EX P ER I ME NTAL C E LL RE S E ARCH

330 (2015) 382 –397

Available online at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/yexcr

Research Article

Quantitative imaging of focal adhesion dynamics and their regulation by HGF and Rap1 signaling Emma Spanjaarda, Ihor Smalc, Nicos Angelopoulosb, Ingrid Verlaana, Alexandre Matova,1, Erik Meijeringc, Lodewyk Wesselsb, Hans Bosa, Johan de Rooija,n a

Center for Molecular Medicine, University Medical Center Utrecht, Utrecht, The Netherlands Division of Molecular Carcinogenesis, Netherlands Cancer Institute, Amsterdam, The Netherlands c Departments of Medical Informatics and Radiology, Erasmus MC – University Medical Center Rotterdam, Rotterdam, The Netherlands b

article information

abstract

Article Chronology:

Cell migration is crucial in development, tissue repair and immunity and frequently aberrant in

Received 26 June 2014

pathological processes including tumor metastasis. Focal adhesions (FAs) are integrin-based adhesion

Received in revised form

complexes that form the link between the cytoskeleton and the extracellular matrix and are thought to

19 September 2014

orchestrate cell migration. Understanding the regulation of FAs by (oncogenic) signaling pathways may

Accepted 12 October 2014

identify strategies to target pathological cell migration. Here we describe the development of a robust

Available online 31 October 2014

FA tracker that enables the automatic, multi-parametric analysis of FA dynamics, morphology and

Keywords: Focal adhesion Computer Vision HGF Rap1 Cell migration Bayesian network inference

composition from time-lapse image series generated by total internal reflection fluorescence (TIRF) microscopy. In control prostate carcinoma cells, this software recapitulates previous findings that relate morphological characteristics of FAs to their lifetime and their cellular location. We then investigated how FAs are altered when cell migration is induced by the metastasis-promoting hormone HGF and subsequently inhibited by activation of the small GTPase Rap1. We performed a detailed analysis of individual FA parameters, which identified FA size, sliding and intensity as primary targets of Rap1. HGF did not have strong effects on any of the FA parameters within the first hours of its addition. Subsequent Bayesian network inference (BNI), using all measured parameters as input, revealed little correlation between changes in cell migration and FA characteristics in this prostate carcinoma cell line. Instead BNI indicated a concerted coordination of cell size and FA parameters. Thus our results did not reveal a direct relation between the regulation of cell migration and the regulation of FA dynamics. & 2014 Elsevier Inc. All rights reserved.

Introduction

metastasis. 2D culture has been instrumental in the elucidation of the fundamental principles, proteins and regulatory mechanisms involved

Regulated cell migration is vital for embryonic development and

in cell migration. Integrin-based adhesion complexes, collectively

tissue morphogenesis and underlies pathological processes like tumor

called Focal adhesions (FAs), are the mechanical units that connect

Abbreviations: TIRF, total internal reflection fluorescence microscopy; FA, focal adhesion; ECM, extracellular matrix; BNI, Bayesian network inference n

Corresponding author. E-mail address: [email protected] (J. de Rooij). 1 Present address: Department of Cell and Tissue Biology, University of California, San Francisco, USA.

http://dx.doi.org/10.1016/j.yexcr.2014.10.012 0014-4827/& 2014 Elsevier Inc. All rights reserved.

EX P ER I ME NTAL C E LL RE S E ARCH

330 (2015) 3 82 – 397

383

the intracellular cytoskeleton to the extracellular matrix (ECM) and

[2] and at least 150 have been established as true FA proteins in

provide the traction necessary for productive cell migration. FAs are

detailed studies [3]. FAs have also been identified in 3D culture of

also signal transduction centers that relay mechanical information

tumor cells [4] and in vivo in vascular endothelium (unpublished

from the extracellular matrix (stiffness, density) to the intracellular

observations). The relevance of the investigation of FAs for cancer cell

machinery. FAs contain a number of proteins that regulate cytoske-

migration is further underscored by the fact that expression levels

letal organization and dynamics to alter local and global cell mec-

and/or signaling activity of FA proteins are often found deregulated in

hanics and a number of proteins that regulate signaling cascades to

metastatic cancer [5,6]. Furthermore, inhibitors of 2 master regulators

alter nuclear processes such as proliferation and differentiation [1].

of FA dynamics, FAK and Src are being developed as clinical inhibitors

Over 1000 putative FA inhabitants have been identified by proteomics

of cancer progression [7–9].

Fig. 1 – hDome detection. (A) The framework of the detection method. (B) The unprocessed image for hDome segmentation, for further explanation zooming in on the red square. (B1) Smoothed image (left) and corresponding 3D intensity plot. (B2) Reconstructed image (left), created with grayscale reconstruction and its corresponding 3D intensity plot (right). (B3) Zoom of the hDome image (left), resulting from subtracting the background image from the smoothed original and its corresponding 3D intensity plot (right). (B4) Image upon weight assignment (left) and its 3D intensity plot (right). (B5) Spot image generated by Monte Carlo simulation and clustering (left) and its corresponding 3D intensity plot (right). (C) Zoom-in of overlay of the markers detected with the hDome method for the watershed segmentation (left), and the regions defined by watershed segmentation (right). (C1) The adaptive threshold line indicating the intensity range and local threshold of the depicted FAs 1 and 2. (C2) Result of thresholding each region that was found in the watershed with the adaptive threshold (left) and the overlay in red on the original image (right). (C3) Overlay of detected FAs on the original image, with the zoom area in green. (D) Determination of protruding and retracting cell areas by subtracting the binary cell image at time (t1) from the image at time (tþ5) (top). In the result image, newly arisen areas have value 1 (white) and disappeared areas have value 1 (black) (bottom left). In this result image triangles are drawn from the centre (red dot) along the edges of the found regions (green lines) to enlarge the protruding and retracting regions towards the centre with a maximum width of 50 pixels (green area) (bottom right). (D1) Color code image of the FAs in the different areas: green in a protruding region, red when in a retracting region and otherwise blue.

384

E XP E R IM ENTA L CE L L R E S EA RC H

FAs come in a range of different sizes and although generally described as elliptic, their shape also varies largely upon close inspection. Differences in size, shape and dynamics have been associated with involvement in different aspects of the migratory process. Small FAs (then called Focal Complexes) form in the front of a migrating cell, where they stabilize the actin-polymerization driven protrusion (lamellipodium) by linking it to the extracellular matrix [10–12]. At the base of the lamellipodium, FAs can either turnover or mature into larger FA. These FAs are static structures that provide the anchor points on which traction force can be exerted to pull the cell body forward. As a cell migrates over its FAs they become more centrally located and eventually end up in the rear where they need to disassemble to allow retraction of the rear of the cell [13,14]. At this stage FAs are no longer stationary and display a sliding motion towards the center of the cell. The formation, turnover, maturation and senescence or death of FAs are subject to regulation by a number of signaling pathways. For instance, key to the decision to mature or turnover is the phosphorylation state of FAK and paxillin [15–17]. FA growth and elongation (maturation) is a tension sensitive process in which the molecular composition changes due to recruitment and activation of other FA proteins like vinculin and zyxin [15,18–21]. FA disassembly in the center and rear of the cell can be mediated by a diversity of mechanisms, including metalloprotease cutting of the integrin ECM connection or calpain-induced proteolysis of FA proteins, like talin, paxillin and FAK [22–24]. Furthermore a decrease of actomyosin-based tension, possibly through microtubule-dependent delivery of relaxation factors [25,26] is observed in cell-rear-located FAs which underscores their mechanosensitivity. Thus, at FAs mechanical signals and biochemical signals converge to regulate FA size, shape, dynamics and lifecycle (reviewed in [27]). Because FA characteristics vary with their role in cell migration and signaling factors that regulate cell migration-dependent processes also affect the characteristics of FAs, it is logical to speculate that the regulation of FA characteristics by such signaling factors is a causative step in their regulation of cell migration, but this has never been directly investigated to our knowledge. To better understand how FA characteristics and dynamics correlate with cell migration and if signaling factors indeed target FAs to regulate cell migration, we have developed a method that automatically detects and tracks individual FAs (in time lapse image series generated by TIRF microscopy of fluorescently tagged FA proteins) and relates those to their cellular region. We have induced migration by the metastasispromoting hormone HGF and inhibited migration by the subsequent activation of the small GTPase Rap1. Surprisingly, our analyses reveal that a very few FA characteristics change upon induction of cell migration by HGF, whereas Rap1 activation, which rapidly blocks cell migration, reduces FA intensity, FA size and FA velocity. Co-analyzing all measured FA parameters and overall cell behavior by means of Bayesian network inference (BNI) showed that cell size and FA behavior are highly connected, whereas cell migration speed and FA behavior are largely uncoupled. In conclusion, we have developed robust automatic FA detection and tracking software, revealed new specific effects of Rap1 on FA behavior, but failed to show correlations between the regulation of FA morphology and dynamics and the regulation of cell migration.

33 0 (2 015 ) 38 2 –3 97

Results Workflow and algorithms of the FA detection and tracking software The workflow of the detection and tracking software that we developed in MatLab is depicted in Fig. 1A. It consists of three main parts: detection of individual FAs, determination of the boundaries of each FA and tracking of each FA from frame to frame. The images that served as input for this FA-tracking algorithm were generated by TIRF microscopy using cells that expressed fluorescently tagged FA proteins, in this case mCherry– paxillin (Fig. 1B, Fig. S1-1 and S1-2and Supplementary movies S1 and S2). Supplementary material related to this article can be found online at http://dx.doi.org/10.1016/j.yexcr.2014.10.012. In the first part of the method we detect individual FAs and determine their localization in every frame of the timelapse, using an hDome based detector [28]. This method has been shown to have a high true positive detection rate even in images with relatively low signal-to-noise ratio [29]. The x and y coordinates of the centers of FAs are found in four consecutive steps of peak extraction, pixel separation, clustering and center of mass determination. First, to get rid of small pixel-to-pixel fluctuations in intensity in the image (noise), the image is smoothed with a Gaussian-shaped kernel with a user-defined deviation σ (Fig. 1B1). To determine the localization of larger peaks in pixel intensities (putative FA signals), regardless of the absolute intensities we apply gray scale reconstruction [30]. This method requires the original, smoothed image and the same image after subtraction of a user-defined value h. It creates a flattened image by dilation of local maxima (Fig. 1B2). A peak image, called hDome image, is then created by subtracting this flattened image from the original (Fig. 1B3). This implies that the hDome image contains peaks that all have a value of h or lower. These peaks thus originate from the concentrated signals from FAs but inevitably also contain some non-FA, peak-like background structures. As a means to discriminate between FA- and backgroundgenerated peaks, all pixels in the hDome image are given a weight value, calculated by their gray value to the power of a user-defined number p (Fig. 1B4). Those weights represent a probability distribution in the image space and we sample 5000 pixels according to the weights, adopting a form of Monte Carlo sampling. We found that sampling 5000 pixels reliably reconstructed all FAs while keeping computation time limited. To finally determine the position of FAs, clusters are generated for each pixel by measuring the number of pixels within a user-defined distance rad. If a cluster contains a user-defined minimum number of pixels, minNum, the pixel is considered to belong to an FA (Fig. 1B5). This separates densely clustered pixels originating from concentrated FA signals from sparsely clustered pixels originating from background structures. The central (x, y) coordinates of each cluster are found by determination of the central pixel. Thus the localization of putative FAs is distinguished from structures generated by background intensity fluctuations in these images and the central coordinates of these FA-clusters are used as input for the FA boundary detection in the next part of the method.

EX P ER I ME NTAL C E LL RE S E ARCH

For precise FA boundary detection, the original image is segmented by a local watershed into smaller regions containing single FAs, using as marker the central coordinates defined in the first part (Fig. 1C1, upper panels). In each region a relative intensity threshold is then applied to define the area of the focal adhesion (Fig. 1C2). The threshold depth is adaptive to the intensity range in the region (Fig. 1C1, lower panel), which is calculated as the difference between the average intensity of the 2% highest intensity pixels (this averages out outliers) and the average intensity of a userdefined percentage (lowpart) of the lowest intensity pixels within the region. The threshold depth (x) for each region is then defined by the line y¼  axþb, where y is the intensity range and a and b are user defined. Thus regions with a high intensity range (bright FAs) get a low threshold depth, whereas regions with a low intensity range (dim FAs) receive a high threshold depth (Fig. 1C1, lower panel). To eliminate remaining dim background structures, regions that have an intensity range below 1/65th of the maximal intensity at that bitdepth are excluded. However, detection of transient structures that are not FAs is not fully eliminated because some background fluctuations do reach the same intensity as dim FAs. Thus, the boundaries of the focal adhesions are defined (Fig. 1C2 and C3) and this is all the information needed for FA tracking and further analysis in the next part of the method. Tracking the focal adhesions is the third part of the analysis. FAs generally do not move very fast. This implies that the image of a given FA will have considerable overlap from frame to frame if the time resolution of the imaging is sufficiently high. This concept is used to track the detected FAs by linking them together in sequential frames if they occupy overlapping pixels. To minimize errors and maximize yield, several restrictions and corrections are applied: first to exclude FAs located outside the cell, the outline of the cell is determined with a threshold based on the bimodal histogram of the TIRF image and detected objects located outside are excluded from the linking algorithm. Second, in order to prevent tracks from jumping between closely neighboring FAs and to deal with splitting or merging FAs, if FAs overlap with multiple detected objects in the following frame, they are linked to the object with maximal overlap. Third, to track FAs as long as possible independently of short fluctuations in illumination or other brief

385

330 (2015) 3 82 – 397

interruptions of signal, in case an FA has no overlapping object in the following frame (tþ1), it is compared to the two subsequent frames (tþ2 and tþ3). If in this comparison an overlapping object is found, the track of this object is continued from the new frame onwards. This results in a track with small gaps at the time points that no FA was detected. These gaps are not repaired to avoid mistakes in the FA feature measurements later on. Fourth, if a track is shorter than three frames, it is excluded from further analysis, to eliminate short-lived structures due to background fluctuations that often occur in TIRF images. This resulted also in the exclusion from our further analyses of all bona fide FAs that lived less than 1 min, because the minimal frame-rate in our experiments is 30 s. FAs that live less than 1 min are for instance the rapid-turnover population in the leading edge described by Horwitz [11]. An additional step in our method consists of the definition of protruding, retracting and non-moving regions of the cell and designation of the FAs in each region. This provides the possibility to classify FA characteristics in different cellular regions. Differences in behavior and in morphology have been observed for FAs in the leading edge compared to the ones in the retracting area of the cell [31]. To define protruding and retracting regions, the binary image of the cell at a certain frame (t) is subtracted from the binary cell image at frame tþ4 (Fig. 1D, upper panels). At a frame rate of 30 s that we used, 5th frame is 2.5 min and this was found to capture stable protrusions, while neglecting rapid, transient protrusions and fluctuations in edge-detection due to imaging instability. In the resulting image, newly formed areas (protrusions) will have the value 1, whereas disappeared areas (retractions) will get the value  1 (Fig. 1D, lower panels). The protrusion and retraction areas are then enlarged towards the cell center such that the diameter of the area in the direction of the cell center is maximal 50 pixels (i.e. 6.5 μm which is roughly the size of the lamellum in these cells when they are migrating at their average speed of 1.5 μm/min) (Fig.1D). This is done for each fifth frame starting from frame 1. In the first and in-between frames the focal adhesions are labeled red when they are in a retraction region, green when they are in protrusive area and the rest (non-moving area) are labeled blue (Fig.1D1 and Supplementary movie S3). With this tool it is possible to separate populations of focal adhesions based on the

Table 1 – Features analyzed by our FA detection and tracking software and the effects of HGF and Rap1 activation on the analyzed cells and whether the feature was taken along for BNI. Response to HGF* Cell characteristics

Size Migration

Cell related FA characteristics

Cell region localization Number of FAs Shortest distance to the cell border

FA characteristics

Size Velocity Intensity Angle between FA axis and line from FA to cell center Roundness Axis length Solidity (area/convex area) Lifetime

n

No. of cells that responded/total no. of cells included for this condition.

Response to Rap1 activation*

In Bayesian network?

↑1/4 ↑4/4

↑6/6 ↓6/6

Yes Yes

0/4 0/4

↑5/6 ↑5/6

No Yes Yes

↑↓2/4 ↑2/4 0/4 0/4

↓6/6 ↓4/6 ↓6/6 0/6

Yes Yes Yes Yes

↓1/4 ↑1/4 ↑1/4 ↓2/4

↓2/6 ↓4/6 ↓2/6 ↑1/6

Yes Yes Yes Yes

386

E XP E R IM ENTA L CE L L R E S EA RC H

localization in the cell and to look at the characteristics of FAs in retracting, protruding or central cell areas. Supplementary material related to this article can be found online at http://dx.doi.org/10.1016/j.yexcr.2014.10.012.

33 0 (2 015 ) 38 2 –3 97

The output of our detection and tracking software contains several files, either csv or excel, with information per track, a binary image of each time point with all the detected FAs and an RGB image of the FAs color-coded in the original image according

Fig. 2 – Optimization of detection settings (A) One of 3 original images for the threshold and detection optimizer, overlaid with the outline of the binary input image (red). (B) Definition of the accuracy measures. Left: ‘r’,the pearson correlation coefficient, where P(input) are the binary pixel intensities of the input image, P(result) the binary pixel intensities of the segmented image and #P (Cell) the total amount of pixels in the cell. Right: FA count, the ratio of the number of FAs in input image over result image was mirrored using 1þ((ratio)   1). Thus FA count is 1 when there is equal amount of FAs in both input and result image and decreases to 0 when equality is lost in either direction. (C) The adaptive threshold optimizer: example results of segmentation, overlaid in red on the original image, with three different adaptive threshold lines (top row) and zooms of the green dotted square (lower panel). Plot of threshold lines and the accuracy of these segmentations (right). Line 3, in green, yielded best result. (D) The hDome parameter optimizer: examples of different parameter sets for the hDome detection (top). Segmentation results of two sets are shown overlaid in red on the original image (top row) and zooms of the green dotted square (lower panel). (E) Overview of the accuracy values of the two sets shown in (D), with in green the set that yielded best result. (F) Graph of the accuracy values of the 10 parameter sets from the table in (D).

EX P ER I ME NTAL C E LL RE S E ARCH

to their cellular location. FA features include lifetime, size, sliding, intensity, orientation and circularity that can be related to the subcellular location of the FA and also to cell size and cell migration since the cell itself is also detected and tracked over time. Table 1 shows a complete list of cellular, subcellular and FA parameters included in the analysis software. In addition, the definition of each feature is listed in the Materials and methods section. Our software (Adquant) will be made publically available at our website.

Parameter optimization In total the detection method contains 8 user defined parameters, an overview of all the input parameters is depicted in Table S1. Five parameters (h, σ, p, rad and minNum) are input for the hDome segmentation and three parameters (lowpart, a and b) for the boundary-detection. To determine the optimal values for these parameters for each dataset to be analyzed, we have developed an automated parameter optimizer that requires 3 manually segmented images and the originals, representative for the dataset (Fig. 2A). The optimizer consists of 2 parts: in the first part the threshold parameters used for FA boundary detection (lowpart,

330 (2015) 3 82 – 397

387

a and b) are optimized and the second part optimizes the hDome parameters (h, σ, p, rad and minNum). To optimize the parameter values, the accuracy of detection is determined by two measures: the first measure is ‘r’, the correlation coefficient between pixel intensities (P) of the binary input image and a binary image of the software generated FA detection (Fig. 2B, left). If the two images have the same value in each pixel, ‘r’ will be 1 and the detection is exactly like the manual detection. If none of the pixels has the same value, ‘r’ will be 0. The second measure is FA count, the ratio of number of FAs between input and software generated images. If there are more FAs in the software generated image than in the input, this ratio is inverted according to the formula depicted in the legend of Fig. 2B (right panel). Thus the optimizer generates the parameter values that result in optimal detection and at the same time quantifies the accuracy of detection (in comparison to the manually segmented input images), which we call the accuracy in case of the adaptive threshold optimizer (Fig. 2C, right panel) and precision value for the hDome detection optimizer (Fig. 2E and F). In the first part of the adaptive threshold optimizer, the 3 original test images are each iteratively segmented using different parameter values for the threshold line, in Fig. 2C examples are

Fig. 3 – (A) Stills of the original image series (top panel) and overlaid with the outline of the segmentation (lower panel) of a representative cell filmed in the transition of being untreated (t-1 min) to the addition of a combination of ML7 and Y27632 (tþ). (B) FA size first averaged over all FAs within a cell and then averaged over all 15 imaged cells (error bars are standard deviations among the 15 cells). Arrow depicts timepoint of addition of ML7 and Y27632. (C) FA intensity first averaged over all FA within one cell and then averaged between all 15 imaged cells (error bars are standard deviations among the 15 cells). Arrow depicts moment of addition of ML7 and Y27632.

388

E XP E R IM ENTA L CE L L R E S EA RC H

shown of segmentation of 1 image using three different threshold lines (Fig. 2C, images and graph). The resulting accuracy measure ‘r’ is shown in Fig. 2C for this image (top row in table) and as a sum of the 3 test images (bottom row in table). The line resulting in the highest cumulative accuracy ‘Total r’ (max¼ 3) is used in the second part of the optimizer for the hDome parameter set. In the second part again the 3 test images are iteratively segmented using different parameter values, different sets are depicted in Fig. 2D (top panel). Resulting segmentations of 2 different sets are shown in Fig. 2D (images). To determine the optimal hDome parameter values, for each image both ‘r’ and ‘FA count’ are calculated (examples shown in Fig. 2E). The 2 measures are summed up and the total precision value is calculated as the sum of this value for all 3 input images (max¼6). The parameter set with the highest precision value is then chosen for the final detection and analysis of FAs in the full dataset (Fig. 2F). Parameter optimization and FA detection using this software was similarly successful in A549-B14 cells expressing GFP–paxillin and MDA-MB-231 cells expressing mCherry–paxillin (Fig. S2). To directly compare our method with the FA detection and tracking method developed by Berginski et. al. [32,33] we uploaded our images to their online function. Fig. S3 and Supplementary movies S4 and S5 show that FAs are detected with similar accuracy, but that boundary detection is more accurate by our software (Fig. S3 and Supplemental movies S4 and S5). Supplementary material related to this article can be found online at http://dx.doi.org/10.1016/j.yexcr.2014.10.012. Thus we conclude that we have developed a robust method for the detection and tracking of FAs that is specifically useful for the analysis of FA features that rely on accurate boundary detection.

Adquant analysis reproduces well-described tensionsensitive changes in FA size and intensity FAs are tension sensitive structures. Both size and intensity are controlled by the contractility of the connected actomyosin cytoskeleton; when contractility is reduced by inhibition of myosinII activity, the FAs shrink and lose intensity until they disappear [20]. To compare changes detected by our software to these previously published observations, we analyzed FA size and intensity in cells in which MyosinII activity was reduced by inhibiting 2 upstream kinases: myosin light chain kinase by ML7 and Rho-kinase by Y27632. Immediately upon addition of ML7 and Y27632 both FA size and FA intensity began to decrease in all imaged cells. This decrease continued for up to 10 min after addition of the inhibitors (Fig. 3B and C). A representative cell and the detection at crucial timepoints are shown in Fig. 3A, the average size and intensity of all FA detected in 15 cells are shown in Fig. 3B and C. This analysis showed that our software accurately detects FAs also when their intensity is strongly reduced and shows that the subsequent analysis of morphology features reproduces well-described effects on FA characteristics. Next we have used our software to analyze a bigger set of FA parameters (see Table 1) in the context of their cellular location and the effects of HGF and Rap1 activation on FA behavior.

HGF and Rap1 regulate cell migration To study the effects of HGF and Rap1 signaling on cell migration and focal adhesion dynamics, we used the HGF responsive

33 0 (2 015 ) 38 2 –3 97

humane prostate cancer cell line DU145. This cell line was reported to express endogenous Epac [34], the cAMP-responsive GEF of Rap1 [35] that can be activated by the Epac selective cAMP analog 007 (2O-Me-8CPT-cAMP) [36]. However, the Epac levels we detected on western blot were very low (Fig. 4B, top) and to boost the magnitude of Rap1 activation by 007, the DU145 cell line was stably transfected with exogenous Epac (Fig. 4B bottom). The resulting cell line, DU145-G4, displayed scattering upon HGF treatment (Fig. 4A, 2nd column) as cells disrupted their cell–cell junctions and increased their migration velocity (Fig. 4D). Upon Rap1 activation by 007, cells flattened (Fig. 4A, 3rd column) and cell migration was slowed down (Fig. 4D). Simultaneous addition of 007 together with HGF completely prevented the scattering response in the DU145-G4 line (Fig. 4A, 4th column and 4D) whereas the DU145-wt line showed an impaired HGF response (Fig. 4C and see Supplementary movie S6). Thus, the DU145-G4 cell line, stably transfected with mCherry–Paxillin as a general FA marker, is used to study effects on FA dynamics of HGF induced migration and subsequent inhibition of migration by Rap1 activation. Supplementary material related to this article can be found online at http://dx.doi.org/10.1016/j.yexcr.2014.10.012. A number of factors limit the long term high magnification TIRF imaging, which would be needed to capture FA dynamics at baseline and upon HGF-induction and subsequent Rap1 activation in one cell: cells migrate out of the field of view; slight drifts in focus reduce FA detection accuracy, bleaching reduces FA detection accuracy, photo-toxicity reduces migration and FA dynamics. Therefore, we separately imaged cells in the transition from baseline-to-HGF and from prolonged HGF-to-Rap1 activation (007 addition). Upon manual inspection we selected the cells that showed the clearest response to HGF and 007. Our dataset therefore consists of 4 cells that were imaged in the transition from baseline-to-HGF and 6 cells that were filmed during the transition of prolonged HGF-to-007 addition. Two stills of each cell, one before and one after addition of either HGF or 007, and the FA detection by Adquant are shown in Supplementary Fig. S1-1 and S1-2. As shown in Fig. 4D, HGF induced a gradual increase of cell displacement in the 4 cells within the first 3 h of addition. This is an increase of up to 2 fold on average and there is variability between cells (Fig. 4D, middle). Addition of 007 to the 6 cells that were stimulated with HGF for 3 h prior to the start of imaging resulted in a rapid decrease in cell displacement, again with some variability between cells (Fig. 4E, middle). Concomitant with the reduction in migration, the 007 stimulated cells flattened which increased the size of the area in contact with the ECM-coated glass surface (Fig. 4E, right).

FA size, intensity and velocity differ among 3 lifetimebased FA subgroups To define baseline FA dynamics, we used our software to detect and track individual FAs in the 4 untreated mCherry–paxillin expressing DU145-G4 cells, prior to HGF. We measured in each frame, for each FA, the size, the background-corrected average fluorescence pixel intensity and the frame to frame velocity. In Fig. S4A the number of FAs is depicted that is analyzed per cell. A large population of FAs lives up to 4 min, whereas only a small subset of FAs lives beyond 30 min (Fig. 5A). It has been observed in numerous studies that short-lived, newly formed adhesions in

EX P ER I ME NTAL C E LL RE S E ARCH

330 (2015) 3 82 – 397

389

Fig. 4 – 007 blocks HGF-induced scattering in Epac-expressing prostate carcinoma cells. (A) Stills from an image series of a representative scatter assay using the Epac-overexpressing DU145_G4 cell line at the start (top row), and at 2 distinct time points after the indicated treatments. (B) Western Blots revealing Epac expression in DU145 cells and concomitant activation of Rap1 by adding the Epac-selective cAMP analog 007. (C) Quantification of the cell velocity by automated cell tracking in DU145 wild type cells [56], averages of 3 image series per condition, arrow depicts time of HGF and 007 addition at frame 16. (D) Similar to (C) but of the DU145-G4 cell line. (D) (Left) The average migration of the 4 different cells that were stimulated with HGF during imaging (red line). Each blue dot is the migration of one of the cells at that moment in time. (Middle) Average cell displacement of the 4 cells prior to (white) and after (gray) HGF treatment. (Right) Average cell size of the 4 cells prior to (white) and after (gray) HGF treatment. (E) (Left) Similar to (D Left) but then showing the 6 different cells that were stimulated with HGF 3 h prior to imaging and in which Rap1 was activated during imaging. (Middle) Average cell displacement of the 6 cells before (gray) and after (black) Rap1 activation. (Right) Average cell size of the 6 cells before (gray) and after (black) Rap1 activation.

390

E XP E R IM ENTA L CE L L R E S EA RC H

the lamellipodia are generally smaller and less motile than longer lived adhesions in the retracting area of the cell, but a systematic investigation into different populations of FAs was to our knowledge not performed.

33 0 (2 015 ) 38 2 –3 97

To investigate whether correlations exist between FA lifetime and FA size, velocity or intensity, we plotted for each of the 4 cells the different FA characteristics against the lifetime per FA and determined the correlation coefficient (Fig. 4B). For FA size

Fig. 5 – Three lifetime based FA populations show differences in FA characteristics. (A) Histogram of lifetime distribution within a representative untreated cell. (B) Scatter plots of FA size (left), FA intensity (middle) and FA sliding velocity (right) versus their lifetime of all FAs within a representative, untreated cell. Each dot represents an FA. (C) Correlation coefficients of each scatter plot for all 4 cells. Significance (po0.05) displayed for the test of the hypothesis to obtain this correlation coefficient if the real correlation coefficient would be 0. (D) Distribution of average size of all FAs and of the FAs in the 3 lifetime-based subpopulations within a representative untreated cell (left) and the average and standard deviation of the distributions (right). Stars indicate if the difference between the populations is significant (po0.01). (E) Distribution of average FA pixel intensity of all FAs and of the 3 lifetime-based subpopulations within a representative untreated cell (left) and the average and standard deviation of the distributions (right). Stars depicted if the difference between the populations is significant (po0.01). (F) Average FA size of the 3-location-based populations of FAs in the 4 cells once HGF is present, averaged over all 4 cells (Error bars ¼standard deviation between cells). (G) Average FA intensity of the 3-location-based populations of FAs in the 4 cells once HGF is present, averaged over all 4 cells (Error bars¼ standard deviation between cells). (H) Average FA sliding velocity of the 3-location-based populations of FAs in the 4 cells once HGF is present, averaged over all 4 cells (Error bars ¼standard deviation between cells).

EX P ER I ME NTAL C E LL RE S E ARCH

and intensity the correlation coefficients of the 4 cells ranged between 0.15 and 0.40 and for velocity they ranged between  0.03 and  0.15. We next calculated the probability of obtaining these correlation values in the case of a non-correlated distribution. This probability was significantly low (po0.05) for the correlation of intensity vs lifetime and size vs lifetime, but not for velocity vs lifetime (Fig. 5C). This indicates that there is a linear correlation among FA size, intensity and FA lifetime. Next, we divided the FAs into lifetime-based populations and determined the distribution and averages of the FA characteristics in these subgroups. Based on literature [17,37], we divided the FAs into 3 groups: one group of FAs that have a lifetime of less than 4 min (short lived), one group of FAs that have a lifetime of 4–30 min (intermediate lived) and a group of FAs that have a lifetime of more than 30 min (long lived). The distribution of average FA size and average FA pixel intensity is significantly different for the 3 groups of FAs (Fig. 5D and E). Both average FA size and average FA intensity increase in the intermediate and long lived population. Together this indicates that FA size and intensity differ depending on how long FAs live. FA features have also been described to differ between the cellular regions. FAs in protruding areas are smaller and less motile

330 (2015) 3 82 – 397

391

than in retracting areas [17,37], whereas FAs in retracting regions are found to slide inwards [38]. To check whether we observed this in our cells, we further analyzed the FAs in the 4 cells that were filmed in the transition to HGF but after HGF had been added, as the cells become more motile. In Fig. S5B, gray bars, the number of FAs is depicted that is analyzed per cell. We formed 3 subpopulations of FAs based on their location: the FAs that spend at least 30% of their lifetime in a protruding area (protruding), at least 30% of their lifetime in a retracting area (retracting) or less than 30% of their lifetime in either and therefore mainly in a central region (central) and measured the average FA size, intensity and sliding velocity (Fig. 5F–H). Overall, a trend of increase is visible among protruding, central and retracting located FAs. FAs in a retracting area were larger than in protruding and central FAs (Fig. 5F), this was significant (po0.01) in 2 out of the 4 cells. Retracting FAs were more intense than the protruding and central FAs (Fig. 5G), again significant in 2 of the 4 cells (po0.01). Protruding FAs were less motile than retracting FAs (Fig. 5H), significant in 3 out of 4 cells (po0.01). Thus, these analyses indicate that different lifetime and location-based FA populations can be distinguished in our data with characteristics that are in agreement with previously described FA features.

Fig. 6 – HGF and Rap1 signaling affect only a few FA features. (A) Histograms of the lifetime of FAs in cells that were treated with HGF 3 h prior to imaging, before and after Rap1 activation. (B) Distribution within one representative cell of the average FA sizes, HGF was added 3 h before imaging, before (black) and after (gray) Rap1 activation. (C) Distribution of the average FA intensities, within one representative cell, HGF was added 3 h before imaging, before (black) and after (gray) Rap1 activation. (D) Average FA velocity per cellular region in a representative cell before (white) and after (gray) Rap1 activation, HGF was added 3 h prior to imaging. Stars depicted if the difference between the regions is significant (po0.01). (G) The result of subtraction of 2 images (t2 t1) at time ‘t’ and ‘tþ11.7 min’ of the same area of a migrating cell showing FA movement in HGF (top) and loss of movement upon Rap1 activation (bottom). The subtraction image is overlaid with the outline of the detected FAs in ‘t’ (magenta) and in ‘tþ11.7 min’ (yellow). (H) The percentage of cells in which the adhesion velocity per population is either significantly decreased (black) or unchanged (gray).

392

E XP E R IM ENTA L CE L L R E S EA RC H

HGF-induced cell migration and subsequent inhibition by Rap1 signaling do not strongly correlate with differences in FA dynamics and morphology HGF and Rap1 have opposite effects on cell migration in 2D culture [39]. Assuming that alteration in FA dynamics drive alterations in cell migration and that these are therefore coregulated processes, we hypothesized that these signaling pathways would thus affect specific measurable parameters of FA morphology and FA dynamics soon after their activation. To test this, we used Adquant to detect and track all FAs in the 10 previously described cells (Fig. S1-1 and S1-2). The number of analyzed FAs per cell is depicted in Fig. S4. The features that were measured and whether and how they changed in response to HGF and Rap1 activation are indicated in Table 1. Surprisingly, during the induction of cell migration (Fig. 4D), none of the cell-related or independent FA parameters strongly changed upon treatment of cells with HGF (Table 1). Rap1 activation did result in a number of clear changes in FA characteristics. Consistent with the induction of cell-spreading (Fig. 4E), new FA formation was induced for a short period of time upon activation, enlarging the total number of FAs for the duration of imaging (Fig. S5). Consistent with earlier observations [39], Rap1 activation caused a shift in the distribution of FA lifetimes averaged over all cells in that the percentage of shortlived FAs decreased and a population of long-lived FAs appeared (Fig. 6A). Most striking, however, are the effects of Rap1 activation on FA size, intensity and velocity (Table 1). As shown in Fig. 6B and C, the average reduction in size after 007 is the result of a shift of the entire population of FAs. The reduction in average intensity represents the loss of a population of highly intense FAs. Because FA sliding is found higher in the rear compared to the front of a migrating cell (Fig. 5H), we further analyzed the effects of Rap1 on FA sliding velocity in different regions. To this end, first the velocity was averaged per FA over time and then this average velocity was averaged over all FAs within the 3 populations both before and after treatment with 007 (Fig. 6D). Rap1 activation decreased FA sliding velocity in the protruding, retracting and central populations of respectively 2, 4 and 5 of the 6 cells analyzed (Fig. 6E and F). The effects of Rap1 activation on FA velocity, size and intensity were also observed in cells stably expressing GFP–vinculin (Fig. S6), indicating that the effects of Rap1 activation on FA characteristics are general and not paxillin specific. To evaluate the consistency of the cellular behavior, the imaging and the detection and tracking results, we performed a separate experiment to produce a second set of imaging data. The results of the analyses of this experiment are shown in Figs. S7 and S8 and reproduce the results depicted in Table 1 and Figs. 4 and 6: HGF induces cell migration (Fig. S7A), but has no strong effects on any of the measured FA characteristics within the first 3 h of stimulation (data not shown); Rap1 activation acutely inhibited cell migration, increased cell size (Fig. S7B) and reduced FA size, intensity and sliding velocity (Fig. S8). In conclusion, Rap1 activation inhibits the sliding of FAs that is observed in the rear of migrating cells and reduces overall FA size, as well as the presence of a population of intense FAs. These specific effects could be explained by a reduction in tension in the FA-associated actomyosin cytoskeleton through the activation of a recently identified Rap1 effector pathway that results in

33 0 (2 015 ) 38 2 –3 97

ArhGAP29-induced inactivation of Rho [40]. Surprisingly, most FA parameters were not clearly affected by HGF within the first 3 h of stimulation, even though cell migration was induced. As a consequence, no FA characteristics could be identified that clearly correlated with the migratory capacity of cells.

Bayesian network inference reveals dependence between FA characteristics and cell size but not migration The fact that effects on cell migration by HGF and Rap1 were not clearly correlating with effects on FA parameters led us to employ BNI to assess dependencies between changes in FA parameters and cell migration parameters in an unbiased, systematic manner. For each cell, and per frame, the FA characteristics, except lifetime (Table 1) and the cell related FA characteristic ‘FA distance to the cell edge’ were averaged over all the FAs present; the cell related FA characteristic ‘number of FAs’ was included; and cell size and cell displacement (distance between center of mass in frame (t) and frame tþ10) were included. As shown in Fig. 7, a negative correlation was calculated by this model between cell size and cell displacement that decreased upon stimulation by HGF. Strikingly, hardly any direct correlations between cell displacement and any of the FA characteristics were observed in any of the conditions. Instead 5–7 of the nine FA characteristics included showed correlation with cell size. Some correlations between FA characteristics did change with the acute or prolonged induction of HGF signaling and upon activation of Rap1, but in all of these conditions strong correlations existed between FA characteristics and cell size, but not between FA characteristics and cell displacement. Together, these network models indicate that: 1) the correlations between FA characteristics and cell migration rate are surprisingly little. 2) Changes in FA characteristics do not clearly influence cell migration. 3) FA characteristics and cell size do show clear interdependence and 4) the effects of Rap1 on FA characteristics represent its induction of cell spreading (size increase) rather than its inhibition of cell displacement.

Discussion In this study we have developed a software tool to track and analyze characteristics of FAs over time. This method is generally applicable to detect and analyze FA features of single FAs over time with high confidence. We used this tool to analyze how HGF and 007-induced Rap1 signaling affect FA dynamics in order to investigate whether these signaling pathways target specific aspects of FA behavior to bring about their opposing effects on cell migration. We found that HGF has surprisingly little effects on FA dynamics and characteristics and that neither in HGF-induced, nor 007-inhibited migratory conditions clear dependencies exist between migratory behavior and FA characteristics. Instead dependence between FA characteristics and cell spreading emerged to be strong. Concomitant with the induction of cell spreading, 007 decreased FA size and intensity and inhibited FA sliding. All of these effects mimic the effects of inhibition of RhoA mediated cytoskeletal contractility, which has recently been identified as a downstream target of Rap1. Over the years, several custom made FA detection software tools have been written. Recently, different automated FA tracking software has been published [32,41,42]. The segmentation

EX P ER I ME NTAL C E LL RE S E ARCH

techniques and research questions differ between methods. Wurflinger et al. designed an 8 step method in which they separate foreground objects using top hat filtering, carefully correct for the merging of close FAs objects by an Otsu threshold and track FAs by object overlap with movement correction and gap closure. It is a very carefully designed method that is powerful in both segmentation and tracking, shown by extensive testing, either with manually segmented images as well as by omitting one by one all the steps they use in the algorithm to validate its strength. Furthermore they reproduce findings on FA dynamics using known experimental data. So far, this method has not been used for the analysis of FA in the context of a biological question. Möhl et al. developed a method to compare measurements from

330 (2015) 3 82 – 397

393

traction force microscopy, actin flow microscopy and fluorescent microscopy monitoring of FAs. For FA segmentation they use a method that has been described previously [43] in which cytoplasmic background staining is reduced by highpass filtering and adhesion patches are segmented by thresholding and linked over time using object overlap. For length aspect ratio measurements, they fit ellipses to the found objects. For comparison between actin flow, FA dynamics and force traction images, they map the datasets into a cell shape restricted to a unit circle. Using this normalization, they find that at the largest adhesion sites f-actin flow is low and traction force is high, which is consistent with previous studies [10,44–46]. Berginski et al. also base their FA detection on the above mentioned algorithm [43], which they

Fig. 7 – Bayesian network inference shows correlation between cell size and FA behavior. Graphical representations of the correlations between FA dynamics and cell behavior calculated by Bayesian network inference. Top row: the transition of adding HGF (right) to unstimulated cells (left). Bottom row: the transition of Rap1 activation (right) in cells treated with HGF 3 h prior to imaging (left). Pairwise correlations are shown in red if negative and in black if positive while the strength of correlations correspond to line thickness. Note that the structure of the graphs is calculated by BNI, whereas correlation is used to visualize the pairwise strength of each line.

394

E XP E R IM ENTA L CE L L R E S EA RC H

slightly adapt, and include a nearest neighbor linking to complement object overlap [32]. As proof of principle they analyze the FA dynamics in NIH3T3 cells stably expressing a paxillin mutant that cannot be phosphorylated anymore by c-Jun N-terminal kinase (JNK) on S178, which inhibits cell migration. In this setting they find that FAs live longer in the paxillin S178A expressing cells. In our new method, we use a 3-step segmentation and detection algorithm. For the initial identification of the FAs we chose to use the hDome method, because it was previously found to perform the best in a comparison of spot-detector algorithms [29]. Then we defined separated image regions, each containing one FA, by watershed segmentation using the hDome detected spots as markers. FA boundary was determined using a unique, local adaptive threshold. FA linking was by object overlap and gap closing, which is very similar to other methods. Compared to other methods, we have specifically focused on reliably determining the boundaries of FAs, which are intrinsically hard to discriminate from the surrounding basal cell area due to the diffusive nature of the transition. The use of a local adaptive threshold increased the precision and consistency of boundary determination and we developed an optimizer that identifies the best set of parameters for the hDome method as well as the adaptive threshold. From visual inspection of the tracking results (Fig. S1-1 and S1-2 and Supplementary movies S1 and S2, Right panels) we conclude that the boundary detection is highly accurate until FAs are in very close proximity, because then they may be merged together by the detection algorithm. The comparison of our software to that of Berginski et al. (Fig. S3 and Supplementary movies S4 and S5) confirms that boundary detection is a specific feature at which our software outperforms other approaches. It should be noted that we have not used the source code and that altering parameters in the Berginski software could very well improve its boundary detection as well. The use of source codes and proper optimization of input parameters on reference datasets would be needed to properly compare the different approaches. The accurate boundary detection renders our method specifically useful for the measurement of overall FA morphology features and for the investigation of overall signal intensities and fluctuations therein within FAs. We have included the automated generation of such measurements within our software. A second specific feature of our tool is that it uses whole-cell dynamics to define protruding and retracting areas of the cell, which allowed the correlation of FA dynamics to their cellular location. Thus 3 image analysis methods were published in time that we developed our method. Image segmentation algorithms and FA detection and tracking capacity of these methods appeared comparable to ours and FA lifetime and dynamics could be measured by these methods for our current study as well. Our method however allows the automated investigation of overall FA morphology and dependent parameters like size and intensity, which is not readily available in the previous methods. To proof the power of our method, we used our software for the analysis of FAs in DU145-G4 cells in TIRF microscopy. Consistent with a wealth of earlier studies that have built a general perception of FA dynamics and behavior in relation to cell migration, we found that small, immobile FAs concentrate at the cell-front, whereas larger, sliding FAs concentrate at the rear. In addition, we determined the effects of HGF and Rap1 signaling on FA dynamics and investigated the relation to their effects on cell

33 0 (2 015 ) 38 2 –3 97

migration. This revealed a surprising resilience of FAs against altering their characteristics even though cell migration was clearly induced within the first hours after HGF. The BNI further emphasized the lack of correlation between cell migration and FA characteristics and indicated that dependencies with cell size are much stronger. It is obvious that requirements on FA function and dynamics do exist to allow cells to migrate. For example, traction force needs to be developed and FAs need to form and disassemble at sufficient rate and in a polarized manner [11,16,47–49]. But it stems from our current analyses that these functional properties are not primarily targeted to cause alterations in cell migration, at least not in the case of its induction by HGF in this particular cell line. Rather, they may change over time to accommodate increased cell migration as dictated by other driving factors such as leading edge actin polymerization [50]. Effects of 007-induced activation of Rap1 on FA characteristics are more apparent than effects of HGF. But also here, correlations between migration and FA characteristics as determined by BNI remained minimal. Again, the strongest correlations existed between cell size (spread area) and FA characteristics. Rap1 was shown to modulate the actin cytoskeleton and increase cell spreading in numerous studies [39,51–54] and it was recently shown that this is mediated by its activation of the RhoGAP ArhGAP29 [40]. In this light, it is interesting to note that all of the FA parameters strongly affected by Rap1 (size, intensity and sliding velocity) were previously shown to be regulated by Rhodependent cytoskeletal contractility. Thus, using this unbiased, systematic analysis of FA behavior, which was developed and conducted prior to – and completely independent of – the identification of ArhGAP29 as a Rap1 target, we identified features that would be predicted to be affected based on the current knowledge. In conclusion, we have developed a new and robust FA tracking and characterizing software. The fact that our analyses recapitulated previous observations of location and lifetime-dependent FA characteristics lends confidence to the general applicability and power of the method. We further confirmed its robustness by obtaining the exact same results from 2 independent datasets. We have used it to investigate HGF and Rap1 regulated cell migration and this has identified – as Rap1 targets – FA size, FA intensity and FA sliding. In addition and surprisingly, we could not find evidence for co-regulation of overall cell migration and of FA characteristics commonly associated with migration. Our software will be made publically available and can be used for the further investigation into the relation between cell migration and FA dynamics and other systematic approaches to understand FA regulation.

Materials and methods Cell lines and culture DU145-G4, A549-B14 and MDA-MB231 cells were transfected with a lentiviral delivery system. Lentivirusal particles were produced by transfection of HEK293 cells with third-generation packaging construct [55]. Cells were plated in full medium in a 6-well plate overnight and then transduced with mCherry– paxillin, mCherry–vinculin or GFP–vinculin virus supernatant in the presence of 40 mg/ml polybrene for 24 h before returning the

EX P ER I ME NTAL C E LL RE S E ARCH

cells to full medium. mCherry–paxillin or GFP/mCherry–vinculin expressing cells were selected by fluorescence-activated cell sorting (FACS) by their fluorescence levels. 48 h prior to imaging cells were plated in Roswall Park Memorial Institute (RPMI) medium (Gibco) supplemented with 10% fetal calf serum (FCS) (Sigma) in collagen-type 1 (10 mg/ml; PureCol)-coated, Lab-Teks #1.0 Borosilicate Chambered System, 8 units. 24 h prior to imaging, the medium was replaced by 0.5% FCS containing RPMI. 25 ng/ml HGF was added either 3 h prior to imaging or after approximately 1 h of imaging. 8-pCPT-20 -O-MecAMP-AM (007AM) (Biolog Life Sciences) was added, at a concentration of 1 mM, to HGF pretreated cells after an hour of imaging. For the scatter assay, cells were treated similarly but plated in a coated 48-well plate. Before imaging the wells were completely filled with medium and the plate was sealed using silicon grease and a glass plate. Either HGF alone, 007AM alone or a combination of HGF and 007AM was added to the cells after 1 h of imaging.

Microscopy For imaging we used a NIKON Ti microscope equipped with a total internal reflection fluorescence (TIRF) system, a CFI Apochromat TIRF objective, 60  oil with an NA of 1.49, a 12-bit electron multiplying CCD (EMCCD) Luka camera (Andor) and controlled by company software (NIS-elements). For the tension inhibition experiment (Fig. 3) we have created a dataset of 15 sets of image series of mCherry–paxillin expressing DU145-G4 in which tension was inhibited after 10–15 min of imaging using a combination of 10 mM of both ML7 and Y27632. Cells were filmed with a frame rate of 10 s. We have created a dataset of image series of the DU145-G4 cells with mCherry–paxillin in the transitions from untreated to HGF and from HGF to 007. The data set consists of 12 cells imaged from epithelial to migratory induced by HGF and of 18 cells that were incubated with HGF 3 h prior to imaging and in which Rap1 was activated during filming via Epac by the cAMP analog 007AM. Within this dataset there are 4 out of 12 cells that clearly responded to HGF, shown by average cell displacement before and after adding HGF (Fig. 4D) and 6 out of 18 that were migratory at the start of imaging and in which Rap1 activation inhibited migration (Fig. 4E). The scatter assay for the characterization of the Du145-G4 cell line was carried out on a Leica AF7000 microscope, with a 10  objective. Images were taken every 10 min and the analysis was carried out using CellTracker, a custom developed Matlab analysis tool [56].

Parameter measurements Cell size: Number of pixels the cell area covers, as defined by the cell segmentation based on the signal in the TIRF image. Cell displacement: Distance traveled by the cell center from frame to frame, in pixels or μm, recalculated using the pixel size of the image (1.3 μm). For BNI the distance traveled was measured from frame t to frame tþ11. Cell migration: Cell displacement in μm divided by the frame rate in seconds.

330 (2015) 3 82 – 397

395

Cell region localization of FA: Protruding and retracting areas are determined as described in Fig. 1D. For each frame the FAs are tagged ‘1’ when in a retracting area, ‘2’ when in an unchanged area and ‘3’ when in a protruding area. Number of FA: Total amount of tracks that exist in the image series. Shortest distance of FA to cell border: Minimum value of the Euclidian distances from the center of the FA to all the pixels on the cell edge. FA size: Number of pixels the FA covers. FA velocity: Distance traveled by the FA from frame to frame, in μm, divided by the frame rate, in seconds. FA orientation: The absolute sine of the angle between the major axis length of the object and the line from the center of the cell to FA. Values are between 0 and 1. 0 means FA is oriented parallel along the line from the cell center to the FA, 1 means the FA is oriented perpendicular to the line from the cell center to the FA. FA roundness: Parameter specifying the eccentricity of the FA, defined in the Image Processing toolbox of MatLab R2013. It is the ratio of the distance between the foci if the object would be an elipse and its major axis length. The value is between 0 and 1: 0 means the object is a circle, whereas 1 indicates the object is actually a line. FA axis length: Maximal possible line that can be drawn within the FA area. FA solidity: The ratio of the number of pixels that are in the region versus the convex area, as defined by the Image Processing toolbox of MatLab R2013. FA lifetime: Number of frames a track exists, recalculated in minutes based on the frame rate. FA Intensity: To determine the average pixel intensity in the FA, it is necessary to correct for a certain background level. To negate fluctuations in the measurement of FA pixel intensity, preferably the background should be measured locally around the FA. In order to do so, we determined a watershed on each frame with the found adhesions as markers and the found cell edge as outer edge. The average pixel intensity in each region, minus the area that belongs to the FA, was used as the local background and this was subtracted from the average intensity of the accompanying adhesion.

Statistics Statistical tests on distributions of FA velocity, FA size and FA intensity in untreated cells were carried out using the “TwoSample Assuming Unequal Variances” student's t-test of the statistical add-in of Excel. The statistical tests on FA intensity distribution per cell were carried out using the Mann–Whitney– Wilcoxon test from MatLab R2013 statistical toolbox. Input per cell were the average FA pixel intensities averaged over the entire lifetime of the FAs, comparing prior to and after addition of HGF or 007. The tests on correlation between lifetime and other FA parameters was carried out using the ‘corrcoef’ command from the MatLab R2013 statistical toolbox. The statistical analysis of the FA sliding data was carried out using the Mann–Whitney–Wilcoxon test in R [57]. Input per cell was the average velocity per FA, comparing prior to and after the addition of a compound. The FAs

396

E XP E R IM ENTA L CE L L R E S EA RC H

were divided over the three subclasses, protruding, retracting and central, and the test was done for each set separately.

Bayesian network inference The FA characteristics (Table 1) were averaged per frame and collected along with the cell size and cell displacement, calculated as traveled distance from frame t to tþ11, into datasets used in learning the network structures. Each variable in these datasets was discretized to four distinct values. The default settings of the Banjo software (Bayesian network inference with Java Objects) [58] were used to run its simulated annealing learning algorithm. Families were constrained to a maximum of 5 parents and annealing moves were made randomly from local perturbations. The fitness of networks to the data was calculated with Banjo's standard likelihood function. There were 1000 re-starts per learning run and the networks presented here were those that best fitted the data of the respective microscopy experiments. Graphviz [59] was used for visualizing the networks and R was used to compute the Pearson correlation between variables. Scripts to coordinate these tasks were written in SWI-Prolog [60] and its Real package “Integrative functional statistics in logic programming” [61].

Acknowledgments This research was funded by a grant from the Netherlands Consortium for Systems Biology (NGI Grant no. 050-060-621). We would like to thank Stefan van der Elst of the flow cytometry facility of the Hubrecht Insitute for FAC-sorting of the cell lines, Holger Rehman and Bastiaan Spanjaard for discussions on statistical analysis and Willem-Jan Pannekoek and Bas Ponsioen for critically reading the manuscript.

Appendix A.

Supporting information

Supplementary data associated with this article can be found in the online version at http://dx.doi.org/10.1016/j.yexcr.2014.10.012.

references [1] S.E. Winograd-Katz, et al., The integrin adhesome: from genes and proteins to human disease, Nat. Rev. Mol. Cell Biol. 15 (4) (2014) 273–288. [2] J.C. Kuo, et al., Analysis of the myosin-II-responsive focal adhesion proteome reveals a role for beta-Pix in negative regulation of focal adhesion maturation, Nat. Cell Biol. 13 (4) (2011) 383–393. [3] B. Geiger, J.P. Spatz, A.D. Bershadsky, Environmental sensing through focal adhesions, Nat. Rev. Mol. Cell Biol. 10 (1) (2009) 21–33. [4] K.E. Kubow, S.K. Conrad, A.R. Horwitz, Matrix microarchitecture and myosin II determine adhesion in 3D matrices, Curr. Biol. 23 (17) (2013) 1607–1619. [5] N.E. Hynes, Tyrosine kinase signalling in breast cancer, Breast Cancer Res. 2 (3) (2000) 154–157. [6] M. Nishimura, et al., Tyrosine phosphorylation of 100-130 kDa proteins in lung cancer correlates with poor prognosis, Br. J. Cancer 74 (5) (1996) 780–787.

33 0 (2 015 ) 38 2 –3 97

[7] K.B. Dunn, M. Heffler, V.M. Golubovskaya, Evolving therapies and FAK inhibitors for the treatment of cancer, Anticancer Agents Med. Chem. 10 (10) (2010) 722–734. [8] C.E. Nunes-Xavier, et al., Protein tyrosine phosphatases as novel targets in breast cancer therapy, Biochim. Biophys. Acta 1836 (2) (2013) 211–226. [9] M.J. van Nimwegen, B. van de Water, Focal adhesion kinase: a potential target in cancer therapy, Biochem. Pharmacol. 73 (5) (2007) 597–609. [10] A.Y. Alexandrova, et al., Comparative dynamics of retrograde actin flow and focal adhesions: formation of nascent adhesions triggers transition from fast to slow flow, PLoS One 3 (9) (2008) e3234. [11] C.K. Choi, et al., Actin and alpha-actinin orchestrate the assembly and maturation of nascent adhesions in a myosin II motorindependent manner, Nat. Cell Biol. 10 (9) (2008) 1039–1050. [12] A. Ponti, et al., Two distinct actin networks drive the protrusion of migrating cells, Science 305 (5691) (2004) 1782–1786. [13] D.A. Lauffenburger, A.F. Horwitz, Cell migration: a physically integrated molecular process, Cell 84 (3) (1996) 359–369. [14] M.P. Sheetz, D.P. Felsenfeld, C.G. Galbraith, Cell migration: regulation of force on extracellular-matrix-integrin complexes, Trends Cell Biol. 8 (2) (1998) 51–54. [15] M. Vicente-Manzanares, A.R. Horwitz, Adhesion dynamics at a glance, J. Cell Sci. 124 (Pt 23) (2011) 3923–3927. [16] D.J. Webb, et al., FAK-Src signalling through paxillin, ERK and MLCK regulates adhesion disassembly, Nat. Cell Biol. 6 (2) (2004) 154–161. [17] D.J. Webb, J.T. Parsons, A.F. Horwitz, Adhesion assembly, disassembly and turnover in migrating cells – over and over and over again, Nat. Cell Biol. 4 (4) (2002) E97–E100. [18] C. Ballestrem, et al., Molecular mapping of tyrosinephosphorylated proteins in focal adhesions using fluorescence resonance energy transfer, J. Cell Sci. 119 (Pt 5) (2006) 866–875. [19] C.G. Galbraith, K.M. Yamada, M.P. Sheetz, The relationship between force and focal complex development, J. Cell Biol. 159 (4) (2002) 695–705. [20] A.M. Pasapera, et al., Myosin II activity regulates vinculin recruitment to focal adhesions through FAK-mediated paxillin phosphorylation, J. Cell. Biol. 188 (6) (2010) 877–890. [21] R. Zaidel-Bar, et al., Hierarchical assembly of cell-matrix adhesion complexes, Biochem. Soc. Trans. 32 (Pt3) (2004) 416–420. [22] K.T. Chan, D.A. Bennin, A. Huttenlocher, Regulation of adhesion dynamics by calpain-mediated proteolysis of focal adhesion kinase (FAK), J. Biol. Chem. 285 (15) (2010) 11418–11426. [23] C.L. Cortesio, et al., Calpain-mediated proteolysis of paxillin negatively regulates focal adhesion dynamics and cell migration, J. Biol. Chem. 286 (12) (2011) 9998–10006. [24] S.J. Franco, A. Huttenlocher, Regulating cell migration: calpains make the cut, J. Cell Sci. 118 (Pt 17) (2005) 3829–3838. [25] I. Kaverina, O. Krylyshkina, J.V. Small, Microtubule targeting of substrate contacts promotes their relaxation and dissociation, J. Cell Biol. 146 (5) (1999) 1033–1044. [26] S. Stehbens, T. Wittmann, Targeting and transport: how microtubules control focal adhesion dynamics, J. Cell Biol. 198 (4) (2012) 481–489. [27] J.T. Parsons, A.R. Horwitz, M.A. Schwartz, Cell adhesion: integrating cytoskeletal dynamics and cellular tension, Nat. Rev. Mol. Cell Biol. 11 (9) (2010) 633–643. [28] I. Smal, et al., Rao-Blackwellized marginal particle filtering for multiple object tracking in molecular bioimaging, Inf. Process. Med. Imaging 20 (2007) 110–121. [29] I. Smal, et al., Quantitative comparison of spot detection methods in fluorescence microscopy, IEEE Trans. Med. Imaging 29 (2) (2010) 282–301. [30] L. Vincent, Morphological grayscale reconstruction in image analysis: applications and efficient algorithms, IEEE Trans. Image Process. 2 (2) (1993) 176–201.

EX P ER I ME NTAL C E LL RE S E ARCH

[31] A. Huttenlocher, A.R. Horwitz, Integrins in cell migration, Cold Spring Harb. Perspect. Biol. 3 (9) (2011) a005074. [32] M.E. Berginski, S.M. Gomez, The Focal Adhesion Analysis Server: a web tool for analyzing focal adhesion dynamics, F1000Research (2013) 682 (2013) 68. [33] M.E. Berginski, et al., High-resolution quantification of focal adhesion spatiotemporal dynamics in living cells, PLoS One 6 (7) (2011) e22025. [34] M. Grandoch, et al., Epac inhibits migration and proliferation of human prostate carcinoma cells, Br. J. Cancer 101 (12) (2009) 2038–2042. [35] J. de Rooij, et al., Epac is a Rap1 guanine-nucleotide-exchange factor directly activated by cyclic AMP, Nature 396 (6710) (1998) 474–477. [36] J.M. Enserink, et al., A novel Epac-specific cAMP analogue demonstrates independent regulation of Rap1 and ERK, Nat. Cell Biol. 4 (11) (2002) 901–906. [37] B. Wehrle-Haller, Assembly and disassembly of cell matrix adhesions, Curr. Opin. Cell Biol. 24 (5) (2012) 569–581. [38] C. Ballestrem, et al., Marching at the front and dragging behind: differential alphaVbeta3-integrin turnover regulates focal adhesion behavior, J. Cell Biol. 155 (7) (2001) 1319–1332. [39] K.S. Lyle, et al., cAMP-induced Epac-Rap activation inhibits epithelial cell migration by modulating focal adhesion and leading edge dynamics, Cell. Signal. 20 (6) (2008) 1104–1116. [40] A. Post, et al., Rasip1 mediates Rap1 regulation of Rho in endothelial barrier function through ArhGAP29, Proc. Natl. Acad. Sci. USA 110 (28) (2013) 11427–11432. [41] C. Mohl, et al., Quantitative mapping of averaged focal adhesion dynamics in migrating cells by shape normalization, J. Cell Sci. 125 (Pt 1) (2012) 155–165. [42] T. Wurflinger, et al., Automated segmentation and tracking for large-scale analysis of focal adhesion dynamics, J. Microsc. 241 (1) (2011) 37–53. [43] E. Zamir, et al., Molecular diversity of cell-matrix adhesions, J. Cell Sci. 112 (Pt 11) (1999) 1655–1669. [44] M.F. Fournier, et al., Force transmission in migrating cells, J. Cell Biol. 188 (2) (2010) 287–297. [45] M.L. Gardel, et al., Traction stress in focal adhesions correlates biphasically with actin retrograde flow speed, J. Cell Biol. 183 (6) (2008) 999–1005.

330 (2015) 3 82 – 397

397

[46] K. Hu, et al., Differential transmission of actin motion within focal adhesions, Science 315 (5808) (2007) 111–115. [47] S.L. Gupton, C.M. Waterman-Storer, Spatiotemporal feedback between actomyosin and focal-adhesion systems optimizes rapid cell migration, Cell 125 (7) (2006) 1361–1374. [48] C.M. Laukaitis, et al., Differential dynamics of alpha 5 integrin, paxillin, and alpha-actinin during formation and disassembly of adhesions in migrating cells, J. Cell Biol. 153 (7) (2001) 1427–1440. [49] R. Zaidel-Bar, et al., Early molecular events in the assembly of matrix adhesions at the leading edge of migrating cells, J. Cell Sci. 116 (Pt 22) (2003) 4605–4613. [50] T.D. Pollard, G.G. Borisy, Cellular motility driven by assembly and disassembly of actin filaments, Cell 112 (4) (2003) 453–465. [51] X. Cullere, et al., Regulation of vascular endothelial barrier function by Epac, a cAMP-activated exchange factor for Rap GTPase, Blood 105 (5) (2005) 1950–1955. [52] M.R. Kooistra, et al., Epac1 regulates integrity of endothelial cell junctions through VE-cadherin, FEBS Lett. 579 (22) (2005) 4966–4972. [53] S.H. Ross, et al., Ezrin is required for efficient Rap1-induced cell spreading, J. Cell Sci. 124 (Pt 11) (2011) 1808–1818. [54] S.H. Ross, et al., Rap1 can bypass the FAK-Src-Paxillin cascade to induce cell spreading and focal adhesion formation, PLoS One 7 (11) (2012) e50072. [55] T. Dull, et al., A third-generation lentivirus vector with a conditional packaging system, J. Virol. 72 (11) (1998) 8463–8471. [56] D. Loerke, et al., Quantitative imaging of epithelial cell scattering identifies specific inhibitors of cell motility and cell-cell dissociation, Sci. Signal. 5 (231) (2012) rs5. [57] R: a language and environment for statistical computing, 2013. [58] J. Yu, et al., Advances to Bayesian network inference for generating causal networks from observational biological data, Bioinformatics 20 (18) (2004) 3594–3603. [59] E.R. Gansner, S.C. North, An open graph visualization system and its applications to software engineering, Software: Pract. Exp. 30 (11) (2000) 1203–1233. [60] J. Wielemaker, et al., Swi-Prolog, Theory Pract. Log. Prog. 12 (2012) 67–96. [61] N. Angelopoulos, et al., Integrative functional statistics in logic programming, in: K. Sagonas (Ed.), Practical Aspects of Declarative Languages, Springer, Berlin Heidelberg, 2013, pp. 190–205.