Application of Quantitative Image Analysis to a Mammalian Cell Line Grown on Microcarriers

Marie-Noelle Pons,* Anne Wagner, Herve Vivier, and Annie Marc L aboratoire des Sciences du Genie Chimique CNRS-ENSIC-INPL, BP 451 F-54001 Nancy Cedex, France

Received September 10, 1991,!4ccepted January 1992

Quantitative image analysis has been applied to the moni- toring of cultures of a mammalian cell line on microcarriers. Procedures have been developed to investigate microcar- rier colonization and cluster formation and to determine the eventual modification of cell size during cultivation using scanning electron microscopy (SEM) microphotography. The human kidney tumor cells (TCL 598) on which the pro- cedures were tested underwent a slight size decrease during the development of the first cell layer on the microcarriers. The cluster size and the cell size remained constant during the culture stationary phase. Key words: quantitative image analysis mammalian cells microcarriers culture monitoring

INTRODUCTION

Biomass characterization techniques suitable for use in control and optimization of bioprocesses are still a chal- lenge due in a large part to the diversity of cells and the number of parameters of interest: total cell number, vi- able cell number, cell volume, cell morphology, and population characterization (degree of contamination). Human sight is certainly a very valuable tool to assess these parameters using microscopy. However, the pro- cedure is time consuming and eye tiring and quantifica- tion is somewhat difficult. Automatic image analysis contributes by making direct cell counts more attractive. Techniques of visualization and image analysis are now applied in the fields of chemistry, biochemistry, and medical engineering. They are used for system and ma- terial characterization and for real-time information collection on processes.

In bioprocesses, quantitative image analysis especially enables data to be obtained on biomass: number of mi- croorganisms, size, morphology. The statistical distribu- tions of these parameters can be obtained as well as their kinetics with respect to culture age. It is more and more recognized that the microorganism morphology is affected by culture conditions’ and is related to its abil- ity to excrete desired metabolites?

Image analysis has been attempted on yeast^,^,^ fila- mentous bacteria,’p””2 and ~e1lets.l~ Here it is applied to the monitoring of a culture of anchorage-dependent mammalian cell lines grown on microcarriers and ob- served by scanning electron microscopy (SEM). The purpose is threefold: to monitor the microcarrier colo- nization by cells, to observe eventual changes in cell

* To whom all correspondence should be addressed.

morphology and/or size, and to investigate what happen to microcarriers when the culture is getting old and clusters are appearing. A cluster is an assembly of sev- eral microcarriers connected by cells forming bridges.

The evaluation of the number of cells growing on mi- crocarriers is a tedious process involving generally the withdrawing of samples and the counting of the trypsin- ized cells. The use of a Coulter counter as a biomass probe has been proposed by Miller et al.,’’ but care needs to be taken to avoid shear-induced detachment of the cells by limiting the number of passages through the Coulter probe orifice. The method has not been applied to clusters.

The difficulties which should be overcome in an image analysis procedure come from the quality of the images, the nature of the cells, and the sample preparation for SEM observation:

1. The cells are observed by SEM but images are two- dimensional views of the three-dimensional reality, and the relief effect of SEM is expressed by an uneven illumination of the image. Furthermore, a correction should theoretically be applied to mea- surements to take into account the sphericity of the microcarrier in order to relate surfaces measured on images (two-dimensional projection) to surfaces on spherical surface^.'^

2. The cells are not fully disconnected one from another and cannot be individualized in most occasions. Therefore, the cells on the image do not constitute a set of disconnected objects (in the sense “one cell- one object”) but form one or several connected object zones. A special technique, granulometry by open- ings, much employed in metallurgical science to study alloys, is here used to obtain a size measurement of object zones.4

3. Sample preparation requires dehydration of the cells and the microcarriers. They undergo shrinkage so the size measured on SEM images is not the actual one in culture. However, careful sample preparation pro- duces a uniform shrinkage and does not destroy the cell

MATERIALS AND METHODS

Mammalian cells (from a human kidney tumor line, TCL 598, from Sandoz Ltd., Basel, Switzerland) were

Biotechnology and Bioengineering, Vol. 40, Pp. 187-193 (1992) Q 1992 John Wiley & Sons, Inc. CCC 0006-3592/92/01018F07$0400

grown on microcarriers (Cytodex 3, Pharmacia, Sweden) in a continuous perfusion reactor. Full details about the culture conditions can be found el~ewhere.'~"~ The cells excreted prourokinase (PUK) into the culture medium.

Two types of culture media have been used: DMEM (Gibco-BRL, Cergy-Pontoise, France) and a mixture of 25% Ham F12-75%DMEM with variable (0%-5%) FCS (Gibco-BRL, Cergy-Pontoise, France) and glucose (11- 25 mM) concentrations (Laboratoires Fandre, Nancy, France) and 2 mM glutamine (Gibco-BRL, Cergy- Pontoise, France). Five grams of microcarriers per liter for a 2-L working volume were retained by a spin filter with a 60-pm mesh. The dilution rate was about 1.5 day-'.

Samples were regularly taken. One part of each sample was used for cell counting. Total cell density was evaluated from the enumeration of crystal violet stained nuclei on a hemacytometer. Cell viability was checked by the trypan blue exclusion method after trypsination of the cells from the carriers. The other part of the sample was prepared for examination under SEM (Jeol T330A, beam energy 10 kV) (Jeol, Rueil-Malmaison, France) after dehydration and gold plating. Samples of 2-3 mL are used. After settling, excess liquid was dis- carded and the microcarriers were rinsed several times with a PBS (phosphate buffer saline) solution contain- ing no calcium or magnesium ions. This solution was finally replaced by a 5% (v/v) glutaraldehyde solution in PBS (stock solution at 25% from Merck, Darmstadt, Germany). Fixation, at 4"C, lasted at least 12 h. Before analysis, glutaraldehyde was taken out. Cells and micro- carriers were dehydrated progressively by means of methanol/PBS solutions with increasing alcohol con- tent. The procedure was the following: 10 min in 20% methanol plus 80% PBS without calcium and magne- sium, 10 min in 50% methanol plus 50% PBS without calcium and magnesium, 10 min in 80% methanol plus 20% PBS without calcium and magnesium, 3 X 10 min in 100% methanol. Microcarriers were deposited on copper supports, air dried, and gold plated under vac- uum. During the fixation process, no cell disruption was noticed: The cells remained attached to the micro- carriers and uniform shrinkage could be assumed.

Images resulted from SEM slides (400 ASA, Ekta- chrome, Kodak). They were captured by a Bosch video camera (Chalnicon tube) with 256 grey levels through a professional vision software package (Visilog by Noesis, Jouy-en-Josas, France). The digitized images (512 lines of 512 pixels) were processed by Visilog or locally made software on a SUN 3/110 color workstation (8 Mbytes memory) connected to the laboratory network.

IMAGE ANALYSIS METHOD

The Appendix contains a basic description of the image analysis operations used in this article. More details can be found in references 4 and 7. The first part of the treatment of grey-level image A (Fig. 1) is devoted to

Figure 4. Examples of circles (in grey) inscribable in object zones (in black).

horizontal axis (Fig. 2). In the case of a disc the Feret diameters are equal to the classical geometric diameter.

Convolution by a Laplacian operator is the enhance- ment method selected to highlight the cell edges on image A. The resulting image is smoothed to eliminate the noise and thresholded to obtain the cell edges as objects. Logical addition (see Appendix) with image C enables retention of only the objects within the zone of interest. The noise elimination not being fully effective, a measurement of the area of objects and an area sort- ing to eliminate the smallest elements are then per- formed. Image complementation and logical addition with image C lead to an image containing the cell prints and the microcarrier uncovered surface as a complex object zone (Fig. 3).

To characterize the size of the object zone, the ra- dius R of the largest circle, centered at pixel i of thejth line, included within the object zone limits, is deter- mined for i = 1,. . . ,512 and j = 1,. . . ,512 (Fig. 4). This is done by means of a series of bidimensional mor- phological openings: and a size distribution character- istic of the object zone is obtained. The sphericity correction4 has not been attempted here as it requires population homogeneity which could not be checked due to the small number of available images per culture age.

RESULTS AND DISCUSSION

It is important to notice that the size measurements are biased by the dehydration procedure. However, as no damage was observed on the microcarrier-cell assembly and on the clusters, the size measurements done on SEM images permit a comparison between samples hav- ing undergone the same dehydration procedure.

Four cultures have been investigated: F25, F26, F27, and F28 (medium enriched with amino acids). Figure 5 presents typical kinetic data.I5 A first phase of active pro- liferation lasts for about 350 h. The second phase, after

350 h, is the stationary cellular phase during which the cell concentration on microcarriers is stabilized at 7 X lo6 cells/mL. With 5 g microcarriers/L confluence is reached between 1 X lo6 and 2 X lo6 cells/mL. Cells then start to form multilayers, and bridges between mi- crocarriers appear for cell densities of about 3 X lo6- 4 X lo6 cells/mL, around 250-300 h of cultivation, to form clusters. Cells seem to multiply inside the clusters, and empty microcarriers may appear. This phenome- non is illustrated in Figure 6 with some characteristic SEM images.

Figure 7a represents the variation of the Feret X- diameter of microcarriers during early ages of cultiva- tion, before 300 h. Analysis has been performed also on empty dehydrated microcarriers to know the size dis- tribution: the mean diameter 81.7 pm with a standard

Figure 5. Typical kinetics of perfused culture.

COMMUNICATIONS TO THE EDITOR 189

Figure 6. Typical growth curve with SEM microphotographies describing the main features of colonization kinetics.

deviation of 8.3 pm is in agreement with the size indi- cated by the manufacturer for dry microcarriers. A slight increase of Feret X-diameter is observed once the microcarriers are fully colonized. The average Feret X-diameter is 100 pm at 300 h, from which a dehydrated cell thickness of 10 pm can be deduced, assuming a cell monolayer on microcarriers.

Figure 7b represents the variations of the Feret X-diameter of clusters. The ratio Feret X-diameter/ Feret Y-diameter is close to 1, and the mean value shape factor 6, defined as the Feret X-diameter times the

Feret Y-diameter divided by the area, is 1.4. For a disc of diameter d, 6 = 4d '/md * = 4/m = 1.27. The average cluster diameter during the stationary phase is 350 pm. It is however difficult to deduce the number of micro- carriers involved in the clusters as the images are pro- jection on a plane of a three-dimensional body and because the number of cell layers is unknown.

The analysis of object zones size by a series of open- ings produces two types of distributions in terms of the radius R of the largest inscribable circle (Fig. 8). For early culture ages, when the microcarriers are not fully

190 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 40, NO. 1, JUNE 5, 1992

0

300 t X

4 X I

100 HS38 mean

0 1 ; ; ; . ; ; . . 0 ux) 400 600 8001ooo1m14001600

Culaue age Q

(b)

Figure 7. (a) Early age (Feret diameter of single colonized micro- carriers) and (b) late-age (Feret diameter of single colonized mi- crocarriers and clusters) kinetics; min, max, and mean are the minimal, maximal, and mean diameters, respectively, of empty mi- crocarrier population; (0) F25 microcarriers, (0) F26 microcarri- ers, (m) F28 microcarriers, (X) F26 clusters, (*) F27 clusters.

Ovz T tiequcncy F26, culture age = 55 hr

peak 2: voidr

0 12 16 20 circle radius (R) (pn)

(4

EL5, culture age = 256 hr

0 4 8 12 16 20 circle radius (R) urn)

(4 Figure 8. Examples of object size distribution: (a) two-peak dis- tribution for early culture age; (b) one-peak distribution for late culture age.

covered, the size distribution exhibits two peaks: the object zone corresponds to cells, the size of which is obtained from the first peak examination, and to zones of uncovered microcarrier surface (voids), which pro- duce the second peak. The characteristic size of the cells, R,, has been chosen to be the upper limit of the first peak, R1,, and not the value of R corresponding to the peak maximum, Rim, in order to take into account the spherical shape of the microcarrier, which intro- duces a bias on the size of the cells situated near the boundary of the mask. The characteristic void size, R,, is the upper limit of the second peak, R2,. For fully colonized microcarriers and for clusters the distribution exhibits only one peak, the upper limit of which is se- lected to be R,. When the microcarrier is almost fully colonized, cells and voids have characteristic sizes of the same order of magnitude: A global one-peak distribu- tion is obtained and R, and R, cannot be distinguished, unless there is a manual selection by the operator of the zones relative to cells before the size measurement pro- cedure is applied. Two-peak distributions are some- times also obtained at the end of the culture: Some cells tend to leave the microcarriers and jump on existing clusters; therefore, empty microcarriers, appearing as large void zones, are attached to some clusters. The maximal theoretical value of R, should be of the order of 40 pm, which corresponds to the half-diameter of an empty microcarrier. A value of 30 pm is obtained, which is in good agreement, as in the very early ages a few cells colonized the microcarriers and as in the late ages the empty microcarrier is attached to the cluster, decreasing the surface to be measured.

Figure 9 summarizes the results obtained on the four cultures for voids and cells. No strong variation of R,, the order of magnitude of which has been checked by direct measurement of high-magnification slides, is no- ticed. The maximal value found for R, is of the order of 10 pm. A slight decrease is noticed during the first hours, while the microcarrier is progressively covered: The av- erage R, value is then 5 pm. The cells tend to spread more when free space is available on the microcarrier.

CONCLUSION

Quantitative image analysis has been used to monitor the kinetics of some morphological parameters of mam- malian cells grown on microcarriers. Special procedures have been developed to deal with pictures cbtained by SEM. Results indicate a slight decrease of the cell size during the microcarrier colonization and a rather ho- mogeneous cluster size during the stationary phase.

More generally these procedures are independent of the cell shape and can be applied to other mammalian cell lines. The main difficulty is the quality of SEM pic- tures and their illlumination uneveness. SEM is indeed an expensive off-line method which requires special

COMMUNICATIONS TO THE EDITOR 191

25.

20

15 .

10

5

A

O J I

0 200 400 600 800 lo00 1200 1400 1600 1800 culture age (hr)

(4

10 15 1% ’ t O J I

0 200 400 600 800 lo00 lux) 1400 1600 1800 culture age (hr)

(b)

Figure 9. (0) F28, (m) F27.

(a) Cells and (b) voids size kinetics: (A) F25, (0) F26,

sample preparation. The dehydration procedure prevents determination of the actual cell size in its hydrated state, and only comparison of results obtained on iden- tically prepared samples is valid. Magnification should be carefully selected, especially in the case of clusters: Images providing information on the cluster size are generally not suitable for cell size characterization. The results should be obtained from a statistically meaning- ful number of images per sample. It is questionable that the method remains complex. It takes a total of about 15 mn to treat one slide. However, due to the dynamics of mammalian cell culture, this drawback should not impede the use of image analysis in research and pro- duction, as it enables us to quantify information other- wise attainable only by approximate means.

The authors thank Dr. Einsele from Sandoz Ltd., Basel, Switzerland, and J. F. RCmy, who conducted the SEM.

APPENDIX

An image A of n l lines of nz picture elements (pixels) having grey levels a,, i = 1,. . . , n l , j = 1,. . . , nz, is equivalent to a nl X n2 ma- trix {arJ}; the a,J take integer values in the interval [0,255].

The result of the convolution of image A {ad by the matrix K (3 X 3) of elements {kk,r} is a new image B {b,J}: b, takes into account the grey-level values of the eight neighbors of pixel ( i , j ) in image A:

bZJ = k i i U r 1 , J - I - k ~ a , - l , , + k i s a , - ~ , ~ + l + k z ~ a , , , - ~ + k n a , , ~

+ kzsa,,,+l + k31a,+l,~-1 + k32a#+l,~ + k33a,+l.J+~

A calibration is necessary to obtain b,J E [0,255]; the Laplacian ma- tr ix K, which enhances the edges, is

K = [ -1 0 4 0 -1 0 1

-1 0 -1

Thresholdmg is the operation which transforms a grey-level image A {u,~} with a,J E [0,255] into a binary image B {bs}, b,,J E [0,1]. In upper limit thresholding the pixel ( i , j ) of A , which has a value a,J lower (respectively higher) than a preset value (threshold) has in image B the value b, = 1 (respectively 0).

Object filling is the operation by which all the holes (correspond- ing a,J = 0) within object zones of a binary image A are filled (cor- responding b, = 1 in new image B).

The result of the logical addition of two binary images A {ad and B {b,J} is image C {ctJ}: ctJ = 1 if a4 = 1 and b, = 1, cg = 0 otherwise.

The complementary image B {b,J} of image A is such that

1 when^,^ = 0 0 when^,^ = 1 bq =

The result of erosion of binary image A { a d is image B {brj}:

1 0 otherwise

if all arf = 1 (i’ = i - 1, i + 1, j‘ = j - 1, j + 1) bij = {

The result of dilation of image A {azJ} is image C {cd :

1 if one of the a,g at least is equal to 1 c , ~ = ( i ’ = i - l , i + l , j ’ = j - l , j + l ) I 0 otherwise

These operations can be combined: An opening is an erosion fol- lowed by a dilation and a closing is a dilation followed by an ero- sion. The four operations can also be performed iteratively.

The convex hull of an object S is the smallest convex object con- taining S.

References

1. Adams, H. L., Thomas, C. R., 1988. The use of image analysis for the morphological measurements on filamentous microor- ganisms. Biotechnol. Bioeng. 32: 707-712.

2. Borrelli, M. J., Wong, R. S. L., Dewey, W. C. 1986. A direct correlation between hyperthermia-induced membrane blebbing and survival in synchronous G I CHO cells. J. Cell. Physiol. 126:

3. Costello, P. J., Monk, P. R. 1985. Image analysis method for the rapid counting of Succharomyces cerevisiae cells. Appl. Environ. Microbiol. 49(4): 863-866.

4. Coster, M., Chermant, J.-L. 1989. Pricis d’analyse d’images, 2nd edition. CNRS, Paris.

5. Csauscher, R., Mundry, K.W. 1990. Visualization of cell physi- ology withvideo intensification microscopy (VIM). pp. 114-115 In: M. Reuss, H. Chmiel, E. D. Gilles, and H. J. Knackmuss (eds.), Biochemical engineering. Gustav Fischer, Stuttgart.

6. Goebel, N. K., Kuehn, R., Flickinger, M. C. 1990. Methods for determination of growth-rate dependent changes in hybridoma volume, shape and surface structure during continuous recycle. Cytotechnology 4: 45-57.

7. Gonzalez, R.C., Wintz, P. 1987. Digital image processing. Addison-Wesley, Reading, MA.

8. Hill, G. A., Robinson, C.W. 1988. Morphological behaviour of Saccharomyces cerevisiae during continous cultures. Biotechnol. Lett. 10: 815-820.

9. Matsumura, M., Imanaka, T., Yoshida, T., Taguchi, H. 1980. Morphological differentiation in relation to cephalosporin C

181-190.

192 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 40, NO. 1, JUNE 5, 1992

synthesis by Cephalosprium acremonium. J. Ferment. Technol.

10. Miller, S. J. O., Henrotte, M., Miller, A. 0. A. 1986. Growth of animal cells on microbeads. I. In situ estimation of numbers. Biotechnol. Bioeng. 28: 1466-1473.

11. Packer, H.L., Belmar Camper, M.T., Thomas, C . R . 1989. Morphological measurement on filamentous micoroorganisms by image analysis. pp. 23-35 In: N.M. Fish, R.I. Fox, and N. F. Thornhill (eds.), Computer applications in fermentation technology; modelling and control of biotechnological pro- cesses. Elsevier, London.

12. Packer, H. L., Thomas, C. R. 1990. Morphological measure- ments on filamentous microorganisms by fully automatic image analysis. Biotechnol. Bioeng. 35: 870-881.

58(3): 197-204. 13. Reichl, U., Gilles, E. D. 1990. Investigations of pellet-forming

microorganisms by means of an image processing system. pp. 336-3395 In: M. Reuss, H. Chmiel, E. D. Gilles, and H. J. Knackmuss (eds.), Biochemical engineering. Gustav Fischer, Stuttgart.

14. Serra, J. 1982. Image analysis and mathematical morphology. Academic, London.

15. Wagner, A. 1990. Ph.D. Thesis, Institut National Polytech- nique de Lorraine.

16. Wagner, A., Marc, A., Engasser, J. M., Villermaux, S., Einsele, A. 1990. Continuous production of prourokinase in feed har- vest and perfusion cultures. Biotechnol. Bioeng. 36: 623-629.

COMMUNICATIONS TO THE EDITOR 193