Molecular diversity of cell-matrix adhesions

Focal contacts (FC) are specialized adhesion sites formed between cultured cells and the underlying extracellular matrix (ECM) (Burridge and Fath, 1989; Geiger et al., 1995; Gilmore and Burridge, 1996; Yamada and Geiger, 1997). In these regions, originally identified by electron microscopy (Abercrombie et al., 1971; Heath and Dunn, 1978) or by interference reflection microscopy (Abercrombie and Dunn, 1975; Izzard and Lochner, 1976), cells form close and tight adhesions to the underlying substratum. At their cytoplasmic aspects, FC are associated with bundles of actin filaments through an electron-dense, multi-molecular complex, the sub- membrane plaque. In typical adherent cultured cells a large number (tens to hundreds) of distinct FC can be detected, ranging in size from less than a square micron to several square microns. The number of FC, their size and distribution, can vary greatly from one cell to the other or even within a single cell, and their morphological diversity may be affected by multiple factors including the nature of the substratum, composition of the culture medium, incubation time after plating, and cell density.

During the last several years, considerable information has accumulated on the molecular composition of FC. It was shown that the transmembrane ‘adhesion receptors’ present in these sites are members of the integrin superfamily (Hynes, 1987, 1992; Schwartz et al., 1995). These molecules react, through their ecto-domains, with specific ECM molecules, and through their cytoplasmic tails with components of a submembrane plaque that links them to the microfilament system (Geiger et al., 1995; Yamada and Geiger, 1997). This plaque contains a multitude of proteins, some of which participate in the physical linkage of the membrane receptors to the cytoskeleton, while others are signal transduction molecules, including different protein kinases, their substrates and various adapter proteins (Geiger et al., 1995; Jockusch et al., 1995; Kam et al., 1995; Yamada and Geiger, 1997). It is believed that the recruitment of these molecules to FC and their activation plays a central role in the generation of ‘adhesion signals’ that are involved in the regulation of many cellular processes such as cell growth, differentiation or apoptosis (Clark and Brugge, 1995; Schwartz et al., 1995; Yamada, 1997; Yamada and Geiger, 1997). Recent studies have indicated that the assembly of FC is also stimulated by local tyrosine-specific protein phosphorylation and that the phosphorylation events are triggered by tension applied to the FC-anchored adhesion complexes by the attached bundles of actin filaments (Bershadsky et al., 1996).

It is intriguing that despite the extensive work for over two decades and the vast information available on FC formation and organization, a definitive and unequivocal molecular characterization of these structures is not yet available. For example, it is not clear whether all submembrane plaques are basically equivalent at a molecular level, nor how their composition is regulated. An application of computerized morphometry was reported recently to analyze gravitational effects on vinculin and FC morphology (Usson et al., 1997), but little is known about the molecular classes, organization and quantitative dynamics of these adhesion sites.

To approach these questions, we have developed a microscopic system that enables us to perform detailed quantitative morphometry of matrix adhesions. In this report we describe this experimental approach and its application to evaluate two hypotheses: (a) cell-matrix adhesions may not be uniform, but may instead consist of a range of distinct structures that differ at the molecular level between and within individual adhesions; and (b) it may be possible to switch on or off rapidly the phosphorylation and assembly state of mature cell-matrix adhesions within minutes.

For rigorous, quantitative testing of these and other concepts in cell-matrix adhesion, we present here a detailed procedure for automated identification of matrix adhesions and characterization of their properties including their size, axial ratio and labeling intensity for their various constituents, as well as the calculation of fluorescence ratio imaging (FRI). This digital microscopy approach was applied here to cultured fibroblasts to test for the extent and types of molecular heterogeneity of their matrix adhesions. The results revealed an unexpected level of variability in the composition of these matrix adhesions, extending even to different regions within the same adhesion site. Our findings establish three general categories of adhesion sites within a broad spectrum of populations of matrix adhesions, namely ‘classical FC’ which are extensively labeled for paxillin, vinculin and phosphotyrosine, ‘fibrillar adhesions’ which contain relatively low levels of these characteristic proteins and instead contain high levels of tensin, and ‘mosaic adhesions’ which contain a combination. We also show that treatment with the kinase inhibitor H-7, which blocks actomyosin contraction (Volberg et al., 1994; Tian et al., 1998) leads to a rapid loss of phosphotyrosine from FC, which is then followed by their dissociation. We further show that very shortly (<5 minutes) after H-7 withdrawal, well-formed FC with apparently normal morphology and composition suddenly reconstitute, indicating a high degree of potential plasticity.

Rat embryo fibroblasts (REF52) were plated on glass coverslips in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal calf serum. For immunofluorescence labeling cells were fixed-permeabilized with 3% paraformaldehyde, 0.5% Triton X-100 in PBS for 2 minutes, and post-fixed with 3% paraformaldehyde in

PBS for 30 minutes. The cells were then washed with PBS and stained with the primary antibodies for 1 hour. After washing with PBS, the cells were stained with the secondary antibodies, namely Cy3- or FITC-conjugated goat anti-mouse or goat anti-rabbit IgG. Finally, the cells were washed with PBS and mounted in Elvanol (Mowiol 4-88, Hoechst, Frankfurt, Germany).

To inhibit cellular contractility, REF52 cells were incubated with 60 μM 1-(5-iso-quinolinylsulfonyl)-2-methylpiperazine (H-7; I-7016, Sigma Chemical Co., St Louis, MO, USA) for different periods of time (Volberg et al., 1994; Tian et al., 1998). The cells were then fixed and double-labeled as described. Recovery was achieved by removal of the H-7-containing medium and further incubation of the cells in normal growth medium.

The primary antibodies used in this study included: rabbit anti-vinculin (R695, Geiger 1979), rabbit anti-phosphotyrosine (PT40, kindly provided by Israel Pecht and Arie Licht, The Weizmann Institute), monoclonal antibody (mAb) anti-vinculin (clone hVin-1, Sigma Chemical Co., St Louis, MO, USA), and mAbs anti-paxillin and anti-tensin (Transduction Laboratories, Lexington, KY, USA). Secondary antibodies (all from Jackson ImmunoResearch Laboratories Inc., West Grove, PA, USA) were Cy3-conjugated goat anti-rabbit F(ab′) ₂ fragment, Cy3-conjugated goat anti-mouse IgG H+L, FITC-conjugated F(ab′) ₂ fragment goat anti-mouse antibody and FITC-conjugated goat anti-rabbit IgG F(ab′) ₂ fragment.

Full-length chicken vinculin cDNA (Bendory et al., 1989; Price et al., 1989) and full-length human paxillin cDNA (kindly provided by K. Nakata, S. Miyamoto and K. Matsumoto, National Institute of Dental and Craniofacial Research, National Institutes of Health, Bethesda, MD) were ligated in-frame into pGZ21 vector according to the method of Kioka et al. (1999). pGZ21 contains a cytomegalovirus promoter and a green fluorescent protein (GFP) coding sequence with three amino acid residue substitutions (S65A, V68L, and S72A) to increase its fluorescence in mammalian cells (Cormack et al., 1996). REF52 cells were transfected using the calcium phosphate method (Sambrook et al., 1989). Expression of each fusion protein was verified at the protein level by western blotting utilizing monoclonal antibodies to vinculin or paxillin, as well as anti-GFP monoclonal antibody (data not shown). As an additional control, full-length chicken tensin was also expressed as a GFP fusion protein (B.-Z. Katz, K. Matsumoto, S. Aota, D. Lin, S. Lin and K. M. Yamada, unpublished).

Specimens were examined using an Axioskop microscope (Zeiss, Oberkochen, Germany) with selective fluorescence excitation and emission interference filters and quad-dichroic mirror (Chroma, VT, USA) that were positioned in the optical path by computer-driven filter wheels. Images were acquired using a scientific-grade cooled charged-coupled device (CCD) camera (Model C220, Photometrics Co., AZ, USA) with Texas Instruments 1024×1024 pixels chip readout generating 12-bit digital data. Image acquisition and processing were controlled by a Silicon Graphics workstation model 4D/35 (Mountain View, CA, USA) using modified Resolve3D and Prism software (Chen et al., 1990, 1995; Kam et al., 1993, 1997). This system is currently produced by Applied Precision Inc. (Issaqua, WA, USA). In the present study cells were examined with a ×100/1.3NA plan-Neofluar objective (Zeiss, Oberkochen, Germany) resulting in a pixel length of 0.118 μm. In order to correct for non-homogenous illumination and pixel-to-pixel variations in CCD sensitivities, correction files were prepared from a slide containing a thin layer of homogenous solution of rhodamine, using the Resolve3D software, and applied during the acquisition session. To ensure accurate alignment between the Cy3 image and the FITC image, cells were immunofluorescently stained for vinculin using a monoclonal anti-vinculin antibody followed by a mixture of Cy3-conjugated and FITC-conjugated goat anti-mouse antibodies. This slide was then used to measure, and correct, small shifts between the Cy3 and the FITC images. These shifts were found to remain constant during the acquisition session (Kam et al., 1995).

Matrix adhesions, immunolabeled for their constituent proteins, appear as local intensity maxima (‘patches’) at the substrate focal plane. However, images often contain diffuse background, which is especially prominent in thick regions of the cell. Subtraction of the averaged image intensity within a rectangular box around each pixel (Eq. 1, Fig. 1), for a box size larger then matrix adhesions, but small compared to cell dimensions, ‘flattens’ this background (‘high-pass filtered image’, ⁠) and facilitates interactive determination of a uniform threshold level, ⁠, that is well below staining intensity for all matrix adhesion patches, yet is above noise. The box-filter was chosen since it can be implemented in a fast algorithm. The box size used in this study was 4.7×4.7 μm. Following segmentation (see below); intensities and background levels were calculated from the unfiltered P_ij images.

Image segments corresponding to individual matrix adhesions (‘patches’) were defined using a segmentation algorithm we named ‘water’, which is based on a modified ‘lake’ algorithm (Soferman, 1989). Usually, all adhesion sites in 4 representative cells were used as a basis for generation of quantitative data sets for each treatment. The water algorithm can be visually described as the emergence of peaks and their associated slopes as water recedes from a hilly landscape after a flood. As water declines, new peaks appear (local maxima) and existing hills expand their area. When two previously disjoined mountains meet, the decision to merge them, or to regard them as separate identities, depends on their size. The algorithm is implemented for 2-D and 3-D images for any desired pixel connectivity, but was applied here in 2-D with 4-neighbors connectivity (i.e. each pixel has 4 touching neighbors: up, down, right and left; all the other diagonal pixels are not considered connected). The inputs for the water routine are: the filtered image ⁠, a threshold for minimum pixel intensity ⁠, a threshold for minimum patch area and a critical patch size for merging ⁠. The output assigns a patch number M_ij to each pixel (ij), or zero to pixels that do not belong to any patch. The set of instructions in the water routine can be described as follows:

1. Sort the (ij) pixels in the high-pass filtered image according to their intensity ⁠, forming a list in a decreasing order.

2. Cycle over pixels (ij) in the sorted order, until ⁠. At each cycle, the pixel (ij) either:

   Case 1. Forms a new patch, if it is not touching any existing patch.

   Case 2. Is assigned to an existing patch, if it is touching one.

   Case 3. Merges patches, if it is touching more then one, and the size of at least one of the connected patches is below the critical area for merging (otherwise it is assigned to the first patch it touches).

3. Exclude all patches smaller than the minimum area for patches ⁠.

The critical area for merging was typically set at 40 pixels (0.56 μm ²). This value was empirically found to prevent fragmentation of matrix adhesions due to their internal structure, yet allowed separation of adjacent, yet distinct, matrix adhesions. To identify very large matrix adhesions with a granular texture, the critical area for merging was set to a much higher value of 300 pixels (4.2 μm ²).

The water-generated mask, M_ij (Fig. 1) allows calculation for each matrix adhesion (‘patch’) of various morphometric and fluorescent attributes based on the original image data (Fig. 2). The morphometric parameters used in this study include: area (Eq. 2), center of mass (Eq. 3), axial ratio of best-fitted ellipse (Eq. 4), intensity (Eq. 5), and the ratio between the intensities of two fluorophores (i.e. Cy3 and FITC) for each patch (for double-labeled samples). The equations used for calculating these parameters are listed below:

Equation 2

Equation 3

Equation 4

Equation 5

Images of cells, double-labeled by immunostaining for two protein components of matrix adhesions, were acquired, high-pass filtered and aligned as described above. The high-pass filtration is critical for FRI since it abolishes the contribution of the background to the matrix adhesion intensity, and since it enables separation of background noise from the signal by a single threshold level. The ratio image was then computed and displayed (as schematically shown in Fig. 3). In order to avoid large ratio values due to low-intensity pixels in the denominator, the ratio is clipped below some set threshold. Nevertheless, it is important to include in the ratio image information about the presence of labeling in the numerator, even when the denominator is below threshold. For this purpose the ratio image was calculated with two separate thresholds for the numerator and denominator images, and assigned to each pixel in the ratio image values according to the following considerations: (A) If both numerator and denominator images were below threshold the ratio image was set to zero. (B) If numerator was below threshold, but denominator was above threshold, the ratio image was set to 0.1. (C) If the numerator was above threshold, but the denominator was below threshold the ratio image was set to 10. (D) When both images had pixel values above their respective thresholds, the ratio value was displayed in spectral, log scale, color look-up table (blue at minimum to red at maximum), which permitted presentation of ratio value variations over two orders of magnitude (from 0.1 to 10). To utilize this range optimally, and to compensate for the differences in binding constants and photon yields of different secondary antibodies, all the ratios were normalized linearly by a constant that shifts their average toward a ratio of 1. The ratio image was then scaled from 0.1 to 10, assigning a black color to case A, blue to case B, red to case C, and a color corresponding to the spectral scale that corresponds to the ratio between the Cy3 intensity and the FITC intensity to case D.

FRI of GFP-tagged versus immunofluorescently-labeled FC was performed by transfecting REF52 cells with cDNA encoding GFP-fusion proteins of vinculin, paxillin or tensin. Forty-eight hours later the cells were fixed and immuno-labeled with antibodies to vinculin, paxillin or tensin (respectively), and Cy3-conjugated secondary antibodies. FRI between the GFP and the Cy3 images was performed as described above.

FC of REF52 cells were labeled for vinculin and automatically identified as described above (Fig. 1). The raw images of such cells were often quite noisy, displaying diffuse staining in the nucleus or in a relatively thick region of the cytoplasm (Fig. 1, ‘original image’). This background labeling was reduced using the high-pass filter, generating a relatively high-contrast image without losing small or faint FC (Fig. 1, ‘filtered image’). Using the water algorithm, individual matrix-adhesion sites could be identified and marked either by a dot on the original raw image (not shown) or by random color presentation (Fig. 1, ‘patched image’). The water system was capable of distinguishing close or even touching, yet distinct, adhesion sites. Rarely, apparently single adhesions were subdivided by the water algorithm into two or more smaller FC due to the presence of fine substructures within the FC itself. Such FC could be merged manually. For each matrix-adhesion site, the area, axial ratio, and intensity of patches recognized by the water software were calculated (Fig. 2). In order to segment very large and granular adhesion sites (Fig. 2B,C) we used a different set of threshold values for water segmentation from those used to recognize the more homogeneous and smaller FC.

To determine whether the immunofluorescence labeling intensity indeed represents the local concentrations of the various proteins, we transfected cells with cDNA encoding GFP chimeras of vinculin, paxillin or tensin, and 48 hours later (after the GFP-tagged molecules were fully incorporated into adhesion sites) the cells were fixed and counter-labeled with antibodies to vinculin, paxillin or tensin, and then Cy3-labeled secondary antibodies. FRI of such double labeling experiments (Fig. 4) indicated that, for the three proteins tested, FRI values were highly uniform, confirming that the immunofluorescence labeling intensity accurately represents the local concentration of the respective proteins.

To study the level of molecular heterogeneity of matrix adhesions, REF52 were fixed 48 hours after plating, and double-labeled for different pairs of plaque proteins, using a mouse/rabbit combination of primary antibodies and the corresponding FITC- and Cy3-labeled secondary antibodies. The pairs of primary antibodies used here included: vinculin-paxillin, tensin-vinculin, phosphotyrosine-paxillin, tensin-phosphotyrosine, and phosphotyrosine-vinculin. All of these components were previously known to localize to FC (Geiger et al., 1995; Yamada and Geiger, 1997). As a control for the reliability of the double-labeling approach, we calculated the vinculin-vinculin FRI following double labeling with rabbit and mouse antibodies.

Examination of this control sample (vinculin-vinculin, Fig. 5, left column) revealed uniform ratio values both between and within FC. These nearly identical values of fluorescence ratios were found irrespective of local variations in vinculin labeling intensity. Minor differences apparently do not correspond to definitive FC, and reflect low levels of diffuse background labeling. Vinculin-paxillin FRI (Fig. 5, middle column) revealed largely comparable distributions of the two proteins in FC. Some sites, displaying vinculin-rich structures with essentially no paxillin, corresponded to cell-cell adhesions, which were previously shown to contain vinculin but no paxillin (Turner et al., 1990). Examination of the tensin-vinculin pairs revealed very different distribution patterns (Fig. 5, right column). Three distinct types of adhesions were discerned: centrally-located adhesions which were tensin-rich, elongated FC at the cell periphery that were vinculin-rich, and smaller FC that displayed comparable fluorescence intensities of both proteins. These three populations were not randomly interspersed, but rather clustered in different regions of the cell (Fig. 5). Examination at higher power revealed highly non-uniform distributions of the two proteins, with distinct tensin-rich or vinculin-rich domains within individual adhesions (Fig. 6). High magnification images of vinculin-paxillin FRI (Fig. 6) were largely uniform both within and between FC.

In view of the involvement of tyrosine phosphorylation in FC assembly (Ayalon and Geiger, 1997; Clark and Brugge, 1995; Gilmore and Burridge, 1996; Miyamoto et al., 1995; Volberg et al., 1992; Yamada and Geiger, 1997; Yamada and Miyamoto, 1995), we have compared the distribution of phosphotyrosine to that of paxillin, tensin or vinculin. Phosphotyrosine-paxillin FRI pointed to a largely similar labeling of FC with uniform FRI (Fig. 7). Again, occasional phosphotyrosine-labeled paxillin-negative structures at the cell periphery corresponded to cell-cell adherens junctions. Vinculin-phosphotyrosine FRI showed less uniform patterns both between and within FC (Fig. 7). The most striking differences observed, however, were between phosphotyrosine and tensin (Fig. 7) manifested by low ratios along the tensin-rich fibrillar adhesions located mainly at the cell center, and ‘mosaic’ FRI images in more peripheral FC pointing to mutually exclusive distribution. Examination of individual adhesions revealed that in contrast to vinculin and paxillin, phosphotyrosine was non-uniformly distributed along FC and tended to be associated with the edges or the periphery of FC (Fig. 8). Phosphotyrosine levels in fibrillar adhesions were very low compared to tensin (Fig. 8). Moreover, even when tensin and phosphotyrosine were associated with the same structure, they usually did not overlap (Fig. 8).

To study the relationships between the morphology and composition of matrix adhesions, cells were labeled for tensin and phosphotyrosine, and the labeling intensities were correlated with the morphology of the adhesion sites. Phosphotyrosine was found to be primarily localized in round-to-ellipsoid (axial ratio <7) adhesions measuring 0.5-4 μm2 (Fig. 9A,B,C). In contrast, tensin was associated with three distinct types of structures: (i) small (0.5-3 μm ²) and rounded-to-ellipsoid (axial ratio <5) adhesions, (ii) elongated fibrillar structures (axial ratio >7), and (iii) large (>8 μm ²) and rounded-to-ellipsoid (axial ratio <7) adhesions (Fig. 9A,B,C). In the non-fibrillar structures, the intensities of tensin and phosphotyrosine were variable compared to the rather uniform prevalence of tensin in, and the absence of phosphotyrosine from, the fibrillar structures (Fig. 9A, compare to D). These results established the existence of different molecular forms of adhesion sites with distinct shapes.

To quantitatively determine the molecular relationships between different constituents of cell-matrix adhesions, we compared the labeling intensities of vinculin and phosphotyrosine in matrix adhesions to those of vinculin (control), paxillin and tensin (Fig. 10). As expected, the relation between the paired labeling for vinculin was linear, positive, and with a high coefficient of correlation (R=0.83±0.04). Paxillin labeling was also found to be strongly correlated with that of vinculin (R=0.73±0.08; Fig. 10). Tensin, on the other hand, showed no significant correlation with vinculin (R=0.05±0.21; Fig. 10).

Correlation of phosphotyrosine labeling intensity and that of paxillin, vinculin and tensin provided further insight into the molecular diversity of matrix adhesions. Both vinculin and paxillin were found to be positively correlated with phosphotyrosine intensities (R=0.42±0.17 and R=0.32±0.14, respectively), though the correlation coefficients were significantly lower than those obtained for the paxillin-vinculin pair. Tensin on the other hand was negatively correlated with the phosphotyrosine average intensity (R=-0.28±0.28; Fig. 10).

As previously shown, treatment of cells with the kinase inhibitor H-7 leads to the loss of cellular contractility and deterioration of stress fibers and FC (Volberg et al., 1994; Tian et al., 1998). To monitor the dynamics of the molecular changes occurring in adhesion sites due to the modulation of actomyosin contractility, we treated REF52 cells with 60 μM H-7 and examined the morphology and molecular organization of adhesion sites.

As shown in Fig. 11, after 5 minutes of treatment with H-7, FC still had normal morphology and contained vinculin, paxillin and tensin, yet their phosphotyrosine content was dramatically reduced. Upon longer treatment FC gradually disappeared, and within 30 minutes almost no FC with normal morphology were retained; only dot-like structures containing paxillin and phosphotyrosine could be detected at the cell periphery. Tensin, in these H-7-treated cells, was primarily found in small, scattered patches that were largely devoid of phosphotyrosine, paxillin and vinculin. The nature of these structures is still unclear. A similar effect was also noted following treatment of human foreskin fibroblasts with a specific myosin light chain kinase inhibitor, ML-7 (data not shown).

This effect of H-7 was rapidly reversible, and well-formed adhesion sites were reformed within 5 minutes after H-7 removal. The newly formed FC had apparently normal morphology, and their levels of paxillin, vinculin and phosphotyrosine were similar to those found in untreated cells (Fig. 12). Recovery of tensin in FC appeared to be somewhat slower.

The primary objective of the present study was to characterize the molecular and structural diversity, as well as the plasticity of cell-matrix adhesions. Studies for over two decades have revealed a multitude of so called ‘FC proteins,’ defined as such on the basis of their general distribution pattern. Yet, a standard and unequivocal definition of FC, their overall molecular compositions and potential diversity, have received relatively little attention. A major obstacle in approaching this issue in the past has been the non-quantitative nature of most microscopic studies of FC structure. Thus, matrix adhesions are commonly viewed as a molecularly homogenous entity, containing a full repertoire of FC proteins (Burridge and Fath, 1989; Geiger et al., 1995; Jockusch et al., 1995).

In this study, we challenge this view and establish, with the help of digital microscopy, that matrix adhesions are highly diverse structures, both molecularly and morphologically, that can be subdivided into 3 main categories: (i) ‘classical FC’ which are arrowhead shaped, and contain high levels of phosphotyrosine, paxillin and vinculin, and relatively low levels of tensin, (ii) ‘fibrillar adhesions’ which are elongated (typical axial ratio >7) and contain relatively high levels of tensin and low phosphotyrosine, and (iii) ‘mosaic FC’ with the general appearance of classical FC, yet with a highly non-uniform internal molecular structure. We further show that treatment with the kinase inhibitor H-7, which blocks actomyosin-driven contractility, leads to a rapid loss of phosphotyrosine from FC, followed by the disassembly of the structure. Re-assembly is swift (<5 minutes), suggesting that even large FC can be regenerated rapidly.

The key tool used here for studying matrix adhesion diversity was digital microscopy and multiple images processing, which enabled identification and molecular mapping of these sites. The novel elements in this analysis, described here for the first time, are the use of the water algorithm, the definition of criteria for an automatic recognition and segmentation of adhesion sites, and the introduction of the ratio-imaging approach. Our initial attempts to use uniform (size and/or intensity) thresholds in order to eliminate background labeling failed to resolve individual adhesions, and large clusters of FC were often identified as a single structure. The water algorithm enabled identification of local ‘valleys’ which separate adjacent adhesion sites, resolving such clusters into the individual adhesions. The setting of critical area thresholds was selected empirically, so that fine substructures within individual adhesions would not lead to excessive segmentation. This precaution was especially critical for fibrillar adhesions with granular labeling texture (such as the one shown in Fig. 2C). As indicated above, the segmentation process was carried out interactively, verifying that appropriate values of ⁠, and were selected, and allowing us to manually merge or separate adhesions. It is however noteworthy that such manual interventions were needed relatively rarely, and that the entire procedure enables a very broad and unbiased sampling of matrix adhesion sites, allowing for reliable statistical analyses.

A source of potential concern addressed in this study was the reliability of quantitation by immunofluorescence microscopy, namely, to what extent local fluorescence intensities accurately represent local concentrations of labeled proteins. This issue is relevant to interpretation of immunofluorescence data in general and is particularly critical in a quantitative study such as this report. To approach this issue, we compared the local intensities of fluorescence of GFP chimeras of vinculin, paxillin and tensin to the corresponding immunofluorescence images. Pixel-by-pixel comparisons of these images confirmed that the two were very closely correlated, indicating that, in these systems, levels of immunofluorescence intensity faithfully represent local protein concentrations and are not significantly affected by factors such as antigen exposure or epitope instability. It is nevertheless clear that these values provide information on the relative levels of the proteins in question and do not provide absolute protein concentrations.

A novel procedure described here is FRI of cytoskeletal and plaque proteins. As shown above, this approach was extremely useful in revealing the diversity in the molecular organization and sub-cellular distribution of matrix adhesions. It is, however, important to highlight here both the potential and limitation of this approach. The unique feature of FRI, which is not attainable by simple superposition, is that it is highly sensitive to local differences in relative fluorescence staining intensities of two proteins, irrespective of their absolute intensities, thereby providing valuable information for even faintly labeled structures. The calculations are based on 12-bit digital data, which are highly accurate even for faint structures. It should be emphasized that the calculated ratios are based on normalized fluorescence intensities (compensating for differences in overall labeling intensities with FITC-labeled and Cy3-labeled antibodies) and, as indicated above, they do not reflect the absolute concentrations of each of the molecules examined.

However, the data presented here on the differential distribution of various components of the submembrane plaque of matrix adhesions (vinculin, paxillin, tensin and phosphotyrosine) clearly document the diversity in the molecular organization of adhesion sites, as well as the presence of distinct molecular subdomains within these structures. We have distinguished here between ‘classical’ FC, ‘fibrillar’ adhesions, and ‘mosaic’ adhesions, based on the labeling for the four FC antigens listed above. Yet, it is entirely possible that systematic examination of additional plaque components (i.e. talin, focal adhesion kinase, VASP, α-actinin, actin, etc.; Geiger et al., 1995) will provide more comprehensive information on the detailed molecular organization of each type of adhesive structure. Such analysis is currently underway.

Another parameter examined in this study is the temporal sequence of molecular interaction involved in the formation and modulation of matrix adhesions. Based on previous data, we have used the kinase inhibitor H-7 for the suppression of FC formation, and examined the disruption and reformation of adhesive structures upon its addition and removal. It has been previously shown that H-7 strongly suppresses actomyosin-driven contractility in cultured cells (Volberg et al., 1994) and that this effect leads to deterioration of stress fibers and FC (Bershadsky et al., 1996; Volberg et al., 1994; Tian et al., 1998). FRI analysis of cells at different time points after addition of H-7 indicated that the earliest detectable response to H-7 treatment was the loss of phosphotyrosine from FC. It is not clear yet whether this reduction in local phosphotyrosine levels is due to dephosphorylation in situ or to loss of phosphorylated proteins from the adhesion site. This effect was observed as early as 5 minutes after the addition of H-7, though at that stage proteins such as vinculin and paxillin were still retained in FC. Only following longer incubations a loss of these proteins from FC was observed along with deterioration of FC structure. We therefore propose that local tension applied by the actomyosin system may trigger local tyrosine phosphorylation, which, in turn, is important for the assembly of the submembrane plaque.

In a separate study (B. Z. Katz et al., unpublished), we have examined the integrin specificity of the various matrix adhesions in human foreskin fibroblasts. We found that classical FC contain mainly α_vβ₃, are devoid of fibronectin and are tyrosine-phosphorylated. The fibrillar adhesions, on the other hand, are α₅β₁- and fibronectin-rich and are poorly labeled for phosphotyrosine. It was further shown that local tyrosine phosphorylation heavily depends on matrix deformability, whereas tensin distribution closely correlates with that of α₅β₁.

The results presented here on the molecular organization of matrix adhesion diversity, and the possibilities offered by the digital microscopic system described here, point to a remarkable molecular heterogeneity of these sites and offer a new classification of matrix adhesion subtypes. Further examination of matrix adhesion diversity after exposure of cells to a variety of additional experimental conditions such as differences in matrix molecules, growth factor stimulation, mechanical perturbations, microtubule reorganization and cell shape, may shed further light on the physiological assembly and reorganization of cell adhesion sites.

We thank Ken Nakata, Shingo Miyamoto and Kazue Matsumoto (NIDCR, NIH) for the human paxillin cDNA. We express our gratitude to Prof. Adi Shamir (Department of Applied Mathematics, the Weizmann Institute, Israel) for illuminating discussions. This study was supported by the Israel Science Foundation and the Minerva Foundation. BG holds the Erwin Neter Chair in Cell and Tumor Biology. ZK holds the Israel Pollak Chair of Biophysics