Correlation of cell membrane dynamics and cell motility
© Veronika et al; licensee BioMed Central Ltd. 2011
Published: 30 November 2011
Essential events of cell development and homeostasis are revealed by the associated changes of cell morphology and therefore have been widely used as a key indicator of physiological states and molecular pathways affecting various cellular functions via cytoskeleton. Cell motility is a complex phenomenon primarily driven by the actin network, which plays an important role in shaping the morphology of the cells. Most of the morphology based features are approximated from cell periphery but its dynamics have received none to scant attention. We aim to bridge the gap between membrane dynamics and cell states from the perspective of whole cell movement by identifying cell edge patterns and its correlation with cell dynamics.
We present a systematic study to extract, classify, and compare cell dynamics in terms of cell motility and edge activity. Cell motility features extracted by fitting a persistent random walk were used to identify the initial set of cell subpopulations. We propose algorithms to extract edge features along the entire cell periphery such as protrusion and retraction velocity. These constitute a unique set of multivariate time-lapse edge features that are then used to profile subclasses of cell dynamics by unsupervised clustering.
By comparing membrane dynamic patterns exhibited by each subclass of cells, correlated trends of edge and cell movements were identified. Our findings are consistent with published literature and we also identified that motility patterns are influenced by edge features from initial time points compared to later sampling intervals.
Cellular populations exhibit phenotypic heterogeneity across various physiological and pathological processes. The causative factors range from biological noise to complex distinct states of cell functions. Different approaches have been reported to study cellular heterogeneity from different fronts. Morphological responses to perturbations in cellular environments have been characterized by patterns of signaling marker colocalization from high content images . Cellular heterogeneity through FACS (fluorescence activated cell sorting) has been captured to provide a large number of cell read outs, but without any spatial information . Earlier studies have profiled cell subpopulations from fluorescent images by computing dynamic features of the cells along with static features by using unsupervised clustering . Cellular morphology is a highly dynamic entity and time-lapse high-content imaging of cells provides an unprecedented opportunity to understand the mechanisms of morphodynamics. Morphodynamics is defined as a correlation of cell morphology and the underlying functional activity with respect to time . This concept has enabled the discovery of functionality of specific biomolecules and demanded new techniques for interpretability, accuracy, and speed. Extensive research has been performed in understanding and application of morphodynamics of cell edges. High throughput analysis of cell morphodynamics has been used to discover functions of specific proteins . A series of studies using quantitative fluorescent speckle microscopy have revealed the power of computer assisted high throughput analysis of time-lapse microscopy images: an analysis of the number of speckles suggested distinct regulation of actin polymerization-depolymerization dynamics in different intracellular regions [6, 7]. The ratio of protrusive to inactive cell perimeter has been used as the measure of cell edge activity . Difference of the cell membrane boundary was reported in the study of cell spread dynamics  and its role in actin transport for protruding lamellipodia , formation of filopodia downstream of SCAR (Suppressor of cAMP receptor) , and the role of cofilin as a promoter of actin polymerization leading to protrusion . Alternatively protrusion rates are measured at multiple locations of the cell boundary. The morphological changes have been studied by placing markers in the cell boundary at regular intervals and tracking their displacement in orthogonal directions to the cell boundary . Instead of direct displacement of tracking, cell boundaries can be analyzed with kymographs . This technique involves high resolution time-lapse microscopy to capture subcellular motion which is widely used for relatively small sample sizes due to highly magnified imaging and for relatively short periods of time. However, these approaches are not suitable for high throughput applications due to computational complexity compounded by elaborate cell shapes and its ever changing dynamics.
In this work, we propose novel morphodynamics concepts to quantify the relationship between whole cell movement and edge dynamics. Whole cell movement as a function of space and time and its possible influence on protrusion retraction dynamics have not been studied in detail. Heterogeneous populations exhibiting characteristic protrusion and retraction patterns have been completely exploited by us in order to identify possible correlations with motility features. Such information is helpful in determining overall motility functions of cells in collective migration. Cell membrane movements are extracted and protrusion/retraction dynamics along the cell edges at different time points were obtained to correlate with whole cell motility features. An approach to extract such patterns from heterogeneous cell populations is presented. Our experiments show that the cells with similar kinetic profiles display different edge movements and that features observed in initial time points have profound influence in determining the type of motility patterns as the cell adapts to its motion.
Results and discussion
Cells used in this experiment were mouse macrophage cell lines IC-21 (American Type Culture Collection (ATCC) TIB-186) treated with solvent DMSO (Dimethyl sulphoxide). Cells were observed over a period of 120 minutes and 12-bit images with 0.5 µ m2 pixels were collected using Cellomics KineticScan at every 10 minutes giving a total of 12 snapshots. Data and statistical analysis were implemented in MATLAB R2008a (The Mathworks, Inc., USA) and R project .
Cell identification and tracking
Cells are bright objects protruding from a relatively uniform dark background in microscopic images. The purpose of segmentation is to identify cells accurately in an automated manner. Segmentation algorithms cluster image pixels based on their features into two groups representing objects of interest and background. Simple methods like thresholding do not work because they are not robust to noise and artifacts of images as well as images with overlapping cells . Methods such as region growing, watershed, clustering and active contours have been attempted on cellular images [17–19]. However, these methods fail on images composed of overlapping or clustered cells. Cell segmentation is crucial to this work since tracking and subsequent analyses depend on the segmentation results. In our analyses, active contour without edges was used since it is not dependent on initialization, noise and boundary leakage by using intensity gradients [20, 21]. The energy functional for regularization term is controlled by the length terms only and it was set according to the resolution of fluorescence intensity. The two phase level-set method is able to identify cells with maximum shape information since it handles sharp corners and cusps of the objects well. Thus, the original shapes of cells are retained yielding accurate features. Since dynamics of cell is dependent on geometric centroid, cell shape has to be accurately segmented. We subjectively evaluated segmentation results from different methods and confirmed that slight changes in the methods could dislodge the cell boundary by several pixels but did not affect the global boundary movement. Since we used run length of the boundary, minor boundary displacement did not affect the overall downstream analysis.
The spatiotemporal tracking method does not assume overlapping of cell boundaries between adjacent frames. It is able to handle dividing cells by using a set of heuristics. Four different scenarios are encountered during matching: (i) a cell in the current frame could match a cell in the proceeding frame (a successful match), (ii) no matching for cells in both frames (cells moving out of focus), (iii) one cell in current frame matches with more than one cell in the proceeding frame (possible cell division), and finally (iv) more than one cell in the current frame matches with only one cell in the proceeding frame (over segmentation). For differentiating case (iii) and case (iv), matching between second and third frame are checked to see whether a cell in multiple matches has its own unique characteristics. If a cell matches its counterparts in second and third frame, then we conclude that this cell has divided in the middle frame. If it has only one match in the third frame, then we conclude that this cell might have been over-segmented in the second frame. We used the same settings for weights as suggested by authors .
Classification of cell features
Chi-square goodness-of-fit for dynamic features
Total path length
Random motility coefficient
Mean path length
Feature values of individual clusters
Total path length (µm)
Total displacement (µm)
Random motility coefficient
Mean path length (µm)
Persistence length (µm)
Classification of edge features
Correlation of cell and edge features
Leave-one-out cross-validation of correlation (mean ± std.dev) ×10–4
55.78 ± 2.93
380.18 ± 3.20
–263.96 ± 0.40
–5.33 ± 0.45
–31.24 ± 1.98
377.82 ± 3.08
95.82 ± 4.64
418.90 ± 3.66
15.80 ± 2.52
426.21 ± 2.11
–88.32 ± 2.22
285.68 ± 1.60
–168.86 ± 1.33
169.92 ± 3.51
–169.26 ± 1.33
169.84 ± 3.51
182.02 ± 1.15
346.26 ± 0.80
329.58 ± 0.65
498.53 ± 0.82
411.92 ± 3.50
489.53 ± 1.28
84.41 ± 0.02
299.57 ± 0.48
427.14 ± 1.57
602.57 ± 0.80
242.12 ± 2.11
413.72 ± 2.78
280.53 ± 1.41
452.87 ± 1.82
280.41 ± 1.41
452.79 ± 1.82
149.57 ± 2.49
127.09 ± 1.82
736.33 ± 0.53
119.63 ± 0.96
599.60 ± 0.12
105.90 ± 0.28
148.64 ± 1.62
123.50 ± 1.37
114.32 ± 0.63
142.40 ± 1.23
137.66 ± 1.47
159.82 ± 1.26
116.66 ± 3.48
148.01 ± 2.52
116.66 ± 3.48
148.01 ± 2.52
776.22 ± 0.01
481.55 ± 0.36
655.13 ± 0.61
504.45 ± 0.44
595.36 ± 0.40
360.12 ± 0.55
828.03 ± 1.53
539.72 ± 0.20
872.38 ± 0.40
562.92 ± 0.40
808.41 ± 0.43
506.85 ± 0.31
796.51 ± 0.19
503.66 ± 0.23
776.57 ± 0.19
503.71 ± 0.23
Class 1: This class consists of cells with low speed and persistence. The pattern shows that active membrane ruffling may not translate into active cell movement. It might have even restricted the cells overall movement which is evident from the low total displacement feature. For example, NRK49F cells with defect in rho or adducin have been shown to have active lamellopodial ruffling, while being unable to migrate  (Fig. 2a and 2b).
Class 2: Cells with medium speed and persistence showing positive correlation for protrusion and retraction. Similar protrusion and high retraction activity may be the reason for multiple peaks of edge features over the length of time (Fig. 2c and 2d).
Class 3: This class is represented by fast moving cells displaying high speed and persistence and is positively correlated with edge movement features. These cells also had the highest edge activity which may help in moving the cell over long distances with high persistence. When the static features of these cells were analyzed they had typical fan shaped morphology (Fig. 2e and 2f).
Class 4: These cells frequently change directionality as indicated by low persistence. Edge features are also positively correlated to dynamic features and this suggests that the frequent change in direction may be accompanied by a respective change in edge movements. Although the cells change direction, they travel within a limited radius more like in spiral motion. This can be seen from the low total displacement and mean path length compared to class 3. Even though, the cell speed is greater than Class 3, the cells do not travel in a constant direction (as indicated by low persistence) and tend to display a spiral or circular concentric motion (Fig. 2g and 2h).
Factor analysis on cell and edge features
Factor name (number)
Initial edge features (1)
Motility features (2)
Intermediate/late edge features (3)
Late retractions (4)
Intermediate retractions (5)
Intermediate retractions (6)
Late protrusions (7)
Factor loading matrix computed from covariance matrix for all classes
Non-genetic heterogeneity in cell populations arises from a combination of intrinsic and extrinsic factors. This heterogeneity has been measured for gene transcription, phosphorylation, cell morphology, drug perturbations, and used to explain various aspects of cellular physiology. Our understanding of individual players in cell migration process is increasing; but there remains a vital gap to be filled concerning how they are coordinated spatially and temporally. New techniques are needed which can quantify dynamic cell movements at the level of single cell resolution in an automated manner.
Here, we report multivariate analysis of different sets of motility features through a meaningful combination of both novel (edge) and existing (centroid based) dynamic features. The first set of measurements has been already proved to improve subpopulation analysis. The second set of features is a novel measurement of edge activity. These features capture pixel movement, either through protrusion or retraction frame by frame over the entire length of observation. Since these measurements are temporally sampled, it is suitable to study cell activity over time. These features are unique and not necessarily a measurement of cell migration, as membrane protrusion-retraction is possible without translocation. Our data indicate different levels of correlation between sets of features, depending on the dynamic classes they belong to. This type of relationship was expected for this cell line due to its highly motile nature. Our findings compare well with previous literature .
The introduction of edge features is the major contribution of this work since it captures edge activity of large number of cells from high throughput imaging platforms in a way that no other profiling methods we are aware of have previously demonstrated. Our profiling method was able to provide additional insights which might have been missed using population based cell migration techniques or classical motility assays. To conclude, we have identified heterogeneous edge patterns of related dynamic profiles and validated our correlation patterns by comparing with previous publications. The dynamic profiles were obtained from cell displacement data by GMM clustering. Edge prints from these subclasses were further used to characterize heterogeneity arising due to different edge movements. The patterns arising from statistical correlation analysis were validated by comparing with previous publications. We also provided statistical evidence that initial time point edge features influence the motility patterns that a cell adapts.
Segmentation and tracking of cells
Level-set was used to segment cells from images, independently at all the time frames . The image gradient was used to stop the evolution of level-sets. Touching cells were further separated by a marker-controlled watershed that uses initially segmented cells as shape markers for marking function . The segmented cells in adjacent frames were correspondingly matched by spatiotemporal matching scheme that uses features like size, intensity, and spatial coordinates for matching . The tracks of cells were subsequently corrected for mismatches and only those cells moving for the entire period of observation were included for further analysis.
Dynamic feature extraction
Dynamic features of cells are classified into two categories based on motility modes: features describing whole cell dynamics and features representing membrane (edge) dynamics. Two different methods were employed to extract the two sets of features.
where is a set of N samples and x i is the i th sample comprising of n features, are the mixture weights, and are component Gaussian densities. Each class density is a n-variate Gaussian function. The mixture weights satisfy the constraints that . The complete Gaussian mixture model is parameterized by the mean vector, covariance matrices and mixture weights from all component densities. These parameters are collectively represented as where (µ k , Σ k ) denotes the mean and covariance of the k th component.
Cells are classified by a set of protrusion and retraction features measured over all the time points. These features provide an idea about the activity level of a cell at respective time instances and are used to cluster the cells. Clustering was performed using K-means algorithm.
This article has been published as part of BMC Bioinformatics Volume 12 Supplement 13, 2011: Tenth International Conference on Bioinformatics – First ISCB Asia Joint Conference 2011 (InCoB/ISCB-Asia 2011): Bioinformatics. The full contents of the supplement are available online at http://www.biomedcentral.com/1471-2105/12?issue=S13.
- Slack DM, Martinez DE, Lani WF, Altschuler JS: Charaterizing heterogeneous cellular responses to pertubations. PNAS 2008, 105(49):19306–19311. 10.1073/pnas.0807038105PubMed CentralView ArticlePubMedGoogle Scholar
- Sachs K, Perez O, Pe’er D, Lauffenburger DA, Nolan GP: Causal protein signaling networks derived from multiparameter single cell data. Science 2005, 308: 523–529. 10.1126/science.1105809View ArticlePubMedGoogle Scholar
- Veronika M, Evans J, Matsudaira P, Welsch R, Rajapakse J: Sub-population analysis based on temporal features of high content images. BMC Bioinformatics 2009, 10: S4.PubMed CentralView ArticlePubMedGoogle Scholar
- Dieterich P, Odenthal-Schnittler M, Mrowietz C, Kramer M, Sasse L, Oberleithner H, Schnittler HJ: Quantitative morphodynamics of endothelial cells within confluent cultures in response to fluid shear stress. Biophysical Journal 2000, 79(3):1285–1297. 10.1016/S0006-3495(00)76382-XPubMed CentralView ArticlePubMedGoogle Scholar
- Bakal C, Aach J, Church G, Perrimon N: Quantitative morphological signatures define local signalling networks regulating cell morphology. Science 2007, 316: 1753–1756. 10.1126/science.1140324View ArticlePubMedGoogle Scholar
- Vallotton P, Gupton S, Waterman-Storer C, Danuser G: Simultaneous mapping of filamentous actin flow and turnover in migrating cells by quantitative fluorescent speckle microscopy. Proc Natl Acad Sci USA 2004, 101: 9660–9665. 10.1073/pnas.0300552101PubMed CentralView ArticlePubMedGoogle Scholar
- Ponti A, Machacek M, Gupton S, Waterman-Storer C, Danuser G: Two distinct actin networks drive the protrusion of migrating cells. Science 2004, 305: 1782–1786. 10.1126/science.1100533View ArticlePubMedGoogle Scholar
- Waterman-Storer C, Worthylake R, Liu B, Burridge K, Salmon E: Microtubule growth activates Rac1 to promote lamellipodial protrusions. Nature Cell Biology 1999, 1: 45–50. 10.1038/9018View ArticlePubMedGoogle Scholar
- Dunn G, Zicha D: Dynamics of fibroblast spreading. Journal of Cell Science 1995, 108: 1239–1249.PubMedGoogle Scholar
- Zicha D, Dobbie IM, Holt MR, Monypenny J, Soong DYH, Gray C, Dunn GA: Rapid actin transport during cell protrusion. Science 2003, 300: 142–145. 10.1126/science.1082026View ArticlePubMedGoogle Scholar
- Biyasheva A, Svitkina T, Kunda P, Baum B, Borisy G: Cascade pathway of filopodia formation downstream of SCAR. Journal of Cell Science 2004, 117: 837–848. 10.1242/jcs.00921View ArticlePubMedGoogle Scholar
- Ghosh M, Song X, Mouneimne G, Sidani M, Lawrence D, Condeelis J: Cofilin promotes actin polymerization and defines the direction of cell motility. Science 2004, 304(5671):743–746. 10.1126/science.1094561View ArticlePubMedGoogle Scholar
- Machacek M, Danuser G: Morphodynamic profiling of protrusion phenotypes. Biophysical Journal 2006, 90(4):1439–1452. 10.1529/biophysj.105.070383PubMed CentralView ArticlePubMedGoogle Scholar
- Woo S, Gomez T: Rac1 and RhoA promote neurite outgrowth through formation and stabilization of growth cone point contacts. Journal of Neuroscience 2006, 26: 1418–1428. 10.1523/JNEUROSCI.4209-05.2006View ArticlePubMedGoogle Scholar
- R Development Core Team: R: a language and environment for statistical computing. In R Foundation for Statistical Computing. Vienna, Austria; 2009. . ISBN 3–900051–07–0 http://www.R-project.org . ISBN 3-900051-07-0Google Scholar
- Otsu N: A threshold selection method from gray-level histograms. IEEE Transactions on Systems, Man, and Cybernetics 1979, 9: 62–66.View ArticleGoogle Scholar
- Gonzalez R, Woods R: Digital Image Processing. In IEEE Transactions on Systems, Man, and Cybernetics. New Jersey, USA: Prentice Hall; 2003.Google Scholar
- Vincent L, Soille P: Watersheds in digital spaces: and efficient algorithm based on immersion simulations. IEEE Transactions on Pattern Analysis and Machine Intelligence 1991, 13(6):583–598. 10.1109/34.87344View ArticleGoogle Scholar
- Kass M, Witkin A, Terzopoulos D: Snakes: active contour models. International Journal of Computer Vision 1997, 1(4):321–331.View ArticleGoogle Scholar
- Chan T, Vese L: Active contours without edges. IEEE Transactions on Image Processing 2001, 10(2):266–277. 10.1109/83.902291View ArticlePubMedGoogle Scholar
- Mark N, Albert A: Feature extraction and image processing. Oxford, UK: Academic press; 2008.Google Scholar
- Rajapakse J, Veronika M, Cheng J: Spatiotemporal cell profiling for cell phase identification. Submitted to BioinformaticsGoogle Scholar
- Dove A: Membrane specilization. Journal of Cell Biology 1999., 145(2):Google Scholar
- Heckman AC, Jamasbi JR: Describing cell shape dynamics in transformed cells through latent factors. Experimental Cell Research 1999, 246: 69–82. 10.1006/excr.1998.4242View ArticlePubMedGoogle Scholar
- Kaiser HF: The application of electronic computers to factor analysis. Educational and Psychological Measurement 1960, 20: 141–151. 10.1177/001316446002000116View ArticleGoogle Scholar
- Cheng J, Rajapakse J: Segmentation of clustered nuclei with shape markers and marking function. IEEE Transactions on Biomedical Engineering 2009, 56(3):741–748.View ArticlePubMedGoogle Scholar
- Zygourakis K: Quantification and regulation of cell migration. Tissue Engineering 1996, 2: 1–16. 10.1089/ten.1996.2.1View ArticlePubMedGoogle Scholar
- Rissanen J: A universal prior for integers and estimation by minimum description length. Annals of Statistics 1983, 11(2):417–431.View ArticleGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.