Experimental
COMPUTER ANALYSIS
Cell Research 56 (1969) 363-368
OF ANIMAL
CELL MOVEMENT
IN VITRO
G. BARSKI, J. W. BUTLER and R. J. THOMAS Institut Gustave-Roussy, 94 Villejuif and CNRS, France, Argonne National Laboratory, Argonne, Ill. 60439, and De Paw University, Greencastle, Ind. 46135, USA
SUMMARY Time lapse photographs of living cells in vitro have been analyzed by an experimental combined film scanner and computer system to derive quantitative indices of cell motility associated with motions other than simple translation and rotation. Results of preliminary runs show general agreement with subjective impressions of cell activity.
Motility was one of the first properties noticed in animal cells since the time of Leeuwenhoek [lo] and Metchnikow [13]. The pioneers of tissue culture, Harrison [9], the Lewises [l 11,and later Earle and his group [12], described the kinetic activity of animal cells in vitro and made the first attempts to produce quantitative data on the speed of cell movement outside the organism. These attempts remained fragmentary, however, and consisted essentially in measuring in time the length of cell paths recorded by direct observation and drawing or by time lapse photography. The information gained by this elementary approach is quite inadequate. First, for animal cells explanted in vitro the rate of directional change is usually high and the linear segments of their paths short. Consequently, the estimations of their displacements, made by the crude methods used so far, depart considerably from real values with the increase of randomness of their movement. Secondly, and in a more general way, we have to consider that cell motility is a complex phenomenon and that the bulk of cinetic animal cell activity in vitro consists in movement of cell membranes (ruffled membrane activity) and of cytoplasmic processes that do not result necessarily in linear and directional
displacement of the whole cell or its center of gravity. Moreover, this kind of nondirectional activity appears as an essential characteristic for a given cell type and corresponds evidently to its specific energetic balance in given physiologic and environmental conditions. Thus, for instance, macrophages display a quite intense activity of their cytoplasm and its processesthough the displacement of whole cells may be quite limited. This is true not only for macrophages but also for many types of cells especially of connective tissue origin. In many casesthe activity of cell membranes, the character of their movement, their proper speed and rhythm appear as an essential, heritable characteristic for a given cell type. Different cell lines developed by Sanford et al. [15] from subcutaneous tissue of C3H mice as well as cell lines developed in our laboratory from lung tissue of C57Bl mice [3] we studied in detail represent a good example of diversity not only of a static morphology but also of kinetic behaviour in vitro when individual cells or cell associations are observed. Some of these differences, approached so far in a purely descriptive way, seem to be related to presence or absence of invasive properties. Many aspects of these differences in cell activity, which is basically Exptl Cell Res 56
364 G. Bavski et al. random and nondirectional, are entirely overlooked if only the length of cell paths are measured and compared. Similarly, the complex phenomena known under the general denomination of “contact inhibition” [l] can be only partially and inadequately approached when analysis is limited to whole cell localization and displacement. In the light of these facts, it is becoming obvious at the present stage of cell culture studies that a more elaborate, accurate and versatile system of analysis of cell motility is needed to quantitate the different facets of this phenomenon in order to check and express in a mathematical language such particularities of kinetic cell activity as ruffling of cell membranes, movement of cytoplasmic processes, activity of pinocytic vacuoles, rotation of nuclei, “bubbling” of the cytoplasm during mitosis, or cytoplasmic reactions at contact between cells in vitro. More specifically, it appears as very important to bring with the aid of quantitative analysis more precision to the notion of contact inhibition observed between normal fibroblastic or epithelial cells in vitro which inhibition is lost or markedly decreased in malignant cells. MATERIAL
AND METHODS
We undertook, with the aid of the previously described 14,81Chloe film scannercombined with a computer system, to analyze time lapse cinematography records picturing living animal cells, of known nature and origin, maintained in well determined conditions of medium, pH, oxygenation and temperature. We used as biological material in our study (1) liver macrophagesmigrating out from fragments of newborn mouse liver; (2) peritoneal macrophagesobtained from washingsof the ueritoneal cavity of adult mice meDared by thr& consec&ve daily in&peritoneal injeitidns of 0.1 % glycogen in saline; (3) fibroblasts from primary adult mousesubcutaneous&sue maintained in deficient medium in a “starved” state. All cultures were prepared on 13 x 32 mm coverslips in flattened tubes and fed with Eagles MEM medium supplementedwith 15% calf serum.After 3 or 4 daysthe cultures were transferred into special perfusedchambers and mounted for time-lapse microcinematography as previously described[2]. Co&ant temperatureof 37°C & 0.2 inside the culture chamberswas maintained with the aid of water Derfusedthermostatic heating stage.We used the Zeisslo;g focus, phasecontrast, optical $stem with objectivemagnification of x 40 and ocular magnification x 1.07. The time interval between frames of the 16 mm Scientia Gevaert film was uniformly 15 set (see fig. 1).
Exptl Cell Res 56
Mathematical interpretation of the time-lapse sequences was accomplished with the computer hardware and program system previously developed for the automatic computation of karyograms from metaphase chromosome photographs [7].
RESULTS The numerical information obtained from a photograph (transparency) by the Chloe machine consists essentially of edge coordinates (in the film plane) of regions on the film which are denser than a certain fixed clipping level. The result is a representation of the image as a union of connected subregions or “shapes” [5] (see fig. 2). In the work described, these shapes were combined into one “object” by the computer programs and the first seven Euclidean moment invariants [6] computed for the composite region. By this process, each cell image is represented by a point in a seven-dimensional measurement space, the position of this point being independent of the location and orientation of the cell image in the film plane. This simply means that the measurements of the cell image made by the Chloe machine are condensed by the computer programs into seven numbers, the methods of calculation ensuring that these numbers are location- and orientation-independent. A statistic of this kind is exactly what is needed for studying the specific non-directional cell motions consisting of temporal changes in shape and size without regard to linear and rotational displacements. Since each instantaneous cell configuration is represented by a point in the measurement space, the cell motion maps into a continuous trajectory in this space; the individual time lapse photographs correspond to points spaced along the trajectory. A rather obvious strategy is then to use as a measure of cell motility an estimate of the length of trajectory swept out in a given time. To define the length of a cell trajectory, a metric function must be defined on the measurement space. The choice of metric is fundamental to the whole process, since the properties of this function ultimately determine what type of motions are measured by the system. To get the
Computer analysis of cell movement
365
Fig. I. A-A’ “560” a liver macrophage, B-B’ “540” a peritoneal macrophage, and C-C’ “554” a “starved” fibroblast. For each pair 10 min elapsed between the two time-lapse photographs reproduced. Subjectively, high motility of macrophages and low motility of the fibroblast can be appreciated. Phase contrast, magnif. x 963 for A-A’, B-B’ and x 385 for C-C’. Exptl Cell Res 56
366
G. Barski et al.
Fig. 2. Views of three types of cells used in cell motility analysis, together with corresponding reconstituted images from data in computer memory. Identification numbers “MO”, “540” and “554” are explained in the text.
general feel of the data, we simply tried several different metric functions which seemed intuitively reasonable and were easy to compute. The numerical results reported are based on the metric function which is the sum of seven terms of the form Y3, m*(x, V) = Cxi - Y3”/Cxf+ Exptl
Cell Res 56
where x = (x1, x2, ...) and y = (y,, yz, . ..) are two arbitrary points. Agreeably, qualitative relationships between the calculated quantities proved not to depend on the choice of metric, at least at this preliminary stage of the analysis. The next step is the definition of the “chord space” of the trajectory, which is the space
Computer analysis of cell movement
spanned by the chord vectors extending from one measured trajectory point to the next (in time sequence). The coordinates of the point corresponding to a given chord are thus the seven numbers di
= mi(Jn+l,
367
Table 1. Measurements of relative motility of an inert cell (no. 554) and three moving cells Cell
D
no.
554
560
540
540 P
0.19
1.43
0.94
0.88
Jd,
in which J,, represents the nth point along the trajectory in time sequence. Finally, the total length of the trajectory is then estimated as the sum of all of these quantities over all the time steps in a given photographic sequence. The reason for introducing the chord space was to make it possible to filter out apparent cell motions actually due to the measuring process itself. This was accomplished by using the “starved” fibroblast sequence as a relatively inert control and computing a discriminant plane in the chord space to discriminate against motions characteristic of the inert cell. Relative to the inert fibroblast control sequence (identified as no. 554), we used these techniques to estimate motility indices for three other time-lapse sequences;one film of the liver macrophage (no. 560) and two sets of the peritoneal macrophage photographs (nos. 540 and 540P). The number of images in each sequence varied from 200 to 420. Fig. 1 shows cells of these three types as they appear to the camera and fig. 2 as they appear to the camera and to the computing machinery. The small equilateral triangles seen in the computer-reproduced pictures point to (right vertex) the centroids (centers of mass) of the cell images as computed by the programs. Table 1 presents the results of this preliminary measurement. What is actually given is the mean distance per time step (6) instead of the total, making it easier to compare runs with different numbers of images. These values generally agree with subjective estimates of the cell activities, except that the liver macrophage (no. 560) appears to yield too high a value of B. This appears to result from the greater amount of fine scale motion characteristic of this cell type, at least in our particular preparation. Because of the form of the metric function used, these fine scale movements make
a greater impression on the computer program than they make on the eye of an observer. DISCUSSlON
The here reported procedure of quantitative computer analysis of animal cell movement in vitro based on scanning of time lapse phase contrast film sequences is, as we believe, a prospectful approach to the quantitation of animal cell kinetic activity in vitro. What is remarkable in this system is that it permits to obtain quantitative data concerning nonlinear and nondirectional cell motility which was, sofar, hardly accessible to quantitative evaluation excepting some attempts made in this field by Pomerat et al. [14] chiefly on pulsating oligodendrocytes. One of the advantages of the described procedure is its versatility, since the basic information stored following the primary scanning analysis can be used according to different programs. The moving objects inside the cell can be selected or discarded according to their size and one can envisage to analyse movement of different cell organellas such as nuclei, membranes, vacuoli, or granular structures. On the other hand, as was done in the here reported preliminary results of comparative kinetic activity of macrophages and fibroblasts, an integrated global activity measure of cells can be calculated and expressedas an index. As a next step the spectrum of cell motility for different types of cells can be studied. On the basis of information concerning the average standard motility of a known cell type, it may become possible to evaluate in a satisfactory fashion the influence on this essential cell function of such factors as temperature, Exptl Cell Res 56
368 G. Barski et al. oxygen tension, pH or action of different chemical components or drugs introduced into the medium. Consequently, a well defined and working system of quantitation of cell movement in vitro may supply a remarkable tool for pharcodynamic studies on the cellular level. Work performed in part under the auspices of USAEC. The comnetent assistance of Mlle Bernadette Leon in the time-lapse cinematography involved in this work is gratefully acknowledged.
REFERENCES 1. Abercrombie, M, Heaysman, J E M & Karthauser, H M, Exptl cell res 13 (1957) 276. 2. Barski, G & Belehradek Jr, J, Exptl cell res 37 (1965) 464. 3. Barski,.G, Billardon, C, Jullien, P N & Carswell, E, Intern J cancer 1 (1966) 541.
Exptl Cell Res 56
z. Butler, J W, J data management 3 (1965) 32. Butler, J W, Butler, M K & Stroud, A, Data acquisi’ tion and processing in biology and medicine (ed. K Enslein) vol. 5. Pergamon, Oxford (1968).
6. - Ibid vol. 3. (1964); vol. 4 (1966). , Butler, J W, Butler, M K & Marczynska, B, Ann N Y acad sci. In press. 8. Clark, R & Miller, W F, Methods in computational physics vol. 5, p. 47. Academic Press, New York (1966). 9. Harrison, R G, J exptl zoo1 9 (1911) 787. 10. van Leeuwenhoek, A, Arcana naturae (ed. H A Krooneveld). Delphis Batavorum 1695. 11. Lewis. W H. Bull Johns Hookins hosn 49 (1931) 17. 12. McQuilkin, W T & Earle, W R, J nati cancer i&t 28 (1962) 763. 13. Metchnikow, E, Immunite dans les maladies infectieuses p. 189. Masson, Paris (1901). 14. Pomerat, C M, Rounds, D E & Huff, W, J roy sot 83 (1964) 265. 15 microscop Sanford, K K, Likely, G D & Earle, W R, J natl cancer inst 15 (1954) 215. Received January 2, 1969 Revised version received March 17, 1969