Evaluation of Cell Viability by Flow Cytometry
Reliable determination of the relative abundance of viable cells from flow cytometric data requires proper selection of parameters, that is, excitation wavelengths and gating procedures. In Figure 1, scatterplots are shown for live/dead staining of a sample of FACS sorted monocytes. Fluorescence intensity of EHD (blue) and EHD (green) are plotted versus the calcein signal in Figures 1a and 1b, respectively. Original data are shown in each of these scatterplots, that is, no spectral compensation was applied. Besides the three clusters of monocytes (M), the major remainders in the samples were platelets (Plt), which can be stained by calcein but not by EHD because of the lack of DNA. Microscopic inspection showed that no adherence of platelets to monocytes occurred. When comparing Figures 1a and 1b, the advantage of the excitation of the EHD fluorescence at 532 nm is clearly discernible. The excitation wavelength of 488 nm results in an extensive crosstalk of calcein signal in the EHD (blue) fluorescence detection channel, and hence Plt and (viable) monocytes are arranged close to a straight line of upward slope (Fig. 1a). In contrast, when exciting at 532 nm, no influences between both detection channels occur (Fig. 1b) and discrimination of different subpopulations is straightforward by quadrant analysis. This is demonstrated by selection of color gates for dead monocytes (R2, red), live monocytes (R3, green) and platelets (R1, blue) in Figure 1b. These gates cover about 98.5% of all events observed. The corresponding analysis when exciting at 488 nm would not be based on Figure 1a to avoid problematic selection of delineation lines. Instead spectral compensation as demonstrated in Supporting Information 1 would be used to derive the relative cell numbers. This is no longer necessary, since 532 nm excitation results in a significant improvement for the cell differentiation.
Figure 1. Flow cytometric differentiation of calcein and ethidium homodimer-1 (EHD) stained cells (example for a FACS sorted sample); M, monocytes; Plt, platelets. EHD fluorescence of stained cells was excited by 488 nm (a) or 532 nm (b) laser light. Color labeling of regions R1, R2, R3 was applied to compare corresponding cell populations in both scatterplots (gate statistics: total count 200040, R1: 5.31%, R2: 16.94%, R3: 76.21%). [Color figure can be viewed in the online issue which is available at wileyonlinelibrary.com.]
Download figure to PowerPoint
Besides the parameters EHD (green), EHD (blue), and calcein, light scatter at 488 and 640 nm and anti-CD14-APC fluorescence was measured. It should be noted that the fluorescence signal of anti-CD14 labeled monocytes (M) is less intense than usual, since magnetic labeling of the CD14 epitopes was carried out first and consequently the number of free reaction sites decreased. To evaluate the stability of our data analysis, we examined different gating strategies for the identification of monocytes based on the SSC 640 versus FSC 640 scatterplot, the anti-CD14-APC fluorescence versus SSC 488 diagram, and a lower threshold for the SSC 488 light scatter intensity in the FSC 488 versus SSC 488 scatterplot. All these approaches yielded good agreement with respect to cell viability and we chose one parameter gating of light scattering intensity to discriminate of the CD14 target cells against debris and platelets. This is illustrated in Figure 2a, where we included the threshold UG for the SSC 488 in the FSC 488 versus SSC 488 scatterplot. In the context of this work, selection of the light scattering parameter (SSC 488) for flow cytometric analysis is essential to allow a well-founded comparison with microscopic measurement, since an equivalent or corresponding morphological parameter is accessible in bright field microscopy, as discussed below.
Figure 2. Gating for flow cytometric determination of cell viability from scatterplots (example for a FACS sorted sample). (a) Threshold setting for separating target cells (M: monocytes) and background (Plt:platelets) based on light scatter. The solid line indicates the threshold for delineation of target cells and dotted lines mark the thresholds used for uncertainty analysis. (b) Scatterplot for live/dead staining using data from excitation by 532 nm laser light. The result was obtained applying gating by threshold UG. Four regions were used to count calcein and ethidium positive and negative cells, respectively. The color scale is the same for (a) and (b) and gives the number of events on a logarithmic scale (gate statistics: total count above UG: 188794, NI: 153053, NII: 32961, NIII: 86, NIV: 2694). [Color figure can be viewed in the online issue which is available at wileyonlinelibrary.com.]
Download figure to PowerPoint
In Figure 2a, platelets and other particles exhibit small SSC 488 signals below the threshold UG. As result of the gating procedure, which excludes all these events, we obtain the scatter plot in Figure 2b. It should be noted that Figure 2b, depicting EHD (green) versus calcein fluorescence intensities, represents the same measurement as Figure 1b. Further analysis was based on the identification of live and dead cells by quadrant analysis as shown in Figure 2b. In region I, viable cells were located as measured by positive calcein staining, their number being NI. The corresponding relative abundance with respect to all events (NI + NII + NIII + NIV) in Figure 2b is designated as RI = NI / (NI + NII + NIII + NIV). In region II, NII EHD-positive cells were enumerated as dead cells yielding the fraction RII. Region III covers a relatively small fraction of cells, which were positive for both EHD and calcein. By microscopy, this interpretation was supported and coincidences were excluded since these cells simultaneously show a cytoplasm stained with calcein and a nucleus stained by EHD. However, RIII rarely exceeded 0.1% when immediately measured after live/dead staining. Region IV covers cells that neither react with calcein nor with EHD. Typically, frequencies were below 2%, that is, RIV < 2%.
The measurement procedure applied here allows classification of each event according to both, live stain and dead stain and hence allows the exclusion of double positive (region III) and double negative (region IV) events. However, to be consistent with usually applied procedures, where live or dead cells are identified by staining with a single dye, and to facilitate reliable comparison with microscopy, we followed the approach of Lange et al. (17) and derived cell counts corresponding to single staining procedures. The fraction of cells stained with calcein AM determine the relative abundance of viable cells with respect to all cells as Vca = RI + RIII while positive EHD staining yielded the complement to viable cells as DEHD = RII + RIII. These relative abundances were measured with respect to all events in the scatterplot in Figure 2b. Both, that is live and dead staining methods provide complementary information and their sum can be used to prove consistency of results. For FACS- and MACS-treated samples we derived Vca + DEHD = (98.8 ± 0.5)% and (99.2 ± 1.1)%, respectively. The close proximity to the expected value of 100% indicates adequate data analysis. Concerning the procedure for cell accumulation, our measurements revealed that the viability of monocytes is slightly higher when sorted by MACS compared to FACS.
Uncertainty determination is essential to judge comparison of flow cytometric and microscopic viability determination. Besides statistical contributions derived from the number of counted events, for flow cytometric measurements a significant contribution to the uncertainty originates from the delineation of different populations in the scatterplots. For instance, the threshold UG can be shifted leading to different values for Vca and DEHD. To estimate this contribution to uncertainty, data analysis was also carried out using additional gating thresholds U and U (Fig. 2a), symmetrically chosen with respect to UG according to U / UG = UG / U = 0.75. It should be noted that this selection also covers extreme values observed in all measurements discussed here. Besides selection of UG, two additional thresholds were set to define the quadrants in Figure 2b and the contribution to the overall error can be determined using a variation of those thresholds in just the same way as described for the scattering signal. This approach was tested for a subset of nine measurements, showing that the uncertainty contribution caused by setting the fluorescence gate was on average approximately twice as large (median 2.2) as the contribution from varying the threshold for light scattering. It was further observed that the light scattering and fluorescence signal are correlated in some cases whereas in some cases they were not. Therefore, we decided not to use the standard uncertainty propagation, that is, taking the square root of the sum of squared errors. Instead the maximal error (linear sum of individual contributions) was used to estimate the overall uncertainty. Since this maximal uncertainty was typically three times larger than the one derived from light scattering alone, the combined uncertainties arising from setting gates for fluorescence and scattering were determined by multiplying the uncertainty estimated by the variation of UG by a factor of 3. It should be noted that the contribution from counting statistics was found to be of minor importance for flow cytometric measurements.
Evaluation of Cell Viability Using Wide-Field Microscopy
Automated cell counting in microscopic images is typically performed by binary masks for object definition (34, 35). An RGB image was recorded (Fig. 3), where the channels R and G represent the EHD and calcein fluorescence, respectively, and B gives the bright field image. Two binary masks were generated, one of which using the image obtained by calculating the average of R, G, and the inverted bright field image B*. The second mask was derived from B* only (Fig. 3). Cells appear mostly as disks in the averaged image and in G, but occasionally as rings in B* and R. In both cases, rings were obtained for all cells by Sobel operation using manual thresholding to generate binary images. Noise emerging from the Sobel operation was reduced by excluding objects with less or equal 10 pixels. Rings were closed by single octagon (radius 2) expansion and dilatation to obtain the binary masks for automated total cell number determination.
Figure 3. Image analysis based on mask derived from the inverted bright field image. For illustration, only a small fraction of the original image is shown (monocytes obtained by FACS). Grey scale and RGB images are scaled for better visibility. [Color figure can be viewed in the online issue which is available at wileyonlinelibrary.com.]
Download figure to PowerPoint
Figure 4. Gating for microscopic determination of cell viability (same sample as Fig. 1). The histogram of the cell area measured by microscopy shows the major peak for single cells and smaller peaks for clusters of two or more cells, which were not separated by image analysis.
Download figure to PowerPoint
The procedure applied to determine cell viability was consistent with the analysis of flow cytometric data and allows an adequate comparison of both methods. The fluorescence intensities for calcein and EHD were obtained by summing intensities of all pixels attributed to a specific cell. Compared to flow cytometry, the relative number of events in region III (Fig. 2b) was higher and exceeded 1% in some measurements due to cell clusters interpreted as single objects. Cell clustering limits the applicability of microscopy, in particular at high density of cells. Hence, microscopic images showing excessive clustering of cells were excluded from the analysis. Combined measurement uncertainties were derived in the same manner as described for flow cytometric data analysis.
Comparison of Cell Counting Methods
The fractions of live (Vca = RI + RIII) and dead (DEHD = RII + RIII) monocytes (blue symbols) as well as CD34-positive cells (red symbols) are plotted in Figure 5. The straight line indicates equal results for flow cytometry and microscopy. It follows from Figures 5a and 5b that the results agree well for calcein-positive and for EHD-positive cells. The uncertainties included are composed of statistical and systematic uncertainties due to gating of the target cells, as described above. However, several cases are observed where the deviation of the measured values from the straight line is larger than the estimated uncertainties. This could be attributed to additional influence quantities, like cell loss due to adhesion, which would predominantly take place in the tubing system of the flow cytometer, and the clustering of cells in microscopic measurements. Microscopic investigation indicated that live monocytes exhibit pronounced adhesion in contrast to dead monocytes. Such correlation could cause higher losses of live monocytes and hence the measured percentage of viable monocytes could be reduced. Selection of gating strategies based on various parameters also contributes to uncertainty of measurement. For example, the number of events misidentified as target cells is different when gating is based on size-correlated measuring quantities like FSC, SSC, and area in microscopy. Typically, we found differences in the range of 1–2% for flow cytometric data when analyzing fluorescence of CD14-positive target cells compared to size-correlated parameters.
Figure 5. Comparison of relative cell abundance measured by flow cytometry and microscopy for CD14-positive cells (monocytes)—blue, and CD34-positive cells—red: (a) live staining of cells with calcein; (b) dead staining of cells with EHD. The dashed line indicates equal viability. [Color figure can be viewed in the online issue which is available at wileyonlinelibrary.com.]
Download figure to PowerPoint
The CD34-positive target cells were enriched by two steps, that is, by applying density gradient followed by MACS. Although the error bars plotted in Figure 5 were estimated in the same way as for the measurements on CD14-positive cells, their size seems to be overestimated. This can be explained as follows. The delineation of target cells by cell size in microscopy and light scattering in flow cytometry is difficult, since circulating stem cells are not particularly large compared to other blood cells. This leads to a substantially higher contribution to the uncertainty even if the delineation of live and dead cells poses no particular difficulty.
In order to extend the range covered by our analysis to low viability, two sets of samples with artificially reduced viability were prepared by storage for 24 and 48 h (23°C, in the dark), respectively, before live/dead staining and subsequent measurement by flow cytometry and microscopy. Figure 5 reveals that viability of CD34 cells measured by microscope and flow cytometer differs by up to 13%. Whereas for CD14 cell viabilities above 95% were reached, for CD34 the maximum percentage of living cells is 75% (Fig. 5), possibly due to contamination by other dead cells.