix, protein signaling cascades are elicited in these clusters, which can be detected as phosphorylation of tyrosine amino acid residues [8]. At the relatively low resolution of confocal laser-scanning microscopy (CLSM), areas of increased adhesion show up as cell-matrix adhesion complexes [9], while they can be detected as nanoscale adhesion-related particles with high-resolution electron microscopy [10]. Numerous reports have shown that proteins of the cell-matrix adhesion complexes, including proteins such as integrins, focal adhesion kinase, and src family kinases, are functionally altered in metastatic cells [11-13]. The mechanical properties of cells and tissues have been shown to be regulated by the fibrous intracellular network of intermediate filaments. In particular, the intermediate filament protein vimentin is important for the mechanical integrity of cells and tissues. The vimentin filaments have also been found to bind to the cell adhesion complexes and thereby to stabilize cell adhesion, as well as to regulate cell motility [14, 15]. In general, the cytoskeletal arrangement largely determines the biomechanical properties of cancer cells, which from recent biophysical investigations have been found to strongly correlate with invasiveness [16, 17]. In particular, the vimentin intermediate filament system has been found to control cell adhesion, motility and invasion and to enhance tumor cell metastasis in vivo [18].

Taken together, cumulative evidences suggest that the metastatic process is linked to altered spatial organization of proteins that regulate the adhesive and mechanical properties of the cells and their environment. In particular, the organization of cell adhesions and vimentin filaments seem to play a prominent role. In recent years, stimulated emission depletion (STED) microscopy and other far-field, ultrahigh resolution optical microscopy, also termed optical nanoscopy, techniques have undergone a remarkable development [19]. On the basis of fluorescence-based readouts, they offer high sensitivity, specificity and throughput, and the labeling and acquisition can be done with minor perturbations on the cells, and in comparison to CLSM the spatial resolution can be increased by up to an order of magnitude. In this work, we show for normal, and for genetically modified malignant fibroblast cells that STED microscopy can uniquely resolve and detect differences in the nanoscale spatial distribution of cell adhesions and vimentin filaments underlying malignant development and metastatic competence of these cells, and provide strategies for how these differences can be quantitatively analyzed. We conclude that characterization and analyses of subcellular distribution of these and similar proteins can provide novel means to identify metastasizing cells, new insights into the role of these proteins in the metastatic process and may form a basis for early cancer diagnostics on very limited amounts of cells.

Materials and Methods

Cell Culture Plating

Human primary skin fibroblasts BJ, passage 12, were used as normal cells, and the genetically modified, transformed, invasive, and metastasizing counterpart of BJ cells; BJhTERTSV40TH-RasV12 of passage 19 (BJ-metastazing) [20] as metastazing cells. The cells were maintained in Dulbecco's modified Eagle's medium (DMEM), supplemented with 10% FBS (both from HyClone, Thermo Fisher Scientific, Waltham, MA) in a 37°C CO2 cell incubator, as described previously [21]. Before cell plating, a cover-slip glass was placed at the bottom of a cell culture dish containing DMEM with FBS and the dish was put in the incubator for 15 min. Thereafter, we trypsinized the cells and added enough cells to the dish to form a 60% confluent monolayer within 48 h. To ensure the attachment of single cells to the cover-slip, we resuspended the cells carefully and payed attention not to swirl the dish when placing it back into the incubator. The normal cells adhered within 30 min and the metastazing cells within 45 min after seeding as expected, given that the cover slip was exposed to the extracellular matrix proteins of the FBS-containing medium.

Immunofluorescence Staining and Antibodies

Forty-eight hours after seeding, the cells were maximally spread out on the substrate. We then confirmed that the cells were attached, that they had a normal morphology, and that they had reached the desired confluence. The cells were then placed back into the incubator. After 15 min, the cells were thereafter rapidly washed in 37°C PBS and immediately fixed. For visualization of phosphotyrosine, the cells were fixed in 3.7% paraformaldehyde (PFA) in PBS for 15 min at 37°C, then permeabilized with 0.2% Triton X100 in PBS for 5 min at room temperature. For detection of the vimentin, cells were fixed and permeabilized in a mixture of 3.7% PFA, 0.1% glutaraldehyde (GA) and 0.2% Triton X100 for 15 min at 37°C. This latter fixation protocol is optimized for the visualization of the cytoskeleton, because GA and the joint fixation and permeabilization reduced the background of soluble cytoplasmic components while preserving the cytoskeleton [22, 23]. With the exception for this reduction of diffuse vimentin background signal, the different types of fixation methods did not result in any change of signal intensity or localization. Both fixation solutions were freshly prepared and preheated to 37°C before use. As the cytoskeleton is sensitive to temperature changes, we ensured that the transfer and washing of the cells before fixation was done as rapidly as possible. After fixation, the cover glasses were washed in PBS by gently shaking three times 10 min and were thereafter blocked in 1% BSA in PBS at 4°C over night. The anti-phosphotyrosine antibody (PY99, Santa Cruz Biotechnology, Santa Cruz, CA) was diluted 1:100 and the anti-vimentin antibody (V9, Sigma-Aldrich, St Louis, MO) 1:175 in 0.1% BSA in PBS. The secondary sheep, anti-mouse antibody (conjugated with Atto-590 (Atto-tec GmbH, Siegen, Germany) by Anna Perols, Biotech, KTH, Stockholm, Sweden) was used at a concentration of 4 μg/ml. For detection of filamentous actin, cells were also stained with 150 mM Atto647N-labelled phalloidin (Sigma-Aldrich, St. Louis, MO) in 0.1% BSA in PBS. After the staining, cells were washed by gently shaking in PBS for 30 min at room temperature at dark. The cover slips were then dipped in ddH20, and liquid was removed from the glass, without disturbing or drying the cells attached to the glass, and were thereafter mounted on a microscope slide using a mounting solution of 0.3 mg/ml glycerol and 0.12 mg/ml Mowiol in 60 mM Tris pH 8.5 [20, 21, 24].

STED Microscopy

Images were acquired by using a custom built STED microscope which design has been described in detail before [25, 26]. In brief, the system is based on a supercontinuum laser (SC-450-PP-HE, Fianium, Southampton, UK) from which two excitation beams (570 ± 5 and 647 ± 5 nm) and two STED beams (710 ± 10 and 750 ± 10 nm) are selected. The STED beams are passed through separate vortex phase plates (VPP-1, RPC Photonics, Rochester, NY) before being fed into the objective. Together with circular polarization, the phase change induced by the phase plates creates destructive interference in the STED beams at the center of the focal point of the objective. Thereby, a deep central minima is generated for the STED beams in the focal plane. By overlapping the focus of the hollow STED and that of the Gaussian excitation beam, the fluorophores excited within the excitation volume will be de-excited by stimulated emission by the STED beam, unless they are in the very center where there is no STED intensity. Thus, only fluorescence from the very central part of the excitation beam focus will be detected, which leads to the increased resolution. The acquired resolution of the system is 40 nm for both imaged fluorophores, as determined by the procedure described in [27]. The fluorescent emissions were collected at wavelengths of 615 ± 15 for ATTO 594 and 675 ± 15 nm for ATTO 647N. Both selected dyes have fluorescence lifetimes below 4 ns so by separating the excitation and corresponding STED beams of the two dyes by 40 ns before they reach the sample, the cross-talk between the dyes can be efficiently reduced to only a few percent. The pixel size was kept constant at 20 nm for the STED images and 50 nm for the confocal images (acquired in the absence of STED beams into the sample) with a pixel dwell time of 1 ms. Images were always taken at the lamellipodia of the cells with an image size of 15 × 15 μm2.

Image Processing and Analysis

STED images were deconvoluted by 10 iterations of the Richardson-Lucy deconvolution algorithm, deconvlucy in MATLAB (MathWorks, Massachusetts), using a 40 nm Lorentzian point spread function (PSF). For comparison corresponding deconvolution was performed for the confocal images using a 250 nm Gaussian PSF. In all analyses of the STED images (including size, density, and nearest-neighbor analyses of the adhesion-related particles and adhesion complexes, as well as widths and directionalities of vimentin intermediate filament structures), the full area of the images was always included. All images were recorded so that they included the rim of the cells, and with about the same fraction of the images containing cell-free areas. The fraction of cell-free areas may vary somewhat between the images, but averaged over images of 15-20 cells each, the average fraction of cell-free areas in the images from the control and from the malignant cells should be very similar. Image analysis was performed by custom written code in MATLAB.


Cells were imaged from three separately made experiments from different dates for both adhesion and vimentin data. To account for cell to cell variation, between 15 and 20 cells were imaged for each target and cell-type. The error bars in the figures throughout the article are given as one standard deviation of the investigated set. Two-tailed student t-tests with unequal variance were applied on the comparisons of interest.


To identify novel means to distinguish between normal and metastasizing cells, we used STED microscopy to image cells of two isogenically matched cell lines representing normal and metastasizing human cells, followed by image analysis of the nanoscale organization of the vimentin intermediate filament system and the phosphorylated tyrosines of cell-matrix adhesions.

Imaging and Analysis of Densities and Distributions of Nanoscale Adhesion-Related Particles

Phosphorylated tyrosines as a marker for sites of cell adhesion, together with filamentous actin, were imaged in normal and in genetically transformed fibroblast cells by our custom-built, dual-color STED microscope. This allowed us to specifically analyze the actively binding nanoscale sites of cell-matrix adhesion, and their localization in relation to the actin filaments. According to previous studies by Electron microscopy (EM), the size of a single adhesion-related particle is ∼25 ± 5 nm [10]. By our STED microscope, with a lateral resolution of 40 nm, we could discern and separate all but the most neighboring adhesion-related particles while at the same time benefit from the high sensitivity, specificity, labeling, and throughput offered by the fluorescence readout. To account for cell-to-cell and sample variation we looked at between 15 and 20 cells for each cell-type from three separate experiments. For both the control and the metastatic cells, the adhesion-related particles were found to mainly reside at the end of large bundled actin stress fibers in cell-matrix adhesion complexes. An increased number of cell-matrix adhesion complexes, together with a decreased size of these adhesion complexes could be detected in the metastatic cells when compared with control cells (Fig. 1, top panels). To further quantify these observations, the high resolution images were analyzed with regard to the location, size and density of the adhesion-related particles. The locations of the centers of the adhesion-related particles were determined by the coordinates of their respective intensity peaks in the images (Fig. 2 A2). By thresholding, peaks due to noise could be excluded. To avoid any bias, the same thresholding value was used for all images. The nearest neighbor distance was then measured for all particles as the distance to the closest neighboring particle, and the sizes of the particles were calculated as the mean full width at half maximum (FWHM) value of the peaks, measured for 20 different angles. Measured FWHM values exceeding 60 nm (size of adhesion, labeling and resolution) indicate very close proximity of two or several particles that cannot be individually resolved. From these analyses, it could be clearly quantified that the metastatic cells show an overall increase in the density and a decrease in the measured size of the adhesion-related particles (Fig. 1 A1 and A2). Although the cell-to-cell variations are quite large and can differ by a factor of two for both density and size, there is still a distinct separation between the control and metastatic cells when taking the cell variation into account as a standard deviation (Fig. 1 A3).

Figure 1.

Adhesion-related particles in normal and metastazing cells. Adhesion-related particles (green) and filamentous actin (red) images on BJ and BJ-metastatic cells. Scale bars correspond to 2 μm. Analysis results of the size and density of the adhesion-related particles (A1,A2): Bars show average density and size of the particles, measured in altogether 17 control and 17 metastatic cells. Error bars correspond to one standard deviation and stars indicate student-t-test result showing significant difference between the cell types, *P ≤0.05; **P ≤ 0.01. (A3): Combination of average adhesion-related particle density and size for individual control cells (small black squares) and metastatic cells (small red circles). The combined average values for the control and metastatic cells are included with standard deviations (larger symbols). (A4): Size histogram of adhesion- related particles. The histogram shows that most particles are found around 40-60 nm, which likely corresponds to single adhesion-related particles, while there is a tail of larger sizes, which most likely originates from particles in very close proximity to each other. [Color figure can be viewed in the online issue, which is available at]

Figure 2.

Analysis of cell-matrix adhesion complexes and the distribution of adhesion-related particles inside and outside of these complexes. The analysis comprised 17 control and 17 metastatic cells, with adhesion-related particles labeled green and filamentous actin red (top left). (A1): The locations of the adhesion-related particles are taken as the peak intensities of their respective intensity profiles. (A2,A3): The areas of the cell-matrix adhesion complexes are defined by separately localizing each adhesion-related particle and if that particle has at least four neighboring-particles within 300 nm (Green spots A2), it is included in the area of the cell-matrix adhesion complex (Otherwise shown as White spots A2). By connecting the green spots fulfilling this criterion, a map over the cell-matrix adhesion can be made (A3) and further analysis including the number and size of the zones in the image as well as the adhesion-related particle density and nearest neighbor distance inside as compared to outside can be performed. (B): One-dimensional histograms (B1,B2) and a two-dimensional scatter-plot (B3) summarizing the determined numbers and sizes of the cell-matrix adhesion complexes in normal and in metastasizing cells. (C): Corresponding one-dimensional histograms (C1,C2) and a two-dimensional scatter-plot (C3) showing the densities of adhesion related-particles inside and outside of the areas of the cell-matrix adhesion complexes (D): Corresponding histograms (D1,D2) and scatter-plot (D3) displaying the determined nearest neighbor distances between the particles inside and outside of the cell-matrix adhesion complexes. (D4,D5): Nearest neighbor histograms of adhesion related-particles in control and in metastatic cells, with the amount of particles found inside the cell-matrix adhesion complexes (red), and the remaining adhesions found outside (blue). [Color figure can be viewed in the online issue, which is available at]

Number and Sizes of the Cell-Matrix Adhesion Complexes and Their Fraction of Adhesion-related Particles

Next, we investigated if, and to what extent the larger areas with increased adhesion, that is, the areas of the cell-matrix adhesion complexes, differed between normal and metastasizing cells. In the recorded STED images, cell-matrix adhesion complexes could typically be found at the end of actin stress fibers and could be identified as areas of increased densities of adhesion-related particles (Fig. 1, enlarged areas). Thus, these areas can be identified in the images by locating areas of increased density of adhesion-related particles (Fig. 2 A1-A3). More specifically, the cell-matrix adhesion areas were defined as the combined area of neighboring adhesion-related particles, which have at least four other particles within a 300 nm radius (Fig. 2). These areas show up as homogenous blurred intensity areas in the confocal images (Fig. 1).

The metastatic cells showed an increase in number (from an average of 7 to 11 per 15 × 15 μm2) and a reduction of size (from ∼1.1 to 0.8 μm2) for the cell-matrix adhesion complexes (Fig. 2 B1 and B2). Approximately half of all adhesion-related particles (54% for control and 48% for metastatic cells) were found to reside inside the cell-matrix adhesions complexes. To further investigate the overall higher density of adhesion-related particles found in the metastasizing cells (Fig. 1 A2 and A3), and to further resolve differences between the cell types, we analyzed these densities inside and outside of the adhesion complexes in the cells. For both the normal and the metastatic cells the densities of these particles were found to be more than one order of magnitude higher in the cell matrix adhesion-complex areas when compared with outside of these areas (Fig 2, C1 and C2). Although the densities of the adhesion-related particles inside the cell-matrix adhesion complexes were very similar for control and metastatic cells, the densities outside of these complexes were significantly higher for the metastatic cells (Fig. 2 C1, C2, and C3). This indicates that the overall increase of adhesion-related particles in the metastatic cells (Fig. 1 A2 and A3) is mainly due to additional adhesions outside of the cell matrix adhesion-complexes. When considered as an average over the whole image, the nearest neighbor distance between the adhesion-related particles was found to be very similar for the metastatic cells, when compared with the control cells (172 ± 18 nm, 168 ± 10 nm, respectively). Also, when separately considering the neighbor distance inside and outside of the cell adhesion complex areas only slight differences could be discerned, where the metastatic cells have longer average nearest neighbor distances inside these areas when compared with control (104 ± 8 nm, 98 ± 7 nm, respectively) and shorter distances outside of these areas (232 ± 33 nm, 262 ± 34 nm respectively) (Fig. 2 D1, D2, and D3). Histograms of the nearest neighbor distances of the adhesion-related particles could more clearly display differences in the particle distributions inside and outside of the cell-matrix adhesion complexes (Fig. 2 D4 and D5). Hence, the STED images and subsequent analyses can reveal distinct differences between normal and metastasizing cells, both with regard to the size and density of their cell-matrix adhesion complexes, as well as to the distribution of adhesion-related particles inside and outside of these complexes.

Imaging and Analyses of Vimentin Intermediate Filament Fibers

In order to compare the nanoscale structural differences of the vimentin intermediate filamentous system in the control and metastatic cells, a set of STED images were acquired for these cells, with vimentin and filamentous actin labeled via secondary antibodies and phalloidin, respectively (see Materials and Methods section). Already a first overall inspection of the images indicates a more chaotic structure of the filaments in the metastatic cells, when compared with the control cells, where especially the direction and entanglement of the filaments look different (Fig. 3, upper panels). In order to quantify these differences, a MATLAB-based analysis software was developed that determines the angle where most of the vimentin fibers are oriented, as well as the entanglement of the fibers. The direction of the fibers was calculated by converting the standard X-Y image to an angle-bin image by using the radon transformation [28]. A structure that is oriented in a certain angle in the X-Y image will when projected to the bin at the corresponding angle in the angle-bin image give rise to a high pixel value. The same fiber will when projected for all other angles yield much lower values for the corresponding angle-bin pixels. However, the projection of the fiber will be distributed over more pixels and the sum of the pixels for a certain angle will be the same. We found out that by taking each pixel value to the power of three, the high pixel values corresponding to the structure orientation are enhanced, and a summation of the pixels for each angle gives a good estimate of the direction of the structures in the X-Y image (Fig. 4). In order to get a single value indicating the direction of the fibers in the image, a simplification was made where we defined all fibers oriented in the most frequent direction ± 10° to be parallel. The amount of parallel fibers can then be compared to the total amount of fibers in the image (ratio of peak angle ± 10°), giving a value ranging from 1 if all fibers are oriented within 10° to 0.12 where all directions are equally probable (as the peak angle ± 10° will then give a value 21/180 = 0.12). Determining the direction of the fibers in the high-resolution STED images by this procedure, makes it possible to clearly distinguish parallel from more randomly oriented fibers (Fig. 5). With this procedure, the direction of the fibers could be distinguished to be less parallel in the metastatic cells, when compared with control cells (Fig. 3 A1).

Figure 3.

The nanoscale vimentin intermediate filament structures in normal and metastasing cells. Top: Vimentin (green) and filamentous actin (red) on BJ and BJ-metastatic cells, scale bars 2 μm. The difference in resolution between the STED and confocal images of vimentin is exemplified in the lower line images, and corresponds to the regions within the white-lined rectangles in the full images. (A): Analysis results based on full images as above, from 20 metastatic and 18 control cells. (A1,A2): Bars showing average direction (ratio of peak angle ± 10°) and entanglement (average structure width) of the vimentin fibers of control cells as compared to metastatic cells. Error bars correspond to one standard deviation and stars indicate student-t-test result showing significant difference between the cell types, *P ≤ 0.05; **P ≤ 0.01. (A3): Graph showing the average vimentin width and direction value of each individual cell, as well as the total average with standard deviations. Right histogram (A4) shows the measured widths of all images of the control cells and the metastatic cells. [Color figure can be viewed in the online issue, which is available at]

Figure 4.

Algorithms for assessing the fiber directions. The pixels in the X-Y images (a1,b1,c1) are projected onto a line in 180 different angles (with 0° defined as the direction of a vertical line in (a1,b1,c1) and collected in bins (radon-transform). The resulting angle-bin image (a2,b2,c2) will have the highest pixel values where the projected angle corresponds with the structure direction in the X-Y image. To extract these angles of predominant structure direction, each pixel value in the angle-bin images (a2,b2,c2) is taken to the power of three and these cubed pixel values are then summed for each angle to create a graph, which more clearly shows the predominant orientation(s) of the structure (a3, b3, c3). The lower images (c1,c2) show an X-Y image of five lines with an angle α = 10° between them, and the corresponding angle-bin image. The graph (c3) shows a normalized distribution of the structure orientation where α is varied between 0°,5°,10°, and 30°. [Color figure can be viewed in the online issue, which is available at]

Figure 5.

Calculation of fiber directionality, an example. Image showing how the direction of fibers is calculated and comparison of parallel fibers (a) to more chaotically oriented fibers (b). As can be seen the STED image provides a much clearer distinction between the parallel fibers and the more chaotically oriented fibers where the area under the curve for the peak angle ± 10° goes from 0.77 (parallel) to 0.27 (chaotic) for STED and 0.65 (parallel) to 0.33 (chaotic) for the confocal image. [Color figure can be viewed in the online issue, which is available at]

The entanglement of the vimentin fibers can be measured by looking at the width of their intensity profiles in the captured images. The widths of the vimentin fibers were calculated by using the skeleton and pruning functions in MATLAB to create a map of the filament network. The width is then calculated at many positions along this network as the minimum FWHM value measured over 20 angles (Fig. S1 Supporting Information). Since a single vimentin filament is ∼10 nm in diameter, entangled filaments can not be separately resolved, neither in confocal, nor in STED microscopy images, and appear in both type of images as separate but wider fibers [14]. However, the difference in apparent widths is much larger when comparing single fibers with entangled fibers in the high-resolution STED images when compared with the lower resolution confocal images (Fig. 6). Consequently, fiber entanglement can be far better determined from the high resolution images. In these images, the width of the fibers was calculated on ∼1000 different positions for each image, that is, each fiber was measured several times on different locations in the image to get a good average. By this procedure, the width of the fibers was found to be significantly higher for the metastatic compared with the control cells, indicating a clearly higher degree of entanglement in the metastatic cells (Fig. 3 A2). Histograms over the fiber widths show that most of the fibers for both the metastatic and control cells were between 40 and 80 nm wide, which indicate mainly single filaments However, the occurrence rate of larger fiber widths confirms that entanglement is clearly more abundant in the metastatic cells when compared with in control cells (Fig. 3 A4). Taken together, the vimentin image data indicate that the vimentin filaments in metastasizing cells have a more random orientation and are more bundled together into fibers compared with in normal cells. When combining the direction and entanglement data the coordinates of the structure width and ratio peak angle parameters can be clearly separated for the metastatic and the control cells (Fig. 3 A3). Hence, the normal and metastasizing cells could be clearly distinguished with regard to the nanoscale, spatial organization of the vimentin intermediate filamentous system in the cells.

Figure 6.

STED versus confocal detection of entanglement. Image showing that STED is much more sensitive in detecting fiber entanglement than confocal imaging. STED FWHM of single fiber is 47 nm increasing to 94 nm ( + 100%) where four fibers have come together while the corresponding confocal width is 161 nm for the single fiber and 178 nm for the bundled fibers ( + 11%). [Color figure can be viewed in the online issue, which is available at]


Super-resolution techniques, such as EM, STED, dSTORM, and PALM have previously been used for imaging sites of cell-matrix adhesions [10, 29-32]. From our data, we could analyze tyrosine phosphorylated sites of adhesion and the calculated densities correspond well with those determined from EM images of integrin-labeled cells [10]. The altered adhesive properties of the invading tumor cells can induce tumor cell metastasis. Our data show that the capacity of cells to metastasize can be linked to a significant increase of the densities, and a more homogenous distribution of nanoscale adhesion-related particles within the cells. Compared with control cells, the metastasis phenotype was also accompanied by an increase in the number and a reduction in the size of the cell-matrix adhesion complexes. We have previously shown that a decrease in size and an increase in the number of cell-matrix adhesion complexes correlates with an increased contractile force of cells [33]. Metastasizing tumor cells have been shown to have a reduced capacity to adhere and an increased ability to migrate [7, 34]. This, together with our previous finding suggests that an increased density and a more homogenous distribution of cell adhesion-related particles leads to reduced cell adhesion and increased contractility, that in turn provide cells with increased migration capabilities.

The process in which a normal epithelial cell transforms into a malignant mesenchymal-like cell, the epithelial to mesenchymal transition (EMT), is the initial step in tumor cell invasion. During this process, the intermediate filaments in the epithelial cells are replaced by vimentin, which for decades has been used as a clinical marker for tumor cell invasiveness [35]. However, the functional importance of the vimentin in invasive tumor cells still remains elusive [36]. The vimentin intermediate filament has been imaged previously by far-field fluorescence nanoscopy methods such as by STED and PALM, see for example [37, 38]. However, those studies described the far-field nanoscopy methods and did not aim to answer biological questions, or to quantify differences in the filament features between normal and malignant cells as a possible basis for cancer diagnostics. Our data show that the vimentin intermediate filaments are differently organized in the metastasizing cells when compared with control cells. Whereas, the control cells often showed a well-structured network where most of the fibers were oriented as single fibers in similar direction, the filaments were oriented in a more disordered fashion in the metastatic cells, and they displayed more entanglement between the fibers. In the light of the previous findings that the recruitment of vimentin fibers to sites of cell adhesion promotes adhesive strength, our data suggest that the adhesions in metastasizing cells are weaker and less stable than in normal cells [39]. In agreement with this, reduced cell adhesion has an increased focal adhesion turn-over, as have previously been linked to increased cell motility [15, 40, 41]. Taken together with our findings, this suggests that an increased entanglement, changed directionality and a more disorganized arrangement of vimentin fibers reflects the adhesive and mechanical properties of cells that increase cell motility, invasion and metastasis.

The analyzed protein distribution features and the differences found are indeed likely to reflect adhesive and migratory properties of many types of malignant cells. However, cancer is a heterogeneous disease. The differences found between the cell types in this study, and for the particular proteins studied, should thus rather be considered as an example. They demonstrate the principle that far-field nanoscopy, in combination with quantitative image analyses, makes it possible to exploit spatial distribution patterns of proteins in individual cells, not resolvable by confocal imaging, for diagnostic purposes. Given the principle, further studies are needed to find out to what extent the identified features can be found in other malignantly transformed cells, to what extent other proteins and spatial distribution patterns can be used diagnostically, and to identify corresponding parameters in corresponding normal cell types as references.

To summarize, far-field fluorescence nanoscopy offers a combination of spatial resolution and molecular specificity not possible to obtain by either state-of-the art confocal immunofluorescence imaging, or by electron microscopy. We have developed fluorescence nanoscopy and subsequent analyses on fibroblast cells to objectively and quantitatively analyze structural and spatial distribution features in the cell surface adhesions and vimentin intermediate filaments. Our data make it possible to understand more in detail the underlying mechanisms rendering cells metastasis competent. The ability to resolve and analyze this altered spatial organization of nanoscale sites of cell adhesions and of vimentin fibers also suggests that far-field nanoscopy can provide novel means to identify metastasis competent cells. For our analyses, nanoscopy images taken on 15-20 cells each from the normal and metastasizing cell types were enough to clearly tell them apart. Further optimization of targets and analyses are likely to further reduce this number. Given the rapid ongoing development of fluorescence nanoscopy techniques and the dissemination of these techniques outside of advanced imaging laboratories, these single cell analyses are also soon likely to be implementable in clinical environments.


The authors thank A. Perols and A. Eriksson–Karlström, KTH, Stockholm for provision of fluorescent antibodies, and Stefan W. Hell, Lars Kastrup, and Andreas Schönle, MPIBPC, Göttingen for important support in the build-up of the STED instrument.