Journal of Microscopy, Vol. 232, Pt 1 2008, pp. 73–81 Received 19 July 2006; accepted 5 March 2008
Quantifying cell–matrix adhesion dynamics in living cells using interference reflection microscopy M . R . H O L T ∗ , Y . C A L L E ∗ , D . H . S U T T O N †, D . R . C R I T C H L E Y †, G.E. JONES∗ & G.A. DUNN∗ ∗ King’s College London, The Randall Division of Cellular and Molecular Biophysics, New Hunt’s
House, London, SE1 1UL, United Kingdom †Department of Biochemistry, Adrian Building, University of Leicester, University Road, Leicester, LE1 7RH, United Kingdom
Key words. Focal adhesion, image processing, interference reflection microscopy, podosome, vinculin. Summary Focal adhesions and podosomes are integrin-mediated cellsubstratum contacts that can be visualized using interference reflection microscopy (IRM). Here, we have developed automated image-processing procedures to quantify adhesion turnover from IRM images of live cells. Using time sequences of images, we produce adhesion maps that reveal the spatial changes of adhesions and contain additional information on the time sequence of these changes. Such maps were used to characterize focal adhesion dynamics in mouse embryo fibroblasts lacking one or both alleles of the vinculin gene. Loss of vinculin expression resulted in increased assembly, disassembly and/or in increased translocation of focal adhesions, suggesting that vinculin is important for stabilizing focal adhesions. This method is also useful for studying the rapid dynamics of podosomes as observed in primary mouse dendritic cells.
Introduction Many cell types form adhesions with the extracellular matrix that are often known as focal adhesions in the case of cells cultured on a plane substratum. These anchorage points provide a mechanical linkage between the extracellular matrix and the cytoplasmic microfilament system, which is composed of polymerized actin and cross-linked with myosin. Numerous signals are transduced via such linkages into the cell to direct many aspects of cell behaviour, most notably cell motility, growth/survival and differentiation. In cell motility, the force generated by the actomyosin microfilament system is transmitted through focal adhesions to the extracellular
Correspondence to: M.R. Holt. Tel: 020 7848 6438; fax: 020 7848 6435; e-mail:
[email protected] C 2008 The Authors C 2008 The Royal Microscopical Society Journal compilation
matrix, thus creating traction and enabling cell movement (for a review, see Wehrle-Haller & Imhof, 2003). When cells are observed using interference reflection microscopy (IRM), focal adhesions manifest as dark, often elongated, patches on the base of the cell and are referred to also as focal contacts (Abercrombie & Dunn, 1975; Izzard & Lochner, 1976). IRM requires illumination of the specimen with a beam of monochromatic light passing through the objective (Curtis, 1964). It is advantageous if the illuminating beam is linearly polarized so that unwanted internal reflections can be removed by an analyzer in the imaging path. The special Zeiss Antiflex immersion objective (Carl Zeiss, Oberkochen, Germany) incorporates a quarter-wave retardation plate that circularly polarizes the emergent light so that reflections from the specimen become plane polarized at right angles to the original direction and are not eliminated by the analyzer. When not using the Antiflex system, unwanted reflections will be reduced when the illuminated field is kept as small as possible. With an oil-immersion objective, the first major reflection is from the cover glass–medium interface and the reflected light has a half-wavelength phase shift because the medium has a lower refractive index than glass. Reflections from the medium–cell interface do not have this phase shift and so, when this light combines with the first reflection, the total phase difference is determined by the number of waves in the path travelled, from cover glass to cell and back, plus the extra half wavelength. Green light has a wavelength of 546 nm in air, which is about 400 nm in tissue culture medium, and so light reflected from regions where the ventral cell surface is 100 nm from the substratum is retarded by 2 × 100 nm plus the extra half wavelength (200 nm) giving a total of one whole wavelength, which brings it back into phase with the cover glass– medium reflection. This gives a bright combined reflection because the beams interfere constructively. By contrast, focal adhesions appear dark because they are only about
74
M. R. HOLT ET AL.
10 nm from the substratum, so the path-length component of the phase difference is small and the two reflected beams interfere destructively. Further reflections can complicate the interpretation of the IRM image. Fortunately, reflections from the cytoplasm–nucleus interface tend to be weaker because the refractive index difference is not so great. Reflections from the dorsal cell surface also contribute little to image contrast because the interference effect quickly becomes attenuated at high-aperture illumination with increasing separation of the reflecting surfaces. Focal adhesions are mediated by heterodimeric receptors called integrins (Hynes, 1992). Specific integrin receptors interact with extracellular matrix components such as vitronectin, fibronectin and laminin (Hynes, 1992). Immunofluorescence detection of specific integrins indicates that these receptors occupy the dark patches visualized by IRM (Singer et al., 1989). A large number of intracellular proteins are found in these IRM-dark, integrin-rich regions (Bloch & Geiger, 1980; David-Pfeuty & Singer, 1980). These include structural components that couple integrins to cytoskeletal actin, for example vinculin, talin and alphaactinin, as well as regulatory components, for example focal adhesion kinase, phosphatidylinositol 4 -phosphate 5 -kinase, src family kinases and certain phosphatases. The formation of focal adhesions is regulated by RhoA GTPase (Ridley et al., 1992). Podosomes are highly transient adhesion structures and localize behind the leading edge of certain monocytic cells, for example, macrophages, dendritic cells and osteoclasts, as well as v-src-transformed fibroblasts (Linder & Aepfelbacher, 2003; Brunton et al., 2004). Podosomes occur in vivo and are involved in matrix remodelling by migrating cells. In osteoclasts, they form circular clusters that fuse to define a sealed compartment where enzymes are released for bone matrix dissolution. Podosomes share some molecular components with focal adhesions, but differ in their organization. They consist of an F-actin core surrounded by vinculin, talin, alpha-actinin, fimbrin, gelsolin, vimentin and other adapter molecules, which are linked to specific integrins (Linder & Kopp, 2005). Recent work has demonstrated that the assembly of actin filaments into podosomes at the leading edge of dendritic cells requires the coordinated and sequential action of Cdc42, Rac and Rho (Burns et al., 2001). Previous attempts to quantitate focal adhesion dynamics have been laborious and have required tracking of individual structures (Anderson & Cross, 2000; Chandrasekar et al., 2005) or scoring adhesion formation and/or breakdown (Webb et al., 2004; Chandrasekar et al., 2005). Only two studies have attempted to quantify podosome dynamics (Kanehisa et al., 1990; Evans et al., 2003). Kanehisa and co-workers used phase contrast to achieve this although podosomes in monocytic cells cannot be detected in this way. The work of Evans and colleagues used cellular expression of fluorescently tagged proteins but this is not easily applied to
monocytic cells given the difficulty of transfecting primary cell types. Another consideration in choosing the noninvasive IRM technique for this study is that heterologous over-expression of GFP-tagged proteins could lead to expression artefacts and may alter the activity of the labelled protein. Here, we present an automated method for the extraction of focal adhesions, which can also be applied to podosomes, from time-lapse IRM sequences using image-processing algorithms. This method normalizes image background, thresholds the image and detects focal adhesions based on size and shape of objects present in the thresholded image before determining the extent of coincidence of focal adhesions at different time points. We have used this method to analyze focal adhesion dynamics in primary mouse embryo fibroblasts (MEFs) expressing or lacking vinculin as well as podosome dynamics in primary mouse dendritic cells. Materials and methods Cell culture MEFs lacking one (vin+/− ) or both (vin−/– ) alleles of the vinculin gene were derived from mouse embryos of the appropriate genotype (Xu et al., 1998). Cells were immortalized using SV40 T antigen. MEFs were cultured in DMEM (Gibco BRL, Invitrogen Ltd., Paisley, UK) containing the following supplements: 10% foetal bovine serum, 1 mM sodium pyruvate, 1 × non-essential amino acids, 0.001% 2-mercaptoethanol, penicillin–streptomycin (90% air : 10% carbon dioxide). Cells were seeded onto No. 1 1 / 2 22 × 22 mm glass cover slips, 24–48 h before observation by IRM. Dendritic cells were derived from spleen based on a protocol previously described (West et al., 1999). Spleens from 6- to 8-week-old SV129 mice were homogenized through a cell strainer to obtain a cell suspension. Cells were washed three times with RPMI containing 1% FBS, and then resuspended in RPMI supplemented with 10% FBS, 1 mM pyruvate, 1 × non-essential amino acids, 2 mM glutamine, 100 μg mL–1 kanamycin, 50 μM 2-mercaptoethanol, 20 ng mL–1 recombinant mouse GM-CSF and 1 ng mL–1 recombinant human TGF-β at a density of 2 × 106 cells mL–1 at 37◦ C in a 5% CO 2 atmosphere. After 14 days ex vivo, 75–80% of the cells in culture were dendritic cells as determined by the expression of CD11c and DEC205 by FACS analysis. Dendritic cells were plated on poly-L-lysine-coated cover slips in culture medium and allowed to adhere and spread for 3 h before IRM analysis. Interference reflection microscopy (IRM) Interference reflection images were collected using a Zeiss Standard 18 microscope with IV-FL incident light fluorescence attachment and a Zeiss 63× Neofluar Antiflex oil-immersion
C 2008 The Authors C 2008 The Royal Microscopical Society, Journal of Microscopy, 232, 73–81 Journal compilation
QUANTIFYING CELLULAR ADHESION DYNAMICS
75
objective, NA 1.25. Exciter and barrier filters were removed from the LP420 reflector and replaced with a linear polarizer and a narrow band-pass filter (to isolate the 546 nm line of the mercury arc source) in the illuminating path complete with a computer-controlled shutter and a crossed analyzer in the imaging path. Cover slips with attached cells were placed onto glass chambers containing normal growth medium, sealed with wax and observed in a 38◦ C warm room. Images were collected digitally every 10 (podosomes) or 60 s (fibroblasts) using a Pulnix CCD camera (Jai Ltd., Harefield, UK) coupled to a Matrox Meteor PCI frame grabber and in-house acquisition software (Matrox Video and Imaging Technology Europe Ltd., Harefield, UK) and saved as stacked TIFF files. Image processing using MathematicaTM 5.2 Single-frame TIFF images were extracted from TIFF stacks and processed using MathematicaTM 5.2 (Wolfram Research, Champaign, IL) as described in the Results section. Analysis was performed on a Dell Precision PWS380 with an Intel Pentium 4 (3.0 GHz, 1 GB RAM) running Microsoft Windows XP Professional ×64 edition (SP2). Results Eight-bit greyscale images obtained by IRM of live mammalian cells were low in contrast (see Figs 1(A) and (B)). This was mostly due to the short exposure times (20 ms) used to minimize phototoxic damage to the live cells. In addition, the background was uneven due to uneven illumination across the field of view. Since thresholding of such images were unable to capture all of the required details (see Figs 1(G) and (H)), we have devised some image-processing steps to even out the background and increase contrast between the focal adhesions and the rest of the image. Image stacks of time-lapse sequences in TIFF format were imported as shown in expression 1 (Exp 1) and converted to MathematicaTM 5.2 format (Exp 2): Exp1: tiffStack = Import[ X: \ \ directory \ \ folder \ \ filename.tif ]; Exp 2: rasterStack = Table[tiffStack[[i,1,1]], {i,Length[tiffStack]}]/255.;
The first processing step was to increase contrast in collected TIFF stacks without compromising image detail. This was achieved in MathematicaTM 5.2 by applying a custom autolevel function (Exp 3–5): Exp 3: imageHistogram[imageRaster ,frame ]: = BinCounts[Flatten[imageRaster[[frame]]∗255], {0,255,1}]; Exp 4: autoLevel[imageRaster List]: = Module[{imageProfile,gLevels,newLevels}, imageProfile = imageHistogram[imageRaster,All]; gLevels = Flatten[Position[imageProfile, ?(#>0.05∗ Mean[imageProfile]&)]]/255.;
Fig. 1. Image processing of IRM micrographs of a fibroblast (A) and a dendritic cell (B) were extracted from digitally recorded IRM images (A, B) using a process of high-pass filtering (C, D), followed by autolevelling (E, F). The effect is to increase image contrast. (G, H) The image from part A was thresholded with two separate values (120 and 90 on a 0–255 scale, respectively). Note that each value is only capable of highlighting some of the adhesions, but not all. The scale bar represents 10 μm and is applicable to all images in this figure.
newLevels = (imageRaster-gLevels[[1]]) /(gLevels[[-1]]-gLevels[[1]])]; Exp 5: autolevelRaster = autoLevel[#]& /@ rasterStack;
Here, a function (Exp 3) is defined that determines the number of pixels per intensity level using the BinCounts command from the MathematicaTM Statistics‘add-on package. An intensity offset was then calculated and grey level with intensity values that fell below this offset were discarded (Exp 4). The offset was calculated as 5% of the mean number of pixels per grey level from the pixel histogram. The highest and lowest grey levels that fell above this offset were then
C 2008 The Authors C 2008 The Royal Microscopical Society, Journal of Microscopy, 232, 73–81 Journal compilation
76
M. R. HOLT ET AL.
used to linearly stretch intervening grey levels so that they covered the entire dynamic range. The module was applied to the image stack as shown in Exp 5. Next, images were filtered, convolving with a 41 × 41 high-pass kernel: Exp 6: highpassFilter[size ]: = Module[{highpassKernel}, middle = (size-1)/2 + 1; template = Table[-1,{size},{size}]/(size)ˆ 2; highpassKernel = ReplacePart[template,1Sqrt[template[[1,1]]ˆ 2;],{middle,middle}]]; Exp 7: highpassRaster = ListConvolve[highpassFilter[41],#]+0.5)& /@ autolevelRaster;
The module in Exp 6 constructs the filter kernel that is applied to the image stack as shown by Exp 7. The autolevel function (Exp 3–4) was applied once more to the filtered images as shown in Exp 8: Exp 8: autolevelHPFRaster = autoLevel[#]& /@ highpassRaster;
The high-pass filter reduced differences between pixels in locally repetitive regions of an image and increased differences between non-repetitive areas (Figs 1(C) and (D)). The filter does this by converting pixel intensities of equal or similar values (or low spatial frequency) to zero. Pixels with intensities that fall in regions of high-intensity variability (high spatial frequency) are converted to positive and negative intensity values. An offset (of 0.5) is then added to the image such that lowfrequency pixels appear mid-grey, whereas high-frequency pixels are darker or lighter than mid-grey. Autolevelling stretched out the pixel grey level histogram linearly over the whole dynamic range to accentuate the differences between high-frequency light and dark pixels leaving the lowfrequency pixels mid-grey. The overall effect was to increase contrast and sharpness in the image as well as to flatten heterogeneities in the background due to uneven illumination (Figs 1(E) and (F)). Since the high-pass filtering introduced speckled noise (very high spatial frequency), images were low-pass filtered (or smoothed) using the module in Exp 9 to construct a 5 × 5 convolution kernel to smooth the images using Exp 10: Exp 9: lowpassFilter[size ]: = Module[{lowpassKernel}, lowpassKernel = Table[1,{size},{size}]/(size)ˆ2;]; Exp 10: lowpassRaster = ListConvolve[lowpassFilter[5],#]& /@ autolevelHPFRaster;
The effect of using a high-pass filter followed by a low-pass filter, as shown here, is equivalent to applying a band-pass filter to the image. Indeed, it is possible to combine the high- and low-pass filter steps into one step by applying the low-pass filter kernel to the high-pass filter kernel to create a band-pass filter
kernel using, for instance, 2D Fourier filters. The image can then be convolved with this derived band-pass filter kernel and is computationally faster than sequentially applying the two separate low- and high-pass filters. Because of the large size of the high-pass filter kernel used here, the band-pass filter is quite broad-ranged, that is letting through most spatial frequencies, but removing the very highest and lowest spatial frequencies. The range of the band-pass filter created in this fashion can be altered by changing the sizes of the high- and low-pass filter kernels. Comparison of Figs 1(A) and 1(E) reveals the striking effect of these steps on image contrast and clarity such that the presence of focal adhesions is more obvious. Similarly, the effect on the clarity of podosome-containing images can be seen in Figs 1(B) (original) and 1(F) (enhanced). The image was then inverted and thresholded using the function in Exp 11 and applied to images in Exp 12 to produce a binary image (pure black and white pixels encoded as intensity values 0 and 1, respectively) that represented the very darkest regions of the contrast-enhanced IRM images as white pixels on a black background (Figs 2(A) and (G)): Exp 11: imageThreshold[imageRaster_List, thresholdLevel Integer]: = Ceiling[(thresholdLevel/255.)-imageRaster]; Exp 12: thresholdedRaster = imageThreshold[#,90]& /@ lowpassRaster;
The normalization that resulted from the high-pass filter and autolevelling meant that all images from the same image stack, as well as image stacks for all observed cells (irrespective of cell treatments, cell type or date of image collection) could be treated with the same empirically derived threshold. Information about focal adhesion dynamics could now be extracted from processed time-separated images. Areas of the images that did not represent focal adhesions were removed manually using the eraser function in Adobe Photoshop 7.0 and/or automatically using our own segmentation algorithms written in MathematicaTM . The segmentation algorithms allowed removal of objects based on various size and shape parameters (as discussed further on in this Results section). These objects represented the majority of artefacts present and included the image of the field diaphragm and small spots within the cell that appeared to be vesicles or small particles. Following removal of these artefacts, each thresholded image (frames 1, 21, 41 and 61 from a 61-frame stack) was then inverted and divided by the number of time-lapse images used for the analysis, in this case four (Exp 13; Fig. 2(B)). Exp 13: overlaygreyscaleFrames = Plus@@ (1 - thresholdedRaster[[{1,21,41,61}]])/4;
White pixels corresponding to adhesions, therefore, became dark grey, that is 0.25 on a scale of 0–1 in MathematicaTM . The four images were then added to each other to produce one
C 2008 The Authors C 2008 The Royal Microscopical Society, Journal of Microscopy, 232, 73–81 Journal compilation
QUANTIFYING CELLULAR ADHESION DYNAMICS
Fig. 2. The IRM image of a fibroblast was thresholded and further processed by removal of areas of the images that did not represent focal adhesions, such as those resulting from the microscope field diaphragm, the cell nucleus, contamination etc., and then inverted so that adhesions were white on a black background (A). The same processing was applied to the dendritic cell (G). The fibroblast-derived image was then divided by four (B) and repeated for three other time-separated images in the same time-lapse series. The four images were then overlaid (C) and then inverted to give an adhesion turnover map (D). Further maps can be seen in E and F. Fibroblasts either contained (D, E) or lacked (F) vinculin expression. Note the presence of more black pixels in D and E, and more light grey pixels in F. This indicates greater focal adhesion turnover in fibroblasts lacking vinculin expression (F). The same image processing produced an adhesion turnover map for the dendritic cell (H). The scale bar represents 10 μm and is applicable to all images in this figure.
composite image consisting of a black background and four different grey levels representing focal adhesions or clusters of podosomes (Eq. 13; Fig. 2(C)). In MathematicaTM , these grey levels were 0.25, 0.5, 0.75 and 1.0, respectively. Finally, the image was re-inverted to give a white background with grey levels in MathematicaTM given as 0 (black), 0.25 (dark grey),
77
0.5 (medium grey) and 0.75 (light grey) (Exp 13; Figs 2(D)–(F), and (H)). Black, dark grey, medium grey, light grey and white pixels indicated that the region was adherent in four, three, two, one or none of the four frames, respectively. Black pixels therefore reflect regions of low adhesion dynamics, whereas light grey pixels represent regions of higher adhesion dynamics. The representative focal adhesion maps shown were derived from four images taken 20 min apart for vin+/− MEFs (Figs 2(D) and (E)), and vin−/– MEFs (Fig. 2(F)). The podosome map shown was derived from four images taken 10 s apart (Fig. 2(H)). The number of pixels for each of the four non-white levels were then determined using MathematicaTM . From these values, we defined a turnover index, where the number of pixels present in three or four of the four collected images (low dynamic adhesions) was expressed as a percentage of the total number of non-white pixels. For instance, vin+/− MEFs and vin−/− MEFs had significantly different indices (Fig. 3(B)) of 28.9 +/− 3.6% (SEM, n = 16) and 16.3 +/− 3.2% (SEM, n = 16), respectively (Student’s t test, P = 0.006), indicating that the contacts in the vin−/− MEFs were 1.77 times less stable than those of the vin+/− MEFs, that is focal adhesions in cells lacking vinculin are more unstable than those in vinculinexpressing cells. A limitation of the greyscale maps was the lack of detailed temporal information. Without reference to original timelapse sequences, it was difficult to determine if high-turnover contacts resulted from breakdown, de novo formation, growth or translocation (slippage) of existing contacts. To circumvent this limitation, we developed a colour (RGB) version of the above procedures. Here, the images were processed so that pixels representing adhesive regions were coloured mid-grey (Exp 14), red, green or blue (Exp 15) according to whether they were present in the first, second, third or fourth frame, respectively, all other pixels being coloured black, that is where adhesions were absent. The colour-coded images were then overlaid (Exp 14–16): Exp 14: firstframeGreyscale = (1-thresholdedRaster[[1]])/.x −>{0.5,0.5,0.5}; Exp 15: overlaythreeframestoRGB = 1−(Thread[#]& /@ Transpose[thresholdedRaster[[{21,41,61}]]]); Exp 16: overlayRGBwithgreyscaleFrame = (overlaythreeframestoRGB + firstframeGreyscale) /.x /;x>1−>0.8;
The absence, or presence, of overlap resulted in specific colour combinations that indicated the origin and fate of particular contacts. Where contacts were stable, the overlapping frames resulted in a preponderance of white or light grey as a result of merging of the base colours. Where the contacts were highly dynamic, there was little overlap and the four base colours were readily discernible. Translocation of adhesions could be recognized as contiguous regions displaying a characteristic sequence of colours. Adhesions that
C 2008 The Authors C 2008 The Royal Microscopical Society, Journal of Microscopy, 232, 73–81 Journal compilation
78
M. R. HOLT ET AL.
Fig. 3. The adhesion turnover maps produced in Figs 2(D), 2(E) and 2(H) were colour-coded based on the frame they were present in. Pixels present in frame 1 were mid-grey, frame 2 pixels were red, frame 3 pixels were green and pixels in the final fourth frame were blue. Overlaying these colour-coded frames resulted in a temporally encoded turnover map for fibroblasts expressing (A), or lacking (B) vinculin as well as for dendritic cells (C). D, E and F are close ups of the maps in A, B and C. Note in the top right-hand corner of E the sequence of grey, red, green and blue adhesions running towards the centre of the cell. This represents retrograde translocation of the same cluster of focal adhesions. The scale bar represents 10 μm and is applicable to the top three images. The lower three images have been magnified 3-fold.
were grey and/or red, but with no associated green and/or blue were adhesions that had broken down early during the time course, whereas green and/or blue adhesions that were not associated with grey and/or red adhesions would have formed later in the time course. Isolated regions of base colour represented highly transient adhesions. Inspection of Fig. 3(B) towards the top left and right-hand sides reveals the presence of blue pixels that are associated with little or no other colours. This suggests that these adhesions appeared in the last 20 min of filming (since blue is the pseudocolour code for the final frame). Note in the top righthand corner of Fig. 3(E) the sequence of adhesions running from grey (original position) through red and green to blue (final position). This sequence indicates slippage of the focal adhesions along the substratum. Examination of Fig. 3(F) shows the cluster of podosomes moving en masse towards the left-hand side of the field of view. The image represents total translocation over 30 s showing that the podosomes are highly dynamic adhesion structures, as expected. Metastatic cells often show increased focal adhesion turnover when cultured in vitro and some have been shown to form podosome-like structures. The machinery required for cell motility represents an attractive target for drugs to block these motile changes and prevent tumour cells from moving away from the primary tumour site. The method presented here could be useful in testing such drugs although not in high throughput screens due to the manual input required to remove non-focal adhesion artefacts from images.
Therefore, we developed an object-detection method to reject non-focal adhesion artefacts from thresholded, normalized IRM images. The first step involved modification of the thresholding step. It was found that the centre of each focal adhesion was darker than its periphery and the empirically determined threshold level presented earlier was set to capture both these regions. However, this meant that some non-focal-adhesion components were captured by this threshold. Since such objects did not generally contain a dark centre, a harsher threshold step was introduced to capture only the darker centres of focal adhesions. This allowed identification of pixels in the centre of focal adhesions. Next pixels that fell within a 10-pixel radius of these central pixels were extracted to capture the lighter regions of the focal adhesion and a few non-focal adhesion pixels around the periphery of the focal adhesion, but nothing else. Next, the original threshold step was applied to the image to produce a binary image lacking the majority of non-focal adhesion artefacts. However, some objects still remained such as particles on the cover slip outside of the cell and an artefact created by the image processing associated with the edge of the field aperture. These could be distinguished from focal adhesions by their size and shape. The particles were small and round, whereas the aperture artefact was much larger than adhesions. Thus, we set out to remove these based on a set of defined size and shape parameters. To do this, each binary image was segmented and individual objects within the binary image catalogued as separate groups of touching black
C 2008 The Authors C 2008 The Royal Microscopical Society, Journal of Microscopy, 232, 73–81 Journal compilation
QUANTIFYING CELLULAR ADHESION DYNAMICS
79
pixels. The code module used is shown below: Exp 17: detect[image ]: = Module[ {neighbours, binary, position, labels, img, objects, pos2, lengths, labelled}, neighbours = Flatten[Table[ {xx, yy},{xx, −1,1},{yy, −1,1}], 1]; binary = image; position = Position[binary, 1]; Table[binary[[position[[i,1]], position[[i,1]] = i,{i, Length[position]}]; labels[{x ,y }]: = binary[[x, y]] = Min[DeleteCases[Extract[binary, {x, y} + # & /@ neighbours, ?(# < 1 &)]]); img[ ]: = (labels[#] & /@ position; binary); FixedPoint[img, binary]; objects = Extract[binary, pos2 = Position[binary, ?(# > 0 &)]]; lengths = Flatten[Join[{0, Length /@ Split[Sort[objects]]}]]; labelled = DeleteCases[Split[Sort[Thread[ {objects, pos2}]], First[#1] = = = First[#2] &], ?(Length[#]
