Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2015 Jun 1.
Published in final edited form as: Cytometry A. 2014 Feb 11;85(6):512–521. doi: 10.1002/cyto.a.22442

Quantitative Analysis of F-Actin Redistribution in Astrocytoma Cells Treated with Candidate Pharmaceuticals

Stephen Lockett 1,*, Chrissie Verma 2, Alla Brafman 1, Prabhakar Gudla 1, Kaustav Nandy 1, Yoshihiro Mimaki 3, Philip L Fuchs 4, Joseph Jaja 5, Karlyne M Reilly 6, John Beutler 7, Thomas J Turbyville 1
PMCID: PMC4385705  NIHMSID: NIHMS671560  PMID: 24515854

Abstract

Actin fibers (F-actin) control the shape and internal organization of cells, and generate force. It has been long appreciated that these functions are tightly coupled, and in some cases drive cell behavior and cell fate. The distribution and dynamics of F-actin is different in cancer versus normal cells and in response to small molecules, including actin-targeting natural products and anticancer drugs. Therefore, quantifying actin structural changes from high resolution fluorescence micrographs is necessary for further understanding actin cytoskeleton dynamics and phenotypic consequences of drug interactions on cells. We applied an artificial neural network algorithm, which used image intensity and anisotropy measurements, to quantitatively classify F-actin subcellular features into actin along the edges of cells, actin at the protrusions of cells, internal fibers and punctate signals. The algorithm measured significant increase in F-actin at cell edges with concomitant decrease in internal punctate actin in astrocytoma cells lacking functional neurofibromin and p53 when treated with three structurally-distinct anticancer small molecules: OSW1, Schweinfurthin A (SA) and a synthetic marine compound 23′-dehydroxycephalostatin 1. Distinctly different changes were measured in cells treated with the actin inhibitor cytochalasin B. These measurements support published reports that SA acts on F-actin in NF1−/− neurofibromin deficient cancer cells through changes in Rho signaling. Quantitative pattern analysis of cells has wide applications for understanding mechanisms of small molecules, because many anticancer drugs directly or indirectly target cytoskeletal proteins. Furthermore, quantitative information about the actin cytoskeleton may make it possible to further understand cell fate decisions using mathematically testable models. Published 2014 Wiley Periodicals Inc.†

Key terms: fibrillar actin, image analysis, pattern analysis, cytoskeleton, anti-cancer drugs, artificial neural network, fluorescence microscopy

The goal of the study reported herein was to quantitatively characterize the unique redistribution of fibrillar actin (F-actin) induced by three small molecules with potential anti-cancer activity against central nervous system tumors and peripheral nerve sheath tumors, which are largely refractory to conventional chemotherapies (1). Along with microtubules and intermediate filaments, actin microfilaments are one of three major components of the cytoskeleton, participating in many cellular processes in addition to their cell scaffolding function (2). Processes include cell motility, cell division and cytokinesis, vesicle and organelle movement, and cell signaling. Actin is one of the most adaptable and highly conserved proteins in eukaryotic cells. Individual subunits of globular actin (G-actin) assemble into long filamentous polymers (F-actin). These in turn form three types of known higher order complexes of F-actin: gel-like networks of cross-branched filaments, closely spaced bundles of filaments and a looser arrangement of contractile actin that include stress fibers (3). Moreover, F-actin forms a complex array of over 15 different spatially-overlapping structures at various times in cells, with different assembly factors associating with different structures (4). These higher order structures are highly dynamic and change their spatial distribution during normal cell processes and in response to physical and chemical stimuli, including the presence of anti-cancer drugs (5). Furthermore, a large number of natural products, some with therapeutic potential, inhibit actin dynamics, and the interaction of these molecules with actin has been partially characterized at the structural level (6). For example, cytochalasins bind to the plus end of actin filaments, blocking their elongation and leading to a punctate appearance of F-actin in micrographs.

The quantification of microscopic images can provide additional information that may identify underlying molecular mechanisms, because such information can statistically validate visual observations, provide input for mathematical modeling of biological processes and reveal features not visible to the human eye. Although quantification of subcellular features is well developed for many applications (7,8), quantification of F-actin features has been restricted in several respects. First, several studies were limited because they quantified only the F-actin structures that appear as fibers in cells (9–14). Other studies quantified the overall spatial localization of F-actin in cells but did not take into account different structural features of F-actin (15–17). Yet other studies limited quantification of F-actin to certain regions of the cell, such as lamellapodia (18–20) or invadopodia (21), and so fail to give a global understanding of the relative amounts of actin devoted to different structures in the whole cell. A few studies have quantified both fibrous and non-fibrous F-actin structures in response to external stimuli. For example, Leemreis et al. quantified F-actin in endothelial cells revealing a decrease in levels, an increase in fragmentation and changes in F-actin directionality after renal ischemia and reperfusion (22). In another example, Fuseler et al. quantified changes in F-actin in cardiac fibroblasts in response to mechanical stress by measuring the total amount of F-actin, shape descriptors and the fractal dimension per cell, of which the latter measures inherent complexity in the pattern of a structure (23). This study additional compared levels of F and G actin measured from cell images. Other investigators have quantified changes in cells induced by actin targeting reagents, but not explicitly quantified actin. For example, Sumiya et al. used high content microscopy (HCM) measurements of cytoplasm area and protrusion of the nucleus to identify natural products that stabilize or destabilize actin, or had other modes of action (24). Ng et al., also using HCM, automatically identified macrophage features that changed significantly with dosage of actin disrupting reagents and were able to super-cluster cells based on response to reagents (25). In addition, viral infection causes a profound decrease in stress fibers (26).

The three natural products we studied were schweinfurthin A (SA), OSW-1, and 23′-dehydroxycephalostatin 1 (ceph). Surprisingly, these structurally distinct compounds from unrelated plant and marine sources had similar patterns of differential cell growth activity in the NCI 60 cancer cell line assay, suggesting that they might share a common mechanism or mode of action. Furthermore, based on visual inspection of micrographs, all three lead to a unique increase in peripheral actin bundles in cell lines sensitive to the compounds. The cytoskeletal redistribution caused by SA, as well as the effects of SA on growth of certain cell lines have been correlated in part to the loss of neurofibromin (encoded by the NF1 gene in humans and the Nf1 gene in mice) in sensitive cells. Neurofibromin, which is a major tumor suppressor, is lost or mutated in the genetic disease, Neurofibromatosis type 1 and many common cancers (27). Neurofibromin normally acts to down-regulate the Ras signaling pathway and disruption of this function through mutation or deletion of the NF1 gene has major consequences for cell signaling regulation, contributing to the tumor phenotype in affected cells. NF1-deficent cells have constitutive activation of the Ras mitogenic signaling network leading to uncontrolled proliferation (28), as well as activation of the Rho pathway leading to dysregulation of the actin cytoskeleton (29). Importantly, re-introduction of the GTPase activating domain of neurofibromin confers SA-resistance to NF1−/− tumor cells, suggesting that the mechanism of SA action depends on loss of neurofibromin function. Furthermore, in addition to its antiproliferative effects in NF1−/− deficient cancer cells, SA inhibits growth factor induced Rho activation and downstream phosphorylation of the actin regulatory protein myosin light chain (MLC). Taken together, these observations suggest that SA may inhibit tumor cell growth by blocking the Rho signaling pathway and cytoskeleton dynamics in similar ways to the neurofibromin tumor suppressor (29). Little else about the signaling network is known. Oxysterol binding protein (encoded by OSBP), which is involved in lipid processing and transport, and oxysterol related protein (encoded by OSB-PL1A) are the only known direct targets of SA, OSW-1 and ceph (30), but how these small molecules or oxysterol binding proteins affect the actin cytoskeleton or neurofibromin/Ras/Rho signaling pathways has not been characterized.

Given the observed cytoskeletal redistributions of F-actin caused by SA, OSW-1, ceph, and mechanistic evidence that the effect of SA is mediated through the Rho pathway, we developed in this study a computational analysis framework to quantitatively characterize the unique redistribution of F-actin induced by SA, OSW-1, ceph and the Rho/ROCK inhibitor,Y27632. We validated the framework using manually annotated images, using control experiments with non-responsive cells and by treating cells with reagents known to affect F-actin levels and distribution. The accuracy of quantifying F-actin levels from images was checked by assessing the bulk levels of F and G actin in cells treated with SA using Western blot analysis. In test experiments, dose-dependent effects of SA, OSW-1, ceph and Y27632 on F-actin in two different responsive cell lines were contrasted with control studies. In the discussion, we present the potential of this analysis framework to further explore and elucidate the mechanisms of novel small molecules, followed by future steps for technical improvements.

Methods

In our studies, F-actin displayed four visually-distinct structural patterns in cells: peripheral actin bundles along the edges of cells, stress fibers internal to cells, actin at the protrusions of cells and internal punctuate actin (Fig. 1). Protrusive actin appeared as intense clusters of actin at the apparent leading and trailing edges of cells, whereas peripheral bundles appeared as single intense lines around straight cell edges. Stress fibers appeared as multiple, parallel lines in the cytoplasm of lower intensity than peripheral bundles, while punctate actin appeared as multiple small dots of either bright or intermediate intensity. The goal of this study was to quantify these four spatial patterns of F-actin in cells as a function of reagent concentration in order to potentially discriminate between and further elucidate the mechanisms of different small molecules in cancer cells.

Figure 1.

Figure 1

Open in a new tab

F-actin patterns in KR158 cells. A: Untreated cells showing phalloidin staining of F-actin (green) and DAPI-labeled nuclei (blue). B: Image A showing annotated examples of stress fibers (cyan bordered regions), peripheral bundles (red bordered regions), punctate actin (yellow bordered regions), and protrusive actin (white bordered regions). The annotations were delineated by hand. C: Following treatment with 100 nM of SA, cells have a greater proportion of peripheral actin bundles.

Image Analysis

The purpose of the image analysis was to measure image-based features, and to use these features to classify pixels as belonging to one of the four F-actin patterns, or the background. Conceptually, the method is analogous to the enhanced cell classifier of Misselwitz et al., where a classifier was trained from many texture features measured from manually classified objects, and then the classifier was applied to other unclassified objects (31). However in this study, we used only two features: intensity and anisotropy at each pixel, which provided sufficiently accurate and robust classification from very small training sets. The flow diagram in Figure 2 outlines the image analysis procedure. The MATLAB script is available from the corresponding author upon request. Each module in the procedure performed the following tasks:

Figure 2.

Figure 2

Open in a new tab

Flow diagram of the procedure for quantifying the proportions of different F-actin structures in cell micrographs. Briefly an artificial neural network classified each pixel as belonging to one of the four F-actin structures based on the pixel's intensity and local anisotropy in the vicinity of the pixel. Details of each module are given in the main text. HIC: high intensity class, MIC: medium intensity class, BC: background class.

  • Acquired Image: Three dimensional stacks were acquired using confocal microscopy of cells labeled with fluorescent phalloidin. The most in focus two-dimensional slice from each stack, determined by visual inspection, was selected for analysis (Fig. 3A).

  • Image rescaling: Images with a pixel size less than 0.44 μm were down-sampled using linear interpolation to a pixel size of 0.44 μm, a size typical in HCM (25).

  • K-means segmentation: Images were automatically segmented into three classes: background class (BC), medium intensity class (MIC), and high intensity class (HIC) (Fig. 3B). In order to reduce the effects of random noise, each pixel in the input image was converted to a four-element vector where the first element was the pixel's intensity from the input image and the subsequent elements were the pixel's intensity after three successive smoothings of the image by a Gaussian filter with a standard deviation (σ) of 0.5 pixels. The set of vectors for the entire image were the input to the k-means classifier (Statistics toolbox, MATLAB, Mathworks, Natick, MA). The HIC approximately corresponded to peripheral bundles and protrusive F-actin, whereas the MIC approximately corresponded to stress fiber F-actin. Internal punctate F-actin was present in both classes. The average cell size per image was calculated by dividing the area of the MIC and HIC classes by the manually enumerated number of nuclei per image.

  • Normalize image intensity: The background class was eroded with a circular structuring element of 15 pixel diameter so that the eroded background regions were at least 6 μm from any cell signal, and the regions were used to calculate the mean background intensity in the image. The mean intensity in the HIC and the mean background intensity were used to normalize acquired images to approximately the same intensity range from 1.0 (brightest pixels) to 0.0 (background), so that artificial neural network (ANN) classification was independent of absolute intensities. Absolute intensities were still used to determine changes in F-actin concentration in cells in response to reagents.

  • Calculate anisotropy: Following previous work of Weichsel et al. (11), we used “anisotropy,” calculated using the “structuretensor” filter (32) in Dipimage (version 2.0.1, Delft University of Technology, The Netherlands), to measure alterations in the elongation of F-actin. The filter was applied with Gaussian smoothing with a σ of 1.0 pixel to reduce random noise effects and with tensor smoothing with a σ of 3.0 pixels. Tensor smoothing made the filter sensitive to parallel linear structures greater than 3 pixels and spaced by at least 3 pixels. Anisotropy was defined as the difference between the larger and smaller eigenvalues divided by the sum of the eigenvalues. Figure 3C shows anisotropy measured from the image in Figure 3A.

  • Select pixels: The aforementioned tensor smoothing resulted in high anisotropy values particularly from peripheral bundles to be incorrectly assigned to other actin types a few pixels away. To mitigate this problem, pixels in the MIC and within 4 pixels of a pixel in the HIC were left unclassified by the ANN.

  • Artificial neural network: Non background pixels from the “Select pixels” stage above were automatically categorized as belonging to one of the four F-actin categories using a feed forward ANN (Neural Network toolbox, MATLAB) using the features of image intensity normalized between 0 and 1, and anisotropy. The ANN had one hidden layer with three nodes and used a tan-sigmoidal transfer function from each hidden node to the output nodes. Figure 3D shows the ANN classification of the image in Figure 3A.

Figure 3.

Figure 3

Open in a new tab

Analysis of the image in Figure 1A. A: Green (F-actin) channel from Figure 1A. B: Segmentation using the K-means classifier into background class (black), medium intensity class (grey), and high intensity class (white). C: Anisotropy image showing high intensities corresponding to elongated F-actin structures. D: Classification of image into peripheral bundles (red), stress fibers (blue), protrusive actin (green), and punctate actin (yellow).

Sample Preparation

The isolation of the natural products schweinfurthin A (SA) (1) and OSW1 were previously described (33), as well as the synthesis of 23′-deoxycephalostatin 1 (ceph) (34). Though these natural products are structurally different (30), they have a similar anti-proliferative profile across the NCI 60 cell lines and furthermore have been observed visually to cause similar distinctive changes in F-actin organization. The A549 human lung cancer line (carrying an activating mutation in Kras) and the U251 human glioblastoma line (carrying a premature STOP codon in NF1) were obtained from the Developmental Therapeutics Program (NCI-Frederick) from stocks used in the NCI 60-cell assay (35). The KR158 astrocytoma cell line was isolated from a C57BL/6J-Nf1−/+;Trp53−/+cis mouse anaplastic astrocytoma tumor as described previously (36). A549 cells were not expected to be affected by SA based on results from an earlier proliferation assay (29) and therefore served as a negative control. We used two different sensitive cell lines, U251 and KR158 to ensure that the effects of SA were not unique to one cell line. Since both cell lines lack functional neurofibromin, we anticipated that they were sensitive to SA and to the other natural products used in this study. Cell lines were cultured in DMEM containing 10% fetal bovine serum supplemented with 2 mmol/L glutamine in a 37°C, 5% CO2 humidified atmosphere. They were seeded on to cover slips and allowed to reattach overnight. Cells were treated for 18 h with ranges of concentrations (shown in Figs. 5 and 6) of the natural products, the ROCK inhibitor, Y27632 (Sigma), cytochalasin B (CB) a direct actin inhibitor, or with the vehicle control dimethyl sulfoxide (DMSO). In F-actin activation experiments, U251 cells were serum starved in 0.5% fetal bovine serum and DMSO for 18 h and then pulsed with epidermal growth factor (EGF; Invitrogen) for 5 min. Cells were fixed in 4% paraformaldehyde made up in 0.1 M phosphate buffer, permeabilized in 0.1% Triton-X100, stained with Alexafluor 488-phalloidin (Invitrogen) to detect F-actin, and mounted in Prolong Antifade containing 4′,6-diamidino-2-phenylindole (DAPI; Invitrogen) to stain nuclei.

Figure 5.

Figure 5

Open in a new tab

A, B, and C: Relative concentration of F-actin (○) and proportions of F-actin in the categories of peripheral bundles (◆), stress fibers (■), protrusive actin (▲), and internal punctate (X) as a function of reagent concentration. Each data point is the average of four measurements, and standard error bars are shown when larger than the data point. A: A549 cells did not show significant changes in F-actin as a function of the dose of SA. UN = untreated without DMSO solvent. B: EGF caused significant increase in total F-actin in serum-starved U251 cells between 1 and 4 nM, but the patterns of F-actin did not change significantly. C: In KR158 cells, internal punctate actin significantly increased and peripheral bundles significantly decreased as a function of CB concentration, whereas protrusive actin, stress fiber actin and total actin did not change significantly. D and E: Western blot analysis of actin in KR158 cells. β-actin: calibration standards. (β-actin and DMSO measurements were performed twice and are shown in D and E respectively.

Figure 6.

Figure 6

Open in a new tab

Relative concentration of F-actin (○) and proportions of F-actin in the categories of peripheral bundles (◆), stress fiber (■), protrusive (▲) and internal punctate (X) as a function of reagent concentration in KR158 cells. Each data point is the average of 4 measurements, and standard error bars are shown when larger than the data point. A: Treatment with SA. B: Treatment with ceph. C: Treatment with OSW1. D: Treatment with Y27632.

Western Blot Analysis of F and G Actin

Although it is generally accepted that measurement of F-actin labeled with fluorescent phalloidin is quantitative (37), we independently assessed the bulk changes in F and G actin levels in cells treated with SA using a G/F actin assay kit (Cat# BK037, Cytoskeleton Inc., Denver, CO). Briefly, approximately 106 KR158 cells were seeded on to 10 cm plates and allowed to reattach overnight. Cells were treated for 18 hours with DMSO, SA at concentrations of 1,000, 100, or 10 nM, or CB at 2.5 or 1.25 μM. Cells were then lysed in the kit buffer to stabilize the different forms of actin, solubilize G-actin, and to leave F- Actin in a non solubilized form. As an internal control for the assay, a DMSO lysate sample was pulsed with phalloidin to drive the G-Actin into the non-solubilized, pellet fraction. Lysates were centrifuged at 2,000 rpm in a table top centrifuge to remove debris, and then centrifuged at 100,000g in a Beckman ultracentrifuge for 1 h at 37° C to separate the soluble G-actin from the non soluble F-actin. The supernatant was transferred to fresh tubes, and the pellets were solubilized in an actin depolymerizing buffer supplied in the kit. The supernatant and pellet samples were then processed by SDS-PAGE and western blot. Since KR158 cells mainly contain β-actin, we probed the blots with a mouse monoclonal antibody to β-actin (Cat# A1978, Sigma) instead of the kit-supplied rabbit polyclonal actin antibody, and used purified β-actin (Cat# APHL99, Cytoskeleton) as a calibration standard instead of kit-supplied α-actin.

Image Acquisition

Four random fields of cells were imaged per sample on a LSM510 confocal microscope (Carl Zeiss Inc., Thornwood, New York) using either a 63×, 1.4 NA, or 40×, 1.3 NA oil objective lens and a pinhole of 1.0 Airy units, resulting in an axial resolution of approximately 0.5 μm. Alexafluor 488-phalloidin was excited using 488 nm light from an argon ion laser at a nominal intensity between 1% and 5%, and fluorescence emission was collected either between 490 and 530 nm or between 505 and 550 nm. DAPI stained nuclei were imaged using 405 nm for excitation and collecting emitted light between 420 and 480 nm for the purposes of enumerating cells per image. Detector gain was kept constant for each experiment. Four successive images were averaged together to form each final two-dimensional slice image. Adjacent two-dimensional images were separated by either 0.5 or 1.0 μm in the z direction.

Training and Validation of the Image Analysis Algorithms

We manually delineated 239 regions in 22 randomly selected images of KR158 cells treated with different concentrations of SA, ceph, OSW1, CB, Y27632, or DMSO alone. Based on visual inspection, the 239 regions were chosen such that they appeared to contain predominantly one of the four actin patterns, (Fig. 1B). Measurements within these regions were used to train the ANN for 1,000 epochs, using 60% of the regions for training, 20% for validation and 20% for testing and using Levenberg-Marquardt back-propagation for updating weights and biases. Thirty-two regions in eight randomly selected images of U251 cells treated with different concentrations of SA or DMSO alone were used to test (but not train) the ANN.

Results

Validation of the ANN

Figure 4 is a bivariate histogram plotting relative intensity versus anisotropy from the 239 annotated regions from KR158 cells corresponding to the four structural patterns of F-actin. The four patterns lie in distinct segments of the graph, with peripheral bundles having high anisotropy values whereas punctate actin has low anisotropy values. Similarly, stress fibers were significantly less intense than protrusive actin. Figure 4 is overlaid with the areas used by the ANN for classifying images. The performance of the ANN for KR158 cells was tested on 47 regions not used for training or validation, of which 35 (74%) were correctly classified. This performance was considered satisfactory for the purposes of evaluating average changes in F-actin patterns occurring across multiple cells in response to various concentrations of reagents. For U251 cells, the ANN correctly classified 30 out of 32 regions (94%) even though these cells were not used for training, suggesting that there is less variability in the F-actin patterns in U251 cells compared with KR158 cells. Figure 3D shows classification by the ANN of the original image shown in Figure 3A.

Figure 4.

Figure 4

Open in a new tab

Bivariate plot of relative intensity versus anisotropy for all manually-annotated regions in KR158 cells overlaid with the areas used by the ANN for classifying images. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

Control Experiments

To further validate the automatic analysis, three control experiments were conducted: (i) treatment of A549 cells with SA, which in previous studies did not show an antiproliferative response to SA (29), (ii) treatment of serum-starved cells with epidermal growth factor (EGF), which increases F-actin levels in cells (38), and (iii) treatment of cells with CB, which is known to disrupt F-actin in cells by blocking monomer addition to barbed ends. Additionally, the quantification of F-actin levels from images was checked by assessing the bulk levels of F and G actin in cells treated with SA. Figures 5A to 5C show the changes in the total F-actin levels and the proportions of F-actin in the four structural categories for these control experiments determined by our image analysis. A549 cells were insensitive in terms of F-actin changes up to a concentration of 100 nM SA (Fig. 5A) consistent with earlier studies in which their growth was not affected by SA up to this concentration (29). Since DMSO should allow SA to readily penetrate into A549 cells, and furthermore camptothecin enters these cells and induces apoptosis, the lack of response of these cells to SA is likely not due to drug exclusion. Thus, the morphology, survival and growth of A549 cells are governed by molecular pathways apparently not affected by SA. EGF causes actin cytoskeletal rearrangement via the Rho pathway and increases F-actin polymerization in serum-starved cells (38). Thus, as expected, we observed a significant increase in total F-actin in cells when EGF dose escalated from 1 to 4 nM, but the relative actin patterns remained unchanged (Fig. 5B). In the third control study, treatment of KR158 cells with CB caused an expected decrease in peripheral bundles, but total actin levels remained unchanged, indicating that CB is not a very effective depolymerizing reagent as has been previously reported (39) (Fig. 5C). Interestingly, the internal punctate actin significantly increased, which could be due to the production of F-actin fragments that has been observed previously in neurons (40). We additionally observed that the protrusive actin fraction increased slightly with increasing CB concentration while the stress fiber fraction remained unchanged. Taken together, these control experiments showed that quantitative pattern analysis of F-actin yielded the expected results when using well understood reagents.

Figures 5D and 5E show Western blots of F and G actin in KR158 cells treated with SA and CB. Untreated cells had approximately equal amounts of F and G actin (lanes: F-actin DMSO and G-actin DMSO) and these amounts did not significantly change with increasing doses of SA from 0 to 1000 nM. Thus, SA treatment did not significantly affect the levels of F and G actin in KR158 cells. This agreed with the result from image analysis that the total F-actin stayed constant over the SA dose range from 0 to 100 nM (Fig. 6A). Interestingly, over the same dose range the proportions of protrusive actin and internal punctate actin significantly increased and decreased respectively. This suggests that the fluorescence signal from phalloidin-labeled F-actin is unaffected by these structural forms of F-actin. In contrast, image analysis reported significant increase in F-actin at the 1,000 nM SA dose, whereas F-actin levels remained approximately unchanged by Western blot analysis. Tentatively, we attribute this paradox to significant cell death at 1,000 nM resulting in fewer cells being analyzed in the Western blot, while image analysis results were independent of total cell number. Treatment of cells with a low dose of CB (1.25 μM) did not affect average actin levels measured by Western blot analysis. However, a 2.5 μM dose of CB slightly decreased the amount of F-actin (lane: F-actin 2.5μM CB), which would be anticipated from the known disruption of F-actin by CB. As expected, converting G-actin to the pelleted form with phalloidin reduced the soluble fraction to zero (lane: G-actin Phalloidin) and concomitantly increased the non-soluble fraction (lane: F-actin Phalloidin).

Treatment of Cells with SA, ceph, OSW1, and Y27632

Both KR158 and U251 cells were treated with SA, and KR158 cells were additionally treated with ceph, OSW1, and Y27632. SA, ceph, and OSW1 caused significant increase in peripheral bundles (Figs. 6A–6C) while punctate actin significantly decreased, in contrast to CB (Fig. 5C). A similar result was obtained in two experiments with U251 cells for doses over 100 nM SA, where the ratio of peripheral bundles to punctate actin increased significantly in both experiments (P < 5% and P < 0.2% by Student's t test). Measurements also showed that average KR158 cell size shrank by approximately twofold when treated with 100 nM of SA, 10 nM OSW1, 10 nM ceph, or 10 μM CB. However, the correlation of average cell size with dosage measured by Pearson's correlation coefficient was weak and equaled 0.053, 0.52, 0.020, and 0.49 for SA, ceph, OSW1, and CB respectively. Moreover, CB caused cell size to shrink, but it had an opposite effect on actin patterns. Therefore, we do not believe that the actin changes are simply a measure of a general stress response of the cells. We went on to test whether these actin changes were caused by inhibition of the Rho-ROCK pathway by measuring the effects of the ROCK inhibitor, Y27632 on F-actin patterns. However, we did not observe a monotonic increase or decrease in the proportion of each F-actin pattern with increasing dose of Y27632 (Fig. 6D). Instead Y27632 gave a biphasic response with an increase in peripheral actin bundles at 5 μM, a decrease at 10 μM and then another increase at 20 μM. This response is hard to interpret, but it has been reported previously that the proportion of MLC2 phosphorylated at serine 19 is biphasic across similar doses of Y27632 in U251 cells (41).

Discussion

We built a computational framework for quantifying spatial patterns of F-actin in order to further characterize the effects of candidate drugs in cells. Our framework is unique, because it quantifies F-actin into four classifications that correspond to different, known structural forms that F-actin has in cells. Thus it contrasts the work of Fuseler et al. (23), who instead measured one structural feature, fractal dimension, a measure of complexity or irregularity of structure. Our framework is robust because it only requires two features (intensity and anisotropy) measured at each pixel for classification. Using this method, we measured an increase in peripheral bundles around the edges of cells with a concomitant decrease in punctate F-actin as a function of dose when cells were treated with three structurally-distinct small molecules. This F-actin redistribution was opposite to the effects of the well-known F-actin disruptor, CB. Although the three molecules have similar anti-proliferative profiles in the NCI 60 cell line screen and all have been reported to bind OSBPs (30), their similar effects on the distribution of F-actin were unexpected. The molecular mechanisms for redistribution are not understood, but interference of the Rho-ROCK pathway could explain cytoskeletal changes caused by SA (29). In order to further understand the molecular mechanisms underlying specific redistribution patterns, we envision the potential for quantifying spatial patterns of F-actin in response to the perturbation of specific pathways using RNAi and classical genetic and small-molecule strategies. For example, knocking down the Rac pathway has distinct effects on the actin cytoskeleton as compared with the Rho pathway as has been observed in numerous descriptive studies (41).

We believe that the computational framework presented here has broad application for quantifying the effects of compounds, including those with therapeutic potential, on the cytoskeleton. Such quantification can reveal cytoskeletal dynamics in response to compounds and serve to test mathematical models of cellular processes (42). Furthermore, significant advances for this approach will stem from application to micropatterned cells (43), because well-spaced cells ensure accurate single cell analysis. Micropatterns constrain each cell to a pre-defined morphology that in some cases confer known orientation to cells, thus enabling analysis at known locations in the cell. Also, we anticipate that analyzing images acquired using super-resolution optical microscopy may detect subtler structural changes to the cytoskeleton, as well as from imaging actin-binding proteins in association with actin itself.

In conclusion, we have built an automatic classifier that measures changes in the proportions of F-actin among different F-actin structures in cells. We predict a range of applications for quantifying cytoskeletal changes in cells and that such quantification will contribute to understanding the mechanisms of action of proteins and candidate drugs in cells.

Acknowledgments

Grant sponsors: National Cancer Institute and National Institutes of Health, Grant number: HHSN261200800001E

Grant sponsors: Intramural Research Program of the National Institutes of Health, National Cancer Institute and Center for Cancer Research.

Footnotes

†

This article is a US government work and, as such, is in the public domain in the United States of America.

Literature Cited

  • 1.Beutler JA, Shoemaker RH, Johnson T, Boyd MR. Cytotoxic geranyl stilbenes from Macaranga schweinfurthii. J Nat Prod. 1998;61:1509–1512. doi: 10.1021/np980208m. [DOI] [PubMed] [Google Scholar]
  • 2.Vicente-Manzanares M, Ma X, Adelstein RS, Horwitz AR. Non-muscle myosin II takes centre stage in cell adhesion and migration. Nat Rev. 2009;10:778–790. doi: 10.1038/nrm2786. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Cooper GM. The Cell: A Molecular Approach. 2nd. Chapter 11 Sunderland, MA: Sinauer Associates; 2000. [Google Scholar]
  • 4.Chhabra ES, Higgs HN. The many faces of actin: Matching assembly factors with cellular structures. Nat Cell Biol. 2007;9:1110–1121. doi: 10.1038/ncb1007-1110. [DOI] [PubMed] [Google Scholar]
  • 5.Pollard TD, Blanchoin L, Mullins RD. Molecular mechanisms controlling actin filament dynamics in nonmuscle cells. Annu Rev Biophys Struct. 2000;29:545–576. doi: 10.1146/annurev.biophys.29.1.545. [DOI] [PubMed] [Google Scholar]
  • 6.Allingham JS, Klenchin VA, Rayment I. Actin-targeting natural products: Structures, properties and mechanisms of action. Cell Mol Life Sci. 2006;63:2119–2134. doi: 10.1007/s00018-006-6157-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Hamilton NA, Wang JT, Kerr MC, Teasdale RD. Statistical and visual differentiation of subcellular imaging. BMC Bioinformatics. 2009;10:94. doi: 10.1186/1471-2105-10-94. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Coelho LP, Peng T, Murphy RF. Quantifying the distribution of probes between sub-cellular locations using unsupervised pattern unmixing. Bioinformatics. 2010;26:7–12. doi: 10.1093/bioinformatics/btq220. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Yoshigi M, Clark EB, Yost HJ. Quantification of stretch-induced cytoskeletal remodeling in vascular endothelial cells by image processing. Cytometry Part A: J Int Soc Anal Cytol. 2003;55A:109–118. doi: 10.1002/cyto.a.10076. [DOI] [PubMed] [Google Scholar]
  • 10.Lichtenstein N, Geiger B, Kam Z. Quantitative analysis of cytoskeletal organization by digital fluorescent microscopy. Cytometry Part A: J Int Soc Anal Cytol. 2003;54A:8–18. doi: 10.1002/cyto.a.10053. [DOI] [PubMed] [Google Scholar]
  • 11.Weichsel J, Herold N, Lehmann MJ, Kräausslich HG, Schwarz US. A quantitative measure for alterations in the actin cytoskeleton investigated with automated high-throughput microscopy. Cytometry Part A: J Int Soc Anal Cytol. 2010;77A:52–63. doi: 10.1002/cyto.a.20818. [DOI] [PubMed] [Google Scholar]
  • 12.Birkholz H. Extracting the ridge set as a graph for actin filament length estimation from confocal laser scanning microscopic images. J Electron Imaging. 2012;21:021110. [Google Scholar]
  • 13.Xu T, Li H, Shen T, Ojkic N, Vavylonis D, Huang X. Extraction and analysis of actin networks based on open active contour models. Proc IEEE Int Symp Biomed Imaging. 2011;2011:1334–1340. doi: 10.1109/ISBI.2011.5872647. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Basu S, Dahl KN, Rohde GK. Localizing and extracting filament distributions from microscopy images. J Microsc. 2013;250:57–67. doi: 10.1111/jmi.12018. [DOI] [PMC free article] [PubMed] [Google Scholar] [Retracted]
  • 15.Lepidi H, Zaffran Y, Ansaldi JL, Mege JL, Capo C. Morphological polarization of human polymorphonuclear leucocytes in response to three different chemoattractants: An effector response independent of calcium rise and tyrosine kinases. J Cell Sci. 1995;108:1771–1778. doi: 10.1242/jcs.108.4.1771. [DOI] [PubMed] [Google Scholar]
  • 16.Zaffran Y, Lepidi H, Bongrand P, Mege JL, Capo C. F-actin content and spatial distribution in resting and chemoattractant-stimulated human polymorphonuclear leucocytes. Which role for intracellular free calcium? J Cell Sci. 1993;105:675–684. doi: 10.1242/jcs.105.3.675. [DOI] [PubMed] [Google Scholar]
  • 17.Jones TR, Carpenter AE, Lamprecht MR, Moffat J, Silver SJ, Grenier JK, Castoreno AB, Eggert US, Root DE, Golland P, et al. Scoring diverse cellular morphologies in image-based screens with iterative feedback and machine learning. Proc Natl Acad Sci USA. 2009;106:1826–1831. doi: 10.1073/pnas.0808843106. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Schaub S, Meister JJ, Verkhovsky AB. Analysis of actin filament network organization in lamellipodia by comparing experimental and simulated images. J Cell Sci. 2007;120:1491–1500. doi: 10.1242/jcs.03379. [DOI] [PubMed] [Google Scholar]
  • 19.Burnette DT, Manley S, Sengupta P, Sougrat R, Davidson MW, Kachar B, Lippincott-Schwartz J. A role for actin arcs in the leading-edge advance of migrating cells. Nat Cell Biol. 2011;13:371–381. doi: 10.1038/ncb2205. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Urban E, Jacob S, Nemethova M, Resch GP, Small JV. Electron tomography reveals unbranched networks of actin filaments in lamellipodia. Nat Cell Biol. 2010;12:429–435. doi: 10.1038/ncb2044. [DOI] [PubMed] [Google Scholar]
  • 21.Quintavalle M, Elia L, Price JH, Heynen-Genel S, Courtneidge SA. A cell-based high-content screening assay reveals activators and inhibitors of cancer cell invasion. Sci Signal. 2011;4:ra49. doi: 10.1126/scisignal.2002032. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Leemreis JR, Versteilen AM, Sipkema P, Groeneveld AB, Musters RJ. Digital image analysis of cytoskeletal F-actin disintegration in renal microvascular endothelium following ischemia/reperfusion. Cytometry Part A: J Int Soc Anal Cytol. 2006;69A:973–979. doi: 10.1002/cyto.a.20269. [DOI] [PubMed] [Google Scholar]
  • 23.Fuseler JW, Millette CF, Davis JM, Carver W. Fractal and image analysis of morphological changes in the actin cytoskeleton of neonatal cardiac fibroblasts in response to mechanical stretch. Microsc Microanal. 2007;13:133–143. doi: 10.1017/S1431927607070225. [DOI] [PubMed] [Google Scholar]
  • 24.Sumiya E, Shimogawa H, Sasaki H, Tsutsumi M, Yoshita K, Ojika M, Suenaga K, Uesugi M. Cell-morphology profiling of a natural product library identifies bisebromoamide and miuraenamide A as actin filament stabilizers. ACS Chem Biol. 2011;6:425–431. doi: 10.1021/cb1003459. [DOI] [PubMed] [Google Scholar]
  • 25.Ng AYJ, Rajapakse JC, Welsch RE, Matsudaira PT, Horodincu V, Evans JG. A cell profiling framework for modeling drug responses from HCS imaging. J Biomol Screen. 2010;15:858–868. doi: 10.1177/1087057110372256. [DOI] [PubMed] [Google Scholar]
  • 26.Seo JY, Yaneva E, Hinson ER, Cresswell P. Human cytomegalovirus directly induces the antiviral protein viperin to enhance infectivity. Science. 2011;332:1093–1097. doi: 10.1126/science.1202007. [DOI] [PubMed] [Google Scholar]
  • 27.Friedman JM, Gutmann DH, MacCollin M, Riccardi VM. Neurofibromatosis: Phenotype, Natural History, and Pathogenesis. 3rd. Baltimore: Johns Hopkins University Press; 1999. [Google Scholar]
  • 28.Bajenaru ML, Donahoe J, Corral T, Reilly KM, Brophy S, Pellicer A, Gutmann DH. Neurofibromatosis 1 (NF1) heterozygosity results in a cell-autonomous growth advantage for astrocytes. Glia. 2001;33:314–323. doi: 10.1002/1098-1136(20010315)33:4<314::aid-glia1030>3.0.co;2-q. [DOI] [PubMed] [Google Scholar]
  • 29.Turbyville TJ, Gursel D, Tuskan RG, Walrath JC, Lockett S, Wiemer DF, Beutler JA, Reilly K. Schweinfurthin A selectively inhibits proliferation and Rho signaling in glioma and Neurofibromatosis type 1 tumor cells in a NF1 GRD dependent manner. Mol Cancer Ther. 2010;9:1234–1243. doi: 10.1158/1535-7163.MCT-09-0834. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Burgett AW, Poulsen TB, Wangkanont K, Anderson DR, Kikuchi C, Shimada K, Okubo S, Fortner KC, Mimaki Y, Kuroda M, et al. Natural products reveal cancer cell dependence on oxysterol-binding proteins. Nat Chem Biol. 2011;7:639–647. doi: 10.1038/nchembio.625. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Misselwitz B, Strittmatter G, Periaswamy B, Schlumberger MC, Rout S, Horvath P, Kozak K, Hardt WD. Enhanced cell classifier: A multi-class classification tool for microscopy images. BMC Bioinformatics. 2010;11:30. doi: 10.1186/1471-2105-11-30. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.van Vliet LJ, Verbeek PW. In: van Katwijk J, Gerbrands JJ, van Steen MR, Tonino JFM, editors. Estimators for orientation and anisotropy in digitized images; ASCI'95, Proceedings of the First Annual Conference of the Advanced School for Computing and Imaging; Heijen, NL. May 16–18; Delft: ASCI; 1995. pp. 442–450. [Google Scholar]
  • 33.Li W, Fuchs PL. An efficient synthesis of the C-23 deoxy, 17 alpha-hydroxy South 1 hemisphere and its cephalostatin 1 analog. Org Lett. 2003;5:2849–2852. doi: 10.1021/ol034894e. [DOI] [PubMed] [Google Scholar]
  • 34.Kuroda M, Mimaki Y, Yokosuka A, Sashida Y, Beutler JA. Cytotoxic cholestane glycosides from the bulbs of Ornithogalum saundersiae. J Nat Prod. 2001;64:88–91. doi: 10.1021/np0003084. [DOI] [PubMed] [Google Scholar]
  • 35.http://dtp.nci.nih.gov/branches/btb/ivclsp.html
  • 36.Reilly KM, Loisel DA, Bronson RT, McLaughlin ME, Jacks T. Nf1;Trp53 mutant mice develop glioblastoma with evidence of strain specific effects. Nat Genet. 2000;26:109–113. doi: 10.1038/79075. [DOI] [PubMed] [Google Scholar]
  • 37.Cooper JA. Effects of cytochalasin and phalloidin on actin. J Cell Biol. 1987;105:1473–1478. doi: 10.1083/jcb.105.4.1473. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Ridley AJ, Hall A. The small GTP-binding protein rho regulates the assembly of focal adhesions and actin stress fibers in response to growth factors. Cell. 1992;70:389–399. doi: 10.1016/0092-8674(92)90163-7. [DOI] [PubMed] [Google Scholar]
  • 39.Miyake K, McNeil PL, Suzuki K, Tsunoda R, Sugai N. An actin barrier to resealing. J Cell Sci. 2002;115:2603–2611. doi: 10.1242/jcs.114.19.3487. [DOI] [PubMed] [Google Scholar]
  • 40.Furuyashiki T, Arakawa Y, Takemoto-Kimura S, Bito H, Narumiya S. Multiple spatiotemporal modes of actin reorganization by NMDA receptors and voltage-gated Ca2+ channels. Proc Natl Acad Sci USA. 2002;99:14458–14463. doi: 10.1073/pnas.212148999. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Salhia B, Hwang JH, Smith CA, Nakada M, Rutka F, Symons M, Rutka JT. Role of myosin II activity and the regulation of myosin light chain phosphorylation in astrocytomas. Cell Motil Cytoskeleton. 2008;65:12–24. doi: 10.1002/cm.20240. [DOI] [PubMed] [Google Scholar]
  • 42.Pollard TD, Cooper JA. Actin, a central player in cell shape and movement. Science. 2009;326:1208–1212. doi: 10.1126/science.1175862. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Théry M, Racine V, Piel M, Pépin A, Dimitrov A, Chen Y, Sibarita JB, Bornens M. Anisotropy of cell adhesive microenvironment governs cell internal organization and orientation of polarity. Proc Natl Acad Sci USA. 2006;10:19771–19776. doi: 10.1073/pnas.0609267103. [DOI] [PMC free article] [PubMed] [Google Scholar]

RESOURCES