Accuracy of Three Techniques to Determine Cell ...

TISSUE ENGINEERING: Part C Volume 14, Number 00, 2008 ª Mary Ann Liebert, Inc. DOI: 10.1089=ten.tec.2008.0313

Accuracy of Three Techniques to Determine Cell Viability in 3D Tissues or Scaffolds Benjamin Gantenbein-Ritter, Ph.D.,1 Esther Potier, Ph.D.,1,2 Stephan Zeiter, D.V.M.,1 Marije van der Werf, M.Sc.,1,2 Christoph M. Sprecher, Dipl-Ing.,1 and Keita Ito, M.D., Sc.D.2

Several different assays are commonly used to evaluate survival of cells inside tissues or three-dimensional carriers, but their accuracy and reliability have not been evaluated. Here, we compare three methods for cell viability (CV) determination: (i) lactate dehydrogenase (LDH) staining on cryosections, (ii) calcein AM=ethidium homodimer-1 (CaAM=EthH) staining, and (iii) carrier digestion and trypan blue (TB) assay. Living and dead cell populations were generated from bovine chondrocytes and combined to produce approximately 0%, 25%, 50%, 75%, and 100% CV mixtures. CV ratios were measured with TB assay (MIX) before seeding cells into fibrin carriers. CV was then determined using the three methods (n ¼ 5=method). Custom-written macros were used to process LDH- and CaAM=EthH-stained images, and hand counting with hemocytometer was used for the TB method. Absolute error and intraclass correlation (ICC) were used for accuracy and reliability evaluation. All methods estimated CV values close to MIX values. TB method was the most accurate (ICC ¼ 0.99) followed by CaAM=EthH (ICC ¼ 0.98) and LDH (ICC ¼ 0.97). As for absolute quantification of living and dead cells, TB and LDH methods performed well (ICC ¼ 0.75–0.96), whereas CaAM=EthH largely overestimated cell numbers (living, ICC ¼ 0.30; dead, ICC ¼ 0.30). Although TB was the most accurate, LDH and CaAM=EthH provide valuable information on cell shape and spatial distribution of cells in tissue or a scaffold.

blue (TB).9,10 Even though these methods are widely used, few studies have been done to validate them in terms of accuracy and reliability.2,11 Previous studies, however, only validated semiquantitatively accuracy and variance between two methods and failed to demonstrate accuracy relative to given a priori controlled target values. Here, using cell populations of known percent viability seeded into a fibrin carrier, we systematically compare the above-mentioned three methods for determination of CV as well as number of living, dead, and total cells in terms of accuracy and reliability.

Introduction

C

ell viability (CV) is an important parameter in tissue engineering and culture studies to evaluate long-term survival of cells. Currently there are a number of assays available to determine CV in tissue or three-dimensional (3D) scaffolds, including lactate dehydrogenase (LDH) staining1,2; calcein AM with ethidium homodimer-1 (CaAM=EthH) staining,3–5 for example, Live=Dead
; and cell counting after scaffold=tissue digestion.6–8 The LDH assay detects LDH activity in living cells by conversion of nitrotetrazolium blue chloride (NBT) into a nonsoluble brown precipitate (formazan) on histological sections. Dead cells can be detected using nucleic acid staining such as propidium iodide (PI). The CaAM=EthH dyes can be used to stain living and dead cells directly in the tissue or scaffold. The CaAM is enzymatically hydrolyzed into calcein in living cells, turning those fluorescent green. The EthH, on the other hand, is only able to enter cells with a compromised membrane and stains nucleic acid fluorescent red. It is also possible to count cells after tissue=scaffold digestion using different dyes or stainings differentiating living from dead cells, such as CaAM or trypan

Materials and Methods Cell source and expansion Articular cartilage was harvested from the femoral condyles of a 4-month-old calf obtained from the local abattoir. Approximately 25-mm3 tissue pieces were washed with Tyrrode’s balanced salt solution (TBSS) and predigested in spinner flasks with 1 mg=mL pronase I (Roche, Basel, Switzerland) in TBSS for 2 h at standard conditions (378C, 5% CO2, 95% humidity). Predigested cartilage was then washed with

1

AO Research Institute, AO Foundation, Davos, Switzerland. Department of Biomedical Engineering, Eindhoven University of Technology, Eindhoven, The Netherlands.

2

1

2 TBSS and digested with 600 IU=mL collagenase II (Worthington, Allschwil, Switzerland) in Dulbecco’s modified Eagle’s medium (high-glucose DMEM; Gibco, Basel, Switzerland) by stirring overnight (14 h). After washing in TBSS, the primary chondrocytes were expanded up to P2 in low-glucose DMEM, 10% fetal calf serum (FCS), and 1% antibiotic=antimycotic solution and then frozen in high-glucose DMEM containing 10% dimethyl sulfoxide (all Gibco). Cells were thawed and expanded up to P8 in high-glucose DMEM, 10% FCS, and 1% antibiotic=antimycotic solution in standard conditions for experimental use. Cell viability mixtures To prepare the dead cells, 3 days before gel casting, cells were detached with 1% trypsin (Gibco), centrifuged (400 g, 7 min, room temperature [RT]), and resuspended in TBSS. Cell death was then induced by adding 1 mL of 1 N HCl to 9 mL of cell suspension resulting in a pH of *4. This solution was incubated for 10 min at RT. The cells were then centrifuged, rinsed with TBBS, and stored at 48C. The day before gel casting, dead cells were heat treated for 60 min at 608C. On the day of gel casting, living cells were detached, centrifuged, and resuspended in TBSS. The cell concentration of both solutions was determined (TB assay), and the two solutions were then mixed at 0=100%, 25=75%, 50=50%, 75=25%, and 100=0% living=dead cell ratios. Dead and living cells were hand counted (three different aliquots were counted nine times each and averaged) using TB exclusion assay (TB; Sigma, Buchs, Switzerland) to determine CV ratios before gel casting (MIX). Fibrin carriers Mixed cell suspensions were centrifuged and resuspended in a 50 mg=mL fibrinogen solution (Baxter, Vienna, Austria) at 7
106 cells=mL. Within 5 min, 100 mL of thrombin solution (Baxter; 10 U=mL) was then added to 900 mL fibrinogen=cell solution. 230 ml of this solution was transferred into a custommade mould (diameter 8 mm
4 mm height, final volume of 201.05 mm3) and left for 20 min at RT. The final cell density in the fibrin carriers was *7000 cells=mm3, approximately corresponding to that observed in intervertebral disc and articular cartilage.12,13 After removal from the moulds, the carriers were kept in high-glucose DMEM without FCS until assessment of CV using the three different methods. LDH staining Fibrin carriers (n ¼ 5=cell mixture) were transferred into cryocompound (Tissue Freezing Embedding, Jung, Nussloch) and snap-frozen in liquid N2. From each carrier, 12-mm-thick mid-sagittal cryosections were cut (Cryo Star HM 560 MV; Microm, Walldorf, Germany). After mounting, slides were thawed at RT for 10 min and incubated with LDH staining solution1 (2 mM Gly-Gly buffer, pH 8.0, containing 40% Polypep, 60 mM lactic acid, 1.75 mg=mL b nicotinamide adenine dinucleotide (NAD), and 3 mg=mL NBT; all Sigma) for 2.5 h at 378C. Stained slides were then rinsed once with 508C tap water, once with RT phosphate-buffered saline (PBS), and fixed with 4% phosphate-buffered formalin (10 min at RT). Stained and fixed slides were then counterstained in PBS containing 1 mg=mL PI, 0.1% Triton 100X, and 0.5 mg=mL RNaseA (all Sigma) for 12 min at RT. After staining, slides

GANTENBEIN-RITTER ET AL. were rinsed three times in PBS and a coverslip was mounted (Hydromount; National Diagnostics, Atlanta, Georgia). LDH and PI stainings were visualized using a combined light=fluorescence microscope (Axioplan 2; Carl Zeiss, Feldbach, Switzerland) and imaging software (Axiovision software; Carl Zeiss). Three images per section (one section per carrier) were taken at a resolution of 2600
2060 pixels at 10
magnification. Bright and fluorescence images were then individually thresholded by the operator. Using a custommade macro (Zeiss300 software; Carl Zeiss), living cells (LDH positive) were counted on segmented brightfield images as islands ranging from 50 to 1000 mm2 and total cells (LDH and=or PI positive) were counted on the combined segmented fluorescent and brightfield images as island ranging from 30 to 1000 mm2. For LDH staining, the number of cells per carrier were computed based on the volume of the optical section, that is, 0.018 mm3 ¼ 1381.77
1094.78 (calculated from image size)
12 mm (thickness of cryosection). CaAM=EthH staining Fibrin carriers (n ¼ 5=cell mixture) were cut sagittally into halves and incubated in 1 mL high-glucose DMEM without FCS containing 10 mM CaAM and 1 mM EthH (both Fluka, Buchs, Switzerland) for 3 h at 48C followed by 1 h at 378C, 5% CO2, and 100% humidity. The carriers were then scanned from top and bottom surfaces to *200 mm depth at two random locations per side with a confocal laser scanning microscope (cLSM510; Carl Zeiss). Stacks were taken at 10
magnification at a 512
512 pixel resolution (field size of 921.4
921.4 mm) with the pinhole at 1 Airy unit and 50% image overlap and 5.8-mm intervals. Red and green cells were quantified per single image using a custom-made macro in ImageJ software14 (deposited at NIH, http:==rsbweb.nih.gov= ij=plugins=index.html). The macro cellcounter3D consists of a threshold step that passes a binary image with pixels in the range of 100–255 to the plug-in ‘‘nucleus counter’’ (McMaster University, Biophotonics Facility, http:==www .macbiophotonics.ca=imagej), which then uses the ‘‘Otsu’’ method for particle counting. The minimum and maximum island sizes were set to 7–50 and 15–100 pixels for the red and green channels, respectively. The CV was then estimated on a subset of 10 consecutive images, starting at the image with 50% or more of the maximum amount of the total cells per image in a single stack (Fig. 1). To determine this frame, a MatLab routine was written (MatLab 2007; MathWorks, Bern, Switzerland). Figure 1 illustrates a typical profile through a stack of images for the living, dead, and total cells. To validate the cell numbers of the new macro routine, number of living and dead cells and CV were alternately estimated based on projections over the same 10 consecutive image scans, and the same macro was run with the same parameters as described above. The cell numbers per image were transformed to a cell number per carrier assuming an optical slice volume of 0.011 mm3 (921.4
921.4
13.4 mm). Trypsin digestion and TB staining Fibrin carriers (n ¼ 5=cell mixture) were cut into four pieces, washed in TBSS for 5 min at RT, and incubated in 1 mL of TBBS containing 1% bovine pancreas trypsin (Sigma) at 378C until dissolved (70–90 min). Trypsin was then inactivated adding 1 mL of FCS. Cells were centrifuged (500 g, 10 min,

ACCURACY OF CELL VIABILITY METHODS

3 Statistical analysis For each carrier, the repeated measure within each method, for example, cell counting, was averaged. The obtained values were calculated back to the cell number per carrier. CV was defined as CV ¼

l , lþd

where l and d are number of living and dead cells, respectively. Further, in order to quantify the accuracy in terms of the three methods to the MIX, the ‘‘absolute error’’ (AE) was calculated as AEi,j ¼ xi,j  pi ,

FIG. 1. Number of living, dead, and total cells per image through a stack of confocal laser scanning images (CV was 75%). The two shaded lines on the x-axis mark the starting and end image of 10 consecutive images, which were selected as the region of interest. The first image was that which had 50% of the maximum number of cells found on any image in the stack.

RT) and resuspended in TBSS. An aliquot of the resulting cell suspension was then taken and incubated with the same volume of 0.4% TB solution for 20 min at RT. Using a Neubauer improved hematocytometer, dead and living cells were hand counted (three different aliquots per digested carrier, each counted three times) and CV evaluated.

where xi;j is the mean CV for method, j, and pi is the mean CV estimated before gel casting (MIX) for the CV mixture, i. Significance of AE was assessed by testing values, xi,j, of each method, j, for each CV mixture type, i, relative to MIX, pi, with a nonparametric Mann–Whitney U-test. For all tests, a Bonferroni correction was applied where a p-value 0.80 as outstanding.18 Results The MIX values all matched well the targeted CV mixtures, that is, 0.0  0.0%, 21.6  2.1%, 51.8  1.0%, 73.4  2.4%, and 96.6  0.7% corresponding to 0%, 25%, 50%, 75%, and 100% living cells target. The calculated mean cell number per carrier was 1.55  0.28
106 cells. Representative images are shown in Figure 2 for each of the five viability mixtures and for each method. The images of all methods showed a clear gradient from 0% to 100% CV. The 0% viability mixture validated the protocol of cell death induction by acidic shock and heat deactivation of enzymes as there were no living cells observed for any of the methods used (Fig. 2, first column). With respect to CV, all three methods performed well as expressed by high pairwise ICCs between each of the three methods and MIX values (Table 1). TB method was the most accurate (ICC ¼ 0.99) followed by CaAM=EthH (ICC ¼ 0.98) and LDH stainings (ICC ¼ 0.97). In terms of precision, all three methods had similar range of standard deviations (Fig. 3). LDH staining method CV was overestimated by 12.2% and 12.0% for the 25% and 50% viability mixtures, respectively, resulting in a lower ICC of 0.97 (Fig. 3A and Table 1). The method also overestimated number of living cells in viability mixtures $ 50% (ICC ¼ 0.75, Fig. 3B) but, on the other hand, was extremely accurate to count the number of dead cells (ICC ¼ 0.96, Fig. 3C). The mean total number of cells was 2.11  0.63
106 cells per carrier, which was 1.4 times off from the theoretical value (Fig. 3D). CaAM=EthH method CV was overestimated in the 25%, and underestimated in the 75% viability mixtures (Fig. 3A). Number of living cells was always overestimated (ICC ¼ 0.30) with an AE to MIX increasing with amount of viable cells (Fig. 3B). For all CV mixtures, this method also overestimated the number of dead cells (ICC ¼ 0.34, Fig. 3C). The mean total number per carrier was 4.66  1.53
106 cells, which was three times off from the theoretical value (Fig. 3D).

FIG. 3. AE relative to MIX for CV (A), living cell number per carrier (B), dead cell number per carrier (C), and total cell number per carrier (D) as evaluated using three different methods, that is, LDH, CaAM=EthH, and TB, for five different CV mixtures. Values are mean  SD, n ¼ 5.

ACCURACY OF CELL VIABILITY METHODS

5

Table 2. Advantages and Inconveniences of the Compared Methods

LDH CaAM=EthH TB

Recovery of the cells

Backup of the original carrier

Cell morphology

Cell distribution

Equipment

Time consuming

  þ

þþ  

þ þþ 

þþ þþ 

þ þþ 

þþ þ 

, not appropriate; þ, useful; þþ, very useful.

TB method

Applicability=handling of methods

In all CV mixtures, this method estimated the number of living cells extremely close to MIX values (ICC ¼ 0.95, Fig. 3B). Only in the 0% and 25% CV mixtures did this method considerably underestimate the number of dead cells (ICC ¼ 0.79, Fig. 3C). Mean total cell number was estimated to 1.14  0.21
106 cells per carrier, thus, off the target value by a factor of 0.7.

LDH method provides a good estimate of dead cell number, but it can also offer important information regarding cell distribution throughout the tissue=carrier. Using this method, it is possible to detect potential cell layer, cell density gradient, as well as cell aggregates. A potential caveat of this method, however, is the detection of ‘‘freshly’’ dead cells as living cells, due to LDH stability (up to 36–38 h after cell death). This should be considered when analyzing LDH staining results. Compared to the other methods studied here, the CaAM= EthH technique offers the possibility to observe cell morphology, thus, providing additional information on cell behavior. This method, however, only allows analysis of tissue cores of limited size and depth. This has been evaluated by scanning fibrin carriers containing fluorescent microbeads of 6 mm diameter (seeded at similar density, AlignFlow; Molecular Probes, Invitrogen, Basel, Switzerland). The imaging resulted in similar fluorescent intensity profiles with maximum scan depth of *200 mm. Thus, the fading out of both green and red signals deep in the carrier (Fig. 1) is most likely caused by optical limitation than by diffusion problem or cell death. Contrary to the other proposed methods, the trypsin digestion=TB staining method does not provide additional information on cell spreading or spatial distribution of cells. Nevertheless, it was found to be the most accurate to determine CV and living and total cell numbers. Table 2 gives an overview on advantages=disadvantages of the different methods. TB method is advantageous to the other two methods in terms of simplicity, needed infrastructure, and required lab time, but provides no information on cell distribution or morphology. Further, it can only be used with scaffold=tissue that can be digested without killing the cells.

Discussion Cell viability All of the three analyzed methods performed well relative to MIX to determine CV (Fig. 3A and Table 1) with respect to accuracy and reliability, with highest pairwise ICC for the TB method. All of the methods, however, over- or underestimated CV for 25%, 50%, or 75% CV mixtures to different extents. Nevertheless, these deviations were not significant. We did not observe systematic overestimation of CV for the CaAM=EthH method as found in a recent study on chondrocyte viability in osteoarticular allografts but rather an error of a random nature (Fig. 3A).19 All three methods had similar range of standard deviations. Can methods accurately quantify cell numbers? LDH and CaAM=EthH methods overestimated number of living cells with respect to MIX in all CV mixtures (except for 0%), whereas TB method always slightly underestimated this number but was clearly the one closest to the MIX values. In fact, the CaAM=EthH method systematically overestimated living, dead, and total cell number by a factor of 2–4. In order to have a confirmation of these high estimates, we also evaluated cell numbers by projection of 10 consecutive images, and by subsequent cell counting using our cell-counting macro and parameters (instead of counting cell number per optical slice of 0.011 mm3). However, the estimated total cell number with a mean of 4.30  0.40
106 cells was again about three times higher than the 1.55
106 cells seeded per carrier. The overestimation of total cell number by the CaAM=EthH method can partially be explained by inaccurate calculation of the effective volume, a complex product of the refraction index of the media, which itself is a function of fibrin; water content; and the angle of the scans and other unknown factors. Thus, for quantitative purposes, the confocal laser scanning microscope technique may not be applicable, in particular, for tissue screening where water content, extracellular matrix, and its optical properties may be highly variable between samples.

Limitations The current study relies heavily on the accuracy of the MIX, which itself was estimated by the TB method. This could be considered as a potential bias and might explain highest accuracy of TB. This source of error, however, should be minimal since it was done prior to seeding the cells in the carrier, and all methods predicted CV very well. Additional variance could have come from the macros that were used to count cells for LDH and CaAM=EthH methods. This error, however, was systematic and did not contribute to the observed variance. Some of the parameters of the proposed macro for counting cells in laser scanning microscopy (LSM) stacks and on LDHstained sections, such as the minimal pixel island size to define a cell or the overall thresholding step, are values that need to be fixed for the macro to run. Their choice is empirical and

6 may need to be adjusted for a specific cells=tissue or staining (using alternate dyes). Accurate evaluation of CV relies on proper quantification of living and dead cells. Particularly, dead cells can become difficult to detect in long-term cultures due to weakening of the dead cell membrane and partial nucleic acid digestion. Absolute number of living cells, thus, might be a more reliable parameter to evaluate long-term survival of cells into tissues or carriers. However, for LDH and especially for CaAM= EthH, the number of living cells will be overestimated. The conclusions from this study might be applicable to more general 3D tissues and other scaffolds. However, in 3D tissues, many parameters are not fixed. For instance, methods, which involve a digestion step (like the TB in our study), tend to digest selectively dead cells, which then leads to random overestimation of CV in any downstream protocol for CV. The way ‘‘freshly dead cells’’ were induced here may not be comparable to cells that died from natural cell death, which was likely to be a major reason why TB method fitted so nicely with MIX. Replacing fibrin with another hydrogel=or even a tissue changes the optical characteristics, which then influences outcomes of LDH and CaAM=EthH method. Conclusions All three methods predicted CV very well. TB method, however, was the most accurate, followed by CaAM=EthH and LDH stainings. In terms of absolute cell numbers, TB and LDH assays both came very close to the premixture values. CaAM=EthH overestimated cell numbers on average by a factor of 3. These methods, however, possess different advantages and inconveniences (Table 2) that should be taken into account when selecting a method for CV assessment. Acknowledgments We thank M. Stoddart for helpful comments and the new protocol for the CaAM=EthH stain, and K. Schwieger for advised help in statistical analyses. References 1. Stoddart, M.J., Furlong, P.I., Simpson, A., et al. A comparison of non-radioactive methods for assessing viability in ex vivo cultured cancellous bone: technical note. Eur Cell Mater 12, 16–25, 2006; discussion 16–25. 2. Rauch, B., Edwards, R.B., Lu, Y., et al. Comparison of techniques for determination of chondrocyte viability after thermal injury. Am J Vet Res 67, 1280, 2006. 3. Willerth, S.M., Arendas, K.J., Gottlieb, D.I., et al. Optimization of fibrin scaffolds for differentiation of murine embryonic stem cells into neural lineage cells. Biomaterials 27, 5990, 2006. 4. Mauck, R.L., Yuan, X., and Tuan, R.S. Chondrogenic differentiation and functional maturation of bovine mesenchymal stem cells in long-term agarose culture. Osteoarthritis Cartilage 14, 179, 2006.

GANTENBEIN-RITTER ET AL. 5. Park, S.H., Park, S.R., Chung, S.I., et al. Tissue-engineered cartilage using fibrin=hyaluronan composite gel and its in vivo implantation. Artif Organs 29, 838, 2005. 6. Catelas, I., Sese, N., Wu, B.M., et al. Human mesenchymal stem cell proliferation and osteogenic differentiation in fibrin gels in vitro. Tissue Eng 12, 1, 2006. 7. Ho, W., Tawil, B., Dunn, J.C., et al. The behavior of human mesenchymal stem cells in 3D fibrin clots: dependence on fibrinogen concentration and clot structure. Tissue Eng 12, 1587, 2006. 8. Haschtmann, D., Stoyanov, J.V., Ettinger, L., et al. Establishment of a novel intervertebral disc=endplate culture model: analysis of an ex vivo in vitro whole-organ rabbit culture system. Spine 31, 2918, 2006. 9. Melamed, M.R., Kamentsky, L.A., and Boyse, E.A. Cytotoxic test automation: a live-dead cell differential counter. Science 163, 285, 1969. 10. O’Brien, R., and Gottlieb-Rosenkrantz, P. An automatic method for viability assay of cultured cells. J Histochem Cytochem 18, 581, 1970. 11. Imbert, D., and Cullander, C. Assessment of cornea viability by confocal laser scanning microscopy and MTT assay. Cornea 16, 666, 1997. 12. Maroudas, A., Stockwell, R.A., Nachemson, A., et al. Factors involved in the nutrition of the human lumbar intervertebral disc: cellularity and diffusion of glucose in vitro. J Anat 120, 113, 1975. 13. Hunziker, E.B., Quinn, T.M., and Ha¨uselmann, H.J. Quantitative structural organization of normal adult human articular cartilage. Osteoarthritis Cartilage 10, 564, 2002. 14. Rasband, W.S. ImageJ, National Institutes of Health, Bethesda, MD, http:==rsb.info.nih.gov=ij=, 1997–2007, 2007. 15. Sokal, R.R., and Rohlf, F.J. Biometry. W.H. Freeman Company, NewYork, NY, 1995. 16. SPSS Inc. SPSS Base 14.0 for Windows User’s Guide, 2007. 17. Shrout, P.E., and Fleiss, J.L. Intraclass correlations: uses in assessing rater reliability. Psychol Bull 84, 420, 1979. 18. Landis, J.R., and Koch, G.G. The measurement of observer agreement for categorical data. Biometrics 33, 159, 1977. 19. Lightfoot, A., Martin, J., and Amendola, A. Fluorescent viability stains overestimate chondrocyte viability in osteoarticular allografts. Am J Sports Med 35, 1817, 2007.

Address reprint requests to: Benjamin Gantenbein-Ritter, Ph.D. AO Research Institute Clavadelerstrasse 8 CH-7270 Davos Switzerland E-mail: [email protected] Received: May 31, 2008 Accepted: July 4, 2008