An automatic procedure for evaluation of single cell motility

An Automatic Procedure for Evaluationof Single Cell Motility

Anatoliy Dmytriyev, Vadim Tkach, Olga Rudenko, Elisabeth Bock, and Vladimir Berezin*Protein Laboratory, Institute of Molecular Pathology, Copenhagen University, Copenhagen, Denmark

Received 27 July 2005; Revision Received 30 March 2006; Accepted 12 April 2006

Background: Cell motility is vital in many physiologicaland pathological processes, such as embryogenesis,inflammation, wound healing, and metastasis. However,the time-consuming step in the evaluation of individualcell motility is the analysis of hundreds of recorded imagesof cell cultures in general consisting of retrieving images,one at a time, and marking the positions of individual cellsby a computer mouse. Therefore, the aim of the presentstudy was to develop a novel automatic procedure for theevaluation of cell motility.Materials and Methods: The procedure was tested onfibroblasts and glioma and adenocarcinoma cells engi-neered to express the green fluorescent protein by eithertransient transfection or adenovirus transduction, allow-ing automatic recognition of cell coordinates on retrievedimages.

Results: The effects of serum growth factors, teratogeniccompounds, and overexpression of transcription factorson the motile behavior of cultured cells were determined.Cell motility was estimated by both manual and automaticmarking of cell position and subsequently motility parame-ters were computed. The results obtained by the two pro-cedures were found to correlate significantly.Conclusions: We developed a procedure allowing auto-matic video recording of sparsely seeded cells transfectedwith a plasmid or tranduced with a recombinant virus ex-pressing the enhanced green fluorescent protein (EGFP).q 2006 International Society for Analytical Cytology

Key terms: cell motility; object recognition; fibroblasts;adenocarcinoma cells; Fra-1 expression; valproic acid

Cell motility is vital in many physiological and patholog-ical processes, such as embryogenesis, inflammation,wound healing, and metastasis. For quantitative evaluationsof cell motility two main approaches are used. The firstapproach comprises cell population assays, includingmethods such as the Boyden chamber assay and thein vitro wound healing assay (1,2). In the second ap-proach migration of individual cells is evaluated followingthe analysis of coordinates of cell positions at differenttimes of observation using time-lapse video microscopy.Recent advances in microscopy include the use of compu-terized microscope workstations, in which microscope ta-ble positions, filter wheel changes, autofocusing, and aframe grabber for time-lapse video recording all are con-trolled by specially designed software (3). During the lastdecade, application of this technology has made it possi-ble to study the molecular mechanisms regulating themotile behavior of single cells (4). However, the time-con-suming step in the evaluation of individual cell motility isthe analysis of hundreds of recorded images of cell cul-tures in general consisting of retrieving images, one at atime, and marking the positions of individual cells by acomputer mouse. When the positional coordinates of cellsat different time points are fixed, the software can pro-duce data describing the cell tracks, based on which aseries of cell motility parameters can be determined (5).

Advanced commercially available computer-systems forthe analysis of cell motility and changes in cell shape havebeen developed in recent years. Thus Soll Technologieshas developed an image analysis system based on the useof polarized light microscopy and differential interferencecontrast optics (DIC). We, here, report the developmentof a novel automatic procedure for the evaluation of cellmotility. The procedure allows automatic video recordingof sparsely seeded cells transfected with a plasmid or tran-duced with a recombinant virus expressing the enhancedgreen fluorescent protein (EGFP). The system allows auto-matic recognition of cell contours and determination ofcoordinates of cell positions at any time of observation. Tovalidate the developed procedure, the individual motilityof mouse fibroblastoid L929 cells, rat glioma BT4Cn cells,and mouse adenocarcinoma CSML0 cells, grown in a vari-ety of culture conditions, was estimated by both manualmarking of the positions of cells on sequentially retrievedimages, and automatic marking of the same with the newsoftware. In each individual experiment, 100–200 cells

*Correspondence to: Vladimir Berezin, Protein Laboratory, Panum Insti-

tute, Blegdamsvej 3C, Bld. 6.2, DK-2200 Copenhagen N, Denmark.

E-mail: [email protected] online in Wiley InterScience (www.interscience.wiley.com).

DOI: 10.1002/cyto.a.20303

q 2006 International Society for Analytical Cytology Cytometry Part A 69A:979–985 (2006)

were processed. The results indicate a highly significantcorrelation of various computer motility parametersobtained by the two procedures, supporting the validityof the automatic procedure.

MATERIALS AND METHODSExpression Plasmids

To generate the pCMV-Fra-1 expression plasmid, mRNAswere amplified by reverse transcription-PCR as previouslydescribed (6), and cloned into the PCR3-Uni vector (Invi-trogen, Groningen, The Netherlands). The pEGFP-N1 con-struct expressing EGFP was purchased from Clontech(Palo Alto, CA). A recombinant adenovirus, whose expres-sion is under control of a CMV promoter (Ad5.CMV-GFP),was obtained from Qbiogene (Illkirch Cedex, France). Thevirus was amplified in HEK293A cells. The titer of the puri-fied recombinant virus was estimated by the Adeno-XTM

Rapid Titer Kit (Clontech).

Cell Cultures

The L929 cell line was obtained from the European Col-lection of Animal Cell Cultures. The CSML0 cell line origi-nated from a spontaneous mammary adenocarcinoma inan A/Sn mouse (7). The highly invasive rat glioma cell lineBT4Cn was obtained from O. Laerum, University of Ber-gen (8). Cells were grown in Dulbecco’s modified Eagle’smedium (DMEM) supplemented with 10% (v/v) heat inac-tivated fetal calf serum (FCS), penicillin (100 U/ml), andstreptomycin (100 lg/ml) (Gibco BRL, Gaithersburg, MD).

Transfection and Transduction

Transient transfection of cells was performed using theTargefect F-2 reagent according to the manufacturer’sinstruction (Targeting Systems, Santee, CA). SubconfluentL929 cells were infected with Ad5.CMV-GFP virus accord-ing to the manufacturer’s instruction (Clontech). 24 hafter transfection, cells were seeded at a density of 4 3 103

cells/cm2 and grown for 24 h.

Video Recording

The workstation employed comprised an Eclipse TE300inverted microscope (Nikon, Japan; objective Nikon PlanFluor 103/0.30 Ph1 DL) equipped with a heated, mova-ble, computer controlled microscope stage (3). Digitalmicrographs of cells were obtained using a black andwhite CCD video camera (DFA, Denmark) and a framegrabber coupled to a conventional PC. Recordings wereperformed using the PRIGRA software (developed at theProtein Laboratory, Copenhagen, Denmark). Images werestored as 768 3 576 pixels, eight-bit, grayscale GIF image-files.

Computation of Motility Parameters

When making video recording of cells for motility stu-dies, the observations can be performed either continu-ously or (as in our study) at selected time intervals. In thefollowing, the time interval (of uniform duration) between

discrete observations is denoted as s. For example, duringa given recording time (total time of observation, ttot),positions of cells are observed at time-points t 5 0, t 5 s,t5 2s, t 5 3s etc.Cell displacement. The displacement of a cell during

a given time period can be calculated in two differentways, here denoted as the ‘‘simple’’ and the ‘‘complexoverlapping’’ methods. In the simple method, the displa-cement of a cell is calculated as the displacement from itsposition at the initiation of the recording to its position atlater time-points. In the following, such a time-point willbe referred to as the ‘‘time of observation’’, tobs. Thus,when using the simple method of calculation the cellulardisplacement is measured from t 5 0 to t 5 tobs. The term‘‘complex’’ refers to the situation where the calculated dis-placement is an average value of displacement occurringin several, different time intervals of identical duration. Inthe following, these time intervals will be referred to asthe ‘‘time interval of interest’’, ti. When using the com-plex, overlapping method the displacement at ti 5 4s willbe the average displacement observed for all time intervalsof the duration t 5 4s (i.e. from t 5 0 to t 5 4s, t 5 s tot 5 5s, t 5 2s to t 5 6s and so on). To compute cell motil-ity parameters we used the complex method of calcula-tion because it allowed the reduction of random noise.Mean-cell-speed (St). The speed of a cell can be com-

puted by dividing displacement, Dd, of the cell by thetime interval, Dt, within which the displacement tookplace.

Ss ¼ DdDt

ð1Þ

The dispersion of a cell is simply the Euclidean distancebetween two points on a plane (measured in lm). Forexample, the dispersion of a single cell after a given timeof observation is calculated as

dobs ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiðxðtobsÞ � xðt0ÞÞ2 þ ðyðtobsÞ � yðt0ÞÞ2

qð2Þ

where x(Æ) and y(Æ) are the x- and y-coordinates of the cell.The average speed of a sample of a population of cells

calculated for a given time of observation can be calcu-lated using the complex method with overlapping inter-vals [eq. (3)].

St i ¼1

N

1

ttot=s� ti=sþ 1

3XN

k¼1 t¼ti

Xttots

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiðxkðtÞ�xkðt�tiÞÞ2þðykðtÞ�ykðt�tiÞÞ2

qti

ð3Þ

where k denotes a given cell, N is the size of the investi-gated sample of a population of cells, x(Æ) and y(Æ) are thex- and y-coordinates of the cell, ttot is the time elapsedfrom the first to the last image constituting a recording,and s is the time interval between discrete observations.

980 DMYTRIYEV ET AL.

Cytometry Part A DOI 10.1002/cyto.a

In the following, cell speed calculated according toEq. (3) is referred to as ‘‘mean-cell speed.’’

The mean path length, L, for a sample of a populationof cells at a given time of observation, tobs, was calculatedas

L ¼ 1

N

XNk¼1

Xtobss¼1

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiðxkðtsÞ � xkðts�1ÞÞ2 þ ðykðtsÞ � ykðts�1ÞÞ2

q

ð4Þ

The locomotive index, LI, was calculated as the relation-ship LI 5 d/L, and used as a measure of cellular persist-ence in a certain direction (i.e. the tendency of a cell tomove in approximately the same direction for a certainlength of time).

The root-mean-square speed, S, and the persistencetime in direction, P, are estimated after plotting the mean-squared displacement Æd2æ against time by subsequent fit-ting of the curve to the equation

hd2i ¼ 2S2P t � P 1� exp�t

p

� �� ð5Þ

The value Æd2æ can be obtained by tracking the cell posi-tions, as explained by Dickinson and Tranquillo (9).

The curve fit simultaneously permits an estimate of therate of diffusion,

R ¼ 2S2P ð6Þ

Once Æd2æ is available, the values of S, P, and R can be esti-mated by the method proposed by Dunn (10) and revisedby Dickinson and Tranquillo (9), which consists in fittingthe Eq. (5) to the mean-squared displacement points witha generalized nonlinear least-squares regression.

RESULTSAutomatic Object Recognition and Cell Tracking

To identify the contours of cells (object recognition) arecorded image has to be transformed into a binary formusing binary thresholding operation. Cell contours wereobtained using a contour-detection algorithm, and thecentroid coordinates were computed (11–13).Tracks connecting the centroid coordinates of each cell

in all consecutive images were calculated by employing analgorithm based on the following assumption: in a seriesof images from 1 to N after determining of centroid posi-tions of cells on both image x and image x þ 1, the pro-gram will for every centroid from image x try to find thenearest centroid on the consecutive image x þ 1. The pro-gram connects two centroids from images x and x þ 1 ifthe distance between them is less then 39.1 lm (conditionI). 39.1 lm is an empirical coefficient. This limit wasdeduced by analysis of the maximal displacement of 100cells (fibroblasts) in 15 min in the presence of serumwhere we found that 39.1 lm is a maximal distancechange for fibroblasts in 15 min. If a neighbor is not foundthe program will use the following algorithm:

i. Find ‘‘connections’’ between the rest of the cen-troids from images x þ 1 and x þ 2.

ii. If after step (i) condition I is not fulfilled for some ofthe centroids from image x þ 2, the program will tryto connect these centroids with centroids from imagex given that the distance between the centroids is lessthan 78.2 lm (39.1 lm X 2) (condition II).

iii. If condition II is fulfilled the program computes aputative position for centroids on image x þ 1,which is half of the distance between the centroidpositions on images x and x þ 2.

iv. Every centroid on image x, which has no connectionto centroids in either image x þ 1 and image x þ 2,is marked as the end of a cell track. Likewise, every

FIG. 1. Schematic illustration of theautomated image processing and celltrack generation. L929 cells were trans-duced with Ad5.CMV-GFP virus, seededat a density of 4 3 103 cells/cm2 andgrown for 24 h in the presence of se-rum. Microscopic fields were recordedfor 4 h with 15 min intervals. Theupper panel shows images of fluores-cent cells at times 0 (A) and 4 h (B).Contour coordinates of a cell weredetermined by automatic image recog-nition for every 15 min (C) and the mo-tility track was subsequently generated.The cell is shown in green at times 0(A) and 4 h (B).

981AN AUTOMATIC PROCEDURE FOR EVALUATION OF SINGLE CELL MOTILITY


centroid on image x þ 1, which has no connectionto image x (a ‘‘new’’ centroid), is regarded as thestart of a new track.

The developed algorithms were tested on fluorescentimages of cultured cells engineered to express EGFP. Anillustration of the automatic image processing and celltrack generation is shown in Figure 1.

Automatic Estimation of the Effect of Valproic Acidon Serum-Induced Motility of L929 cells

To validate the automated object recognition and celltracking procedure for estimation of cell motility, it wascompared to a conventional procedure in which cell mo-tility was estimated by manually marking the positions ofall cells on each retrieved image taken from each micro-

scopic field at different time points of observation. First,as a model system the mouse fibroblastoid L929 cell linewas used. Since it is known that the presence of growthfactors is crucial for cell motility, the motile behavior ofL929 cells in the presence or absence of 10% serum wasexamined by the two different methods, the manual ver-sus the automatic procedure. From Figure 2A it appearsthat irrespective of the method of evaluation an increaseof approximately 15 times in cell-displacement was foundwhen fibroblasts were grown in the presence of serum incomparison with cultures grown in the absence of serum.We have previously shown that a known human terato-

gen, valproic acid (VPA), effectively inhibits the motile beha-vior of L929 cells in the presence of serum, in a dose-de-pendent manner (14). We therefore tested the effect of vari-ous concentrations of VPA on the motility of L929 cellsusing the two different evaluation methods. Irrespectively

FIG. 2. Effects of serum (A) and var-ious concentrations of VPA (B) on themotile behavior of L929 fibroblastsevaluated by automatic (black sym-bols) and manual cell tracking (redsymbols). Correlation between motil-ity parameters, Ss (C), LI (D), and R(E), determined by the automatic andmanual procedures. L929 cells weretransduced with Ad5.CMV-GFP virus,seeded at a density of 4 3 103 cells/cm2 and grown for 24 h in the ab-sence or presence of serum and vari-ous concentrations of VPA. For eachcondition, 10–12 different micro-scopic fields were recorded for 4 hwith 15-min intervals. The number ofcells in the individual experimentswas from 100 to 150.



of the method used, a similar dose-dependent decrease inthe motile behavior of VPA-treated L929 cells was found(Fig. 2B). Based on the cell-displacement data presented inFigures 2A and 2B, the values of the mean-cell-speed (Ss),the locomotive index (LI) and the rate of diffusion (R) wereestimated and regression lines with 95% confidence limitswere plotted to illustrate the correlation between the twoprocedures for evaluation of cell motility. From Figures 2C,2D, and 2E it appears that values of Ss, LI, and R obtainedemploying the automatic approach correlated positivelywith the values of Ss (r 5 0.991, df 5 4, P < 0.001), LI (r 50.946, df 5 4, P < 0.005), and R (r 5 0.933, df 5 4, P <0.01) obtained by the conventional manual method.

The procedure used in these experiments employedtransformation of L929 cells with a recombinant adenovirusexpressing EGFP, resulting in an almost hundred percent ef-ficiency of transfection. We also tested whether the auto-matic procedure could be used in cases where cells are tran-siently transfected with a plasmid encoding EGFP. The effi-ciency of transfection of L929 cells is usually as low as 10–20%. Therefore, we repeated all the experiments shownabove using transiently transfected L929 cells. Subsequently,a correlation analysis of the motility parameters determinedby the two procedures was performed, and correlation coef-ficients (r) of 0.891 (df 5 11, P < 0.0001), 0.734 (df 5 11,p < 0.005), and 0.806 (df 5 11, p < 0.001) for Ss, LI, and R,were found. Thus, also transient transfection can be used tomake the cells suitable for recognition, but because of theobservation that the fluorescence of the individual EGFPexpressing cells obtained by this procedure is lower thanthat of the in vitro transduced cells, the correlation coeffi-cients are slightly lower.

Automatic Estimation of Serum Induced Motilityof Glioma BT4Cn cells

We also evaluated whether the automatic procedurecould be applied for the estimation of the motile behaviorof other cell types. Rat glioma BT4Cn cells were transi-ently transfected with plasmid cDNA encoding EGFP andthe motile behavior of these cells was analyzed in the ab-sence or presence of serum. From Figure 3 it appears thatirrespectively of the methods used for evaluation of themotility of BT4Cn cells, automatic or manual, addition ofserum to cultures of BT4Cn cells resulted in an increase inthe mean cell speed, Ss, the locomotive index, LI, and therate of diffusion, R.

Automatic Estimation of the Effect of Fra-1Expression on Motility of Adenocarcinoma

CSML0 Cells

We have recently demonstrated that the adenocarci-noma CSML0 cells, which normally do not express Fra-1belonging to the Fos-family members, display a low motil-ity phenotype (6). We therefore tested whether the auto-matic procedure could be applied for the determinationof the motile behavior of cells in which the motility phe-notype was changed by transient expression of the tran-scription factor Fra-1. From Figure 4 it appears that the

expression of Fra-1 markedly increased the rate of diffu-sion (R) of CSML0 cells. Although the absolute values forR determined by the two methods (black bars for controlcells and open bars for Fra-1 transfected cells) differed, asignificant increase of R in Fra-1 transfected cells wasdetected irrespective of the method employed. In con-trast, no significant changes in mean cell speed or locomo-tive index of cells expressing Fra-1, compared with con-trol, were detected employing any of the two methods.

Evaluation of Automatic Estimation of IndividualCell Motility Combining all Obtained Data by

the Two Procedures

To evaluate the correlation between estimates of cellmotility parameters by the automatic and the conventional-

FIG. 3. Effect of serum on the motile behavior of rat glioma BT4Cn cellsevaluated by automatic (black bars) and manual cell tracking (emptybars). BT4Cn cells were transfected with a plasmid encoding EGFP,seeded at a density of 4 3 103 cells/cm2 and grown for 24 h in the ab-sence or presence of serum. For each condition, 10–12 different micro-scopic fields were recorded for 4 h with 15-min intervals. The number ofcells in the individual experiments was from 110 to 140.



approach, all data from the above-described experiments(shown in Figures 2–4) were combined and the correlationbetween the motility parameters determined by the twoprocedures were assessed. A linear correlation was foundwith correlation coefficients of 0.894, 0.789, and 0.754 formean cell speed, locomotive index, and rate of diffusion,respectively (for all of them df5 19, p < 0.0001). Thus, theautomatic method appeared to be an efficient and rapid al-ternative to current time-consuming manual methods forthe determination of motile behavior of cultured cells.

DISCUSSION

Determination of cell motility in vitro is widely used forthe study of the effects of compounds capable of modulat-ing important biological phenomena such as embryogen-

esis, angiogenesis, inflammation, wound healing, and tu-mor invasion (15–18). Despite the fact that cell motility isbeing studied intensively and many methodological aspectshave been systematically addressed (19), the major impedi-ment in using cell motility as a routine end-point biologicalassay is the complexity of the required technique combinedwith the labour-intensive manual processing of thousandsof images obtained by time-lapse video recordings.We have developed an automatic procedure for object

recognition and showed that this procedure can be usedfor routine determination of key cell motility parameters.When comparing a conventional method of manual cellmarking on individual images to the automatic method, ahigh correlation was found. The automatic method is con-siderably faster than the currently employed conventionalmethods. Moreover, it lacks the elements of bias unavoid-able when an operator processes images manually. Thedeveloped method, however, has at least two limitations.One of them is the requirement of cells stained with a flu-orescent dye or transfected with a fluorescent protein.Since a low-power lens has to be used so as to follow thelocomotion of several cells for many hours, the degree offluorescence, i.e. expression levels and transfection effi-ciency of the fluorescent protein, has to be high. The sec-ond limitation is that only sparse cultures of separate cells(e.g. fibroblasts) not forming aggregates or sheets can beused for analysis.In conclusion, we have developed a procedure for auto-

matic analysis of the motile behavior of cultured cellsexpressing a fluorescent marker. The procedure is rapidand very efficient and the obtained results show a highcorrelation to a manual procedure.

LITERATURE CITED1. Boyden S. The chemotactic effect of mixtures of antibody and antigen

on polymorphonuclear leucocytes. J Exp Med 1962;115:453–466.2. Tran POT, Hinman LE, Unger GM, Sammak PJ. A would-induced

[Ca2þ]i increase and its transcriptional activation of immediate earlygenes is important in the regulation of motility. Exp Cell Res 1999;246:319–326.

3. Walmod PS, Berezin A, Gallagher HC, Gravemann U, Belman V,Bacon CL, Nau H, Regan CM, Berezin V, Bock E. Automated in vitroscreening of teratogens. Toxicol Appl Pharmacol 2002;181:1–15.

4. Xie H, Pallero MA, Gupta K, Chang P, Ware MF, Witke W,Kwiatkowski DJ, Lauffenburger DA, Murphy-Ullrich JE, Wells A.EGF receptor regulation of cell motility: EGF induces disassemblyof focal adhesions independently of the motility-associated PLCgsignaling pathway. J Cell Sci 1998;111:615–624.

5. Walmod PS, Hartmann-Petersen R, Berezin A, Prag S, Kiselyov VV,Berezin V, Bock E. Evaluation of individual-cell motility. In: Gavin RH,editor. Methods in Molecular Biology, Vol. 161: Cytoskeleton Methodsand Protocols. Totowa, NJ: Humana Press; 2001. Chapter 6, pp 59–83.

6. Tkach V, Tulchinsky E, Lukanidin E, Vinson C, Bock E, Berezin V. Roleof the Fos family members c-Fos, Fra-1 and Fra-2 in the regulation ofcell motilty. Oncogene 2003;22:5045–5054.

7. Senin VM, Ivanov AM, Afanasjeva AV, Buntsevich AM. New organospe-cific metastatic transplanted tumors of mice and their use for studyinglaser effect on dissemination. Vestnik USSR Acad Med Sci 1984;5:85–91 (in Russian).

8. Laerum OD, Rajewsky MF, Schachner M, Stavrou D, Haglid KG,Haugen A. Phenotypic properties of neuroplastic cell lines developedfrom fetal rat brain cells in culture after exposure to ethylnitrosoureain vivo. Z Krebsforsch Klin Onkol Cancer Res Clin Oncol 1977;89:273–295.

9. Dickinson R, Tranquillo R. Optimal estimation of cell movementindices from the statistical analysis of cell tracking data. Bioeng FoodNat Prod 1993;39:1995–2010.

FIG. 4. Effect of Fra-1 expression on the motile behavior of CSML0 cellsevaluated by automatic (black bars) and manual cell tracking (emptybars). CSML0 cells were transfected with a plasmid encoding the tran-scription factor Fra-1, seeded at a density of 4 3 103 cells/cm2 and grownfor 24 h in the presence of serum. For each condition, 10–12 different mi-croscopic fields were recorded for 4 h with 15-min intervals. The numberof cells in the individual experiments was from 96 to 117.



10. Dunn G. Characterizing a kinesis response: Time averaged measuresof cell speed and directional persistence. Agents Actions Suppl 1983;12:14–33.

11. Pratt WK. Digital Image Processing, 2nd ed. New York: Wiley; 1991.698 pp.

12. Gonzalez RC, Woods RE. Digital Image Processing. New York: Addison-Wesley; 1992. 716 pp.

13. Sonka M, Hlavac V, Boyle R. Image processing, Analysis, and MachineVision, 2nd ed. Pacific Grove: PWS Publishing; 1998. 770 pp.

14. Walmod PS, Foley A, Berezin A, Ellerbeck U, Nau H, Bock E, BerezinV. Cell motility is inhibited by the antiepieleptic compound, valproicacid and it’s teratogenic analogues. Cell Motil Cytoskeleton 1998;40:220–237.

15. Bronner-Fraser M. Neural crest cell migration in the developingembryo. Trends Cell Biol 1993;3:392–397.

16. Hauzenberger D, Klominek J, Holgersson J, Bergstrom SE, Sundqvist KG.Triggering of motile behavior in T lymphocytes via cross-linking of a4 b 1 and a L b 2. J Immunol 1997;158:76–84.

17. Hudson LG, McCawley LJ. Contributions of the epidermal growth factorreceptor to keratinocyte motility. Microsc Res Tech 1998;43:444–455.

18. Pedersen MW, Tkach V, Pedersen N, Berezin V, Poulsen HS. A naturallyoccurring constitutively active variant of the epidermal growth factorreceptor increases cell motility. Int J Cancer 2004;108:643–653.

19. Hartmann-Petersen R, Walmod PS, Berezin A, Berezin V, Bock E. Indi-vidual cell motility studied by time-lapse video recording: Influence ofexperimental conditions. Cytometry 2000;40:260–270.



Documents

An automatic procedure for evaluation of single cell motility