CHAPTER 35
Standardization for Flow Cytometry A. Schwartz* and Emma Fernandez-Repollett * Flow Cytometry
Standards Corporation San Juan, Puerto Rico 00919
t Department of Pharmacology School of Medicine University of Puerto Rico San Juan, Puerto Rico 00936
I. Introduction 11. Considerations for Qualitative Standardization A. Theoretical Considerations B. Practical Considerations 111. Considerations for Quantitative Standardization A. Quantitative Standardization of Light Scatter Data B. Quantitative Fluorescence Intensity C . Quantitative Antibody Binding Capacity IV. Conclusions References
I. Introduction Flow cytometry provides a unique tool for the quantitative analysis of individual particles or cells. To perform this in a reliable manner, appropriate standards are necessary. The emphasis in standardizing flow cytometry has been limited to obtaining the percentage of cells in a population that binds a specific antibody. This has been beneficial for determining both the percentage and absolute number of CD4 lymphocytes in peripheral blood in conjunction with the diagnosis of the onset of AIDS and its therapy. Many factors, however, affect this measurement, ranging from the acquisition and preparation of the sample to METHODS IN CELL BIOLOGY. VOL. 42 Copyrighr 0 1994 by Academic Press, Inc.. All righrr of reproducrion
in
any form rerervcd
605
606
A. Schwartz and E. Fernandez-Repollet
the gating and analysis, to ensure that all populations of the lymphocytes have been accounted for. The effects of these factors on flow cytometric analysis, as well as commercially available standards have been extensively discussed in the recent literature (Horan et al., 1990; National Committee for Clinical Laboratory Standards, 1992; Muirhead, 1992). This chapter focuses on the comprehensive perspective of standardization, specifically identifying cell population patterns, quantitating fluorescence intensities, and determining the antibody binding capacity. Attempts at standardizing these parameters have involved descriptions of weak or strong light scatter to describe size and degree of granularity, as well as references to dim or strong fluorescence intensities. Unfortunately, without standardizing the performance of the flow cytometers, inter-instrument comparisons of data on these basis are, at best, rather inconclusive and the innate potential of these instruments has not been utilized. The need for developing effective standardization procedures is magnified by the dramatic increase in commercial and custom-built flow cytometers which collect four, five, or more parameters. The sensitivity and accuracy in comparing cell population patterns over these parameters require that proper standards are available to normalize the performance of the instruments. In turn, the standards need to have characteristics which closely resemble the samples being analyzed, including apparent size, granularity, fluorescence spectra, and intensity (Schwartz, 1988). It is our intent to review the status of flow cytometry standardization and suggest several new standardization procedures.
11. Considerations for Qualitative Standardization To avoid confusion in our discussion of standardization, “channel”-related terms need to be defined. Fluorescence channels: the optical and electronic section of the flow cytometer which processes signals from specific portions of the spectra, e.g., FLl and FL2, or FITC channel and PE channel. Histogram channels: the linear scaler divisions into which the number of events are stored in list-mode data. These are geometric factors of 2, e.g., 64, 256, and 1024. Histogram channels are often displayed as a linear output of the data. Relative channels: the continuum of linear channels obtained by factoring in gain settings to extend the range covered by linear amplifiers. Relative linear channels: the conversion of histogram channels into a log scale, usually to the base 10, to correspond to the number of decades covered by the log amplifier. This log scale of relative linear channels may also be found on the graphic output displays of data.
35. Standardization for Flow Cytometry
607
A. Theoretical Considerations
1. Sample Space Comparison of cell population patterns may be aided by considering that the sample will have as many dimensions in space as the number of parameters being measured by the flow cytometer. The sample space contains the characteristics of a cell population such as forward (FALS) and right (RALS) angle light scatter, as well as its fluorescent properties. Therefore, when using two fluorescent probes, e.g., fluorescein isothiocynate (FITC) and phycoerythrin (PE), along with the FALS, RALS, the cell population patterns can be described in four-dimensional space. Adding a third fluorescent probe would increase the sample space to five dimensions. Attempting to visualize five-dimensional sample space is a demanding task even for the most abstract thinkers. Although some software is available where three dimensions are presented on the sides of a cube, the usual presentation consists of a series of two-dimensional squares. These two dimensional presentations are usually referred to as dot plots or bivalent histograms. Improvements in software presentation, in which coloration of a particular population is consistent throughout its presentation across the parameters, have greatly helped this visualization task (Loken et al., 1991; Bagwell, 1989).
2. Window of Analysis The concept of a window of analysis is presented to help simplify standarization of the multidimensional flow cytometry data. A window of analysis is the range which is covered by the 256 or 1024 histogram channels of the various parameters of the flow cytometer. It can also be defined as the area of the sample space that can be viewed at particular instrument settings. Consider the bivarient or even single variant histograms of a window of analysis when using a log amplifier. The position of the “window” is dependent on the instrument setting (i.e., PMT or gain settings). Remember, that changing a PMT or amplifier gain does not change the sample properties, but merely how the instrument perceives the sample. It just repositions the window relative to the sample emissions. For example, when using a log amplifier, increasing the PMT voltage appears to move the sample distribution to the right. Some may mistakenly believe that this represents an increase in the fluorescence intensity of the sample; however, the sample is unchanged, and the window of analysis is simply being shifted to the left as illustrated in Figs. l a and Ib. Note that the shape and relative positions of the distributions do not change when using a log amplifier. The range of sample space covered by the window of analysis when using a linear amplifier is very limited, relative to a log amplifier. The use of relative channels, i.e., (channel number) x (maximum gain)/actual gain, can extend the
608
A. Schwartz and E. FernBndez-Repollet
100
101
102
103
Fluorescence 1
lo0
101
102
103
104
Fluorescence 1
Fig. 1 FLI window of analysis using a log amplifier showing six microbead populations where (a) the PMT is set at 600 V and (b) the PMT is set at 700 V. Note that all microbead populations remain in their relative positions and the window of analysis shifts to the left with the increase of PMT voltage.
range of sample space into a linear continuum. With a linear amplifier, a population distribution appears to widen, the further to the right it is positioned in the window of analysis, even though the sample does not change. With an increase of PMT voltage or amplifier gain, the zero point remains fixed and the sample space is stretched to the right with the higher intensities going off scale, as shown in Figs 2a and 2b.
35. Standardization for Flow Cytometry
609
258
-am
8
5 0
200
400
600
800
lo00
800
1000
Fluorescence 1
0
200
400
600
Fluorescence 1 Fig. 2 FLl window of analysis using a linear amplifier (a) at a gain setting of 1.0 showing five fluorescent microbead populations (1-5) and (b) at a gain setting of 9.99 showing four microbead populations (0-3). Note that the (0) microbead is up against the y-axis and cannot be seen in (a); however, it is shown in (b) because the amplifier “stretched out” the sample space near the origin at high gain settings.
3. Defining the Window of Analysis A window of analysis can be defined by two parameters: ( 1 ) a reference point within the window and (2) the window’s dimensions. The reference point in the window may be defined by a target channel. The dimensions of the window are related to the slope of the calibration lines of the respective parameters (e.g., FALS, RALS, and FLl).
610
A. Schwartz and E. Fernandez-Repollet
The reference material, which matches the properties of the samples, is placed in the target channel by adjusting the instrument settings, usually PMT voltages. In the case of fluorescence measurements, the spectra of the reference material must match the spectra of the sample for the relationship to hold among different instruments. The dimensions of the window of analysis are related to the coefficients of response, i.e., the slopes of the calibration lines, of the specific parameters defining the window. As illustrated in Fig. 3, the range of sample space which the histogram channels display is determined by the slope of the calibration line or the projection of the calibration line onto the axis. It should be noticed that a bivarient histogram has two independent dimensions for the window of analysis and the slope of each paramater needs to be determined independently. B. Practical Considerations 1. Qualitative Standardization of Light Scatter Data The size of a cell or particle is related to the FALS signal, whereas its granularity is related to the RALS signal. Since the major components of peripheral blood, i.e., lymphocytes, monocytes, and granulocytes, each have different size and granularity, they can easily be distinguished using FALS and RALS on most commercial flow cytometers. However, it is difficult to obtain complete separation of these populations using only light scatter data. For example, large lymphocytes and degranulated neutrophils may occur in the monocyte region, whereas small lymphocytes may occur where erythrocytes and debris appear.
labeled I
I I
Projection of Slope A Projection of Slope B I
I
FL1 Sample Space
Fig. 3 An illustration of the ranges in samples space covered by two windows of analysis which have different slopes, A and B, due to having a different number of log decades, four and three, respectively, for their amplifiers. The slopes are represented by equal lengths of the 1024 histogram channels.
61 1
35. Standardization for Flow Cytometry
Such overlaps can cause significant errors in quantitating the percentage of subpopulations of cells labeled with a specific antibody. Corrective statistics regarding purity of the light scatter gate should be included in the report. a. Reference Point Comparisons of the cluster patterns of leukocytes require a common reference point in the light scatter window of analysis. A population of highly uniform microbeads may serve this purpose since their light scatter properties are not significantly affected by the aqueous solution in which they are suspended. Since polymeric microbeads have a different refractive index than biological cells (Lakowicz, 1983), they will have different light scatter properties than cells, especially RALS, even if they have the same physical size, and thus will usually appear as a distinct population providing a reference point.
b. Eflect of Instrument Diflerences Light scatter results are very sensitive to the configuration of the instrument optics. Significant differences in the position of cell populations occur in the light scatter window of analysis among different makes and models. Figures 4a and 4b depict the appearance of microbeads mixed with whole normal peripheral blood lysed with the same lysing solution, but analyzed on two different instruments. Note the significant difference of the forward angle light scatter profile of each of the populations on the different instruments.
. . ........ ...... . .\...... ....
.... ..... .....
.
. .
..
.. . .
. I .
..
2
....
.... .....
,
Microbeads
2 5-.'
FALS
I
;.&Lymphocytes
FALS
Fig. 4 Light scatter windows of analysis showing the relative positions of polymeric microbeads and leukocytes in whole lysed blood in (a) the Profile I1 (Coulter Electronics Inc.) and (b) the FACScan (Becton-Dickinson lmmunocytometry Systems, Inc.) flow cytometers. Note, that all leukocyte populations look smaller, i.e.. lower FALS, on the Profile I1 than the FACScan, relative to the microbeads.
612
A. Schwartz and E. Fernandez-Repollet
c. Eflect of Lysing Solution
The degree of swelling induced by different lysing solutions on a given cell population is also a factor which affects the relative positions of different cell populations in the light scatter window of analysis. The effect of different lysing solutions, i.e., FACS lyse (BDIS) and Coulter lyse (Coulter Electronics) on the light scatter profile of normal whole peripheral human blood was compared using the same flow cytometer (Schwartz and Fernhdez-Repollet, 1991). Figures 5a and 5b illustrate the variability in light scatter properties, due primarily to the swelling response of the cells to a particular lysing solution. This finding is in agreement with that of Carter and collaborators (Carter et af., 1992). It is evident from the above discussion that the goal of obtaining a common window of analysis for light scatter data is extremely difficult to achieve, especially with the current status of having various optical configurations in the different commercial flow cytometers and a variety of lysing solutions in use by the different laboratories. Qualitative standardization of light scatter where cluster patterns fall in specific positions in a common window of analysis will require an organized effort within the flow cytometry community, combined with focused cooperation among the manufacturers.
2. Qualitative Standardization of Fluorescence Intensity Data Currently, the majority of the fluorescence data analysis obtained from a flow cytometer involves the use of region makers to determine the percentage of positive labeled cells in a population. As analysis and interpretation of flow
FALS
FALS
Fig. 5 Light scatter windows of analysis (FACScan, BDIS) showing the relative positions of
polymeric microbeads and leukocytes in whole blood when lysed with (a) BDIS lysing solution and (b) Coulter lysing solution. Note that the granulocytes appear “shrunken” and the lymphocytes and monocytes appear “swollen” when lysed with Coulter lysing solution as compared to those lysed with BDIS lysing solution, when using the microbeads as a reference point.
35. Standardization for Flow Cytometry
613
cytometry data progress, more emphasis is being placed on cluster pattern recognition. This has proven to be especially important in the diagnosis, therapy monitoring, and prognosis of bone marrow-related disorders (Terstappen ef al., 1988), as well as activation of subpopulations in peripheral blood (Terstappen et al., 1990). a . Fluorescence Window of Analysis The key to standardizing qualitative fluorescence intensity data is obtaining common windows of analysis for the fluorescence sample space. As previously mentioned, this can be accomplished by having a common reference point, e.g., target channels, and having the response coefficients (i.e., the slopes of the calibration lines) be the same for all instruments. Fortunately, the differences in design of flow cytometers with respect to their fluorescence measurements is in some ways less critical than that of the light scatter measurements (Shapiro, 1988). Moreover, the barrier filters, PMTs, and amplifiers of commercial instruments are rather similar, often coming from the same sources. These factors, together with the availability of proper standards to establish normalized target channels, permit qualitative standardization of fluorescence intensity data and direct comparison of cell cluster position data among instruments. b . Fluorescence Target Channels When considering individual instruments and if the comparison of data among instruments is disregarded, any highly uniform fluorescent material (e.g., fixed chicken red blood cells or fluorescent microbeads of nonspecified spectra) may be employed as the reference material. However, when fluorescence data need to be compared among different instruments, a common window of analysis is essential and, consequently, the fluorescent reference material must have excitation and emission spectra which match those of the labeled samples. With matching spectra, the relative positions of the reference material and labeled cells are fixed in sample space. Therefore, a particle labeled with both fluorescein and phycoerythrin would be a suitable reference material to set target channels for samples labeled with the same two fluorochromes, as shown in Fig. 6. c. Noise Level As with other kinds of laboratory instruments, the noise level of a flow cytometer is the limit of detection of a real sample signal. The fluorescence threshold is equivalent to the noise level and nonfluorescent particles may be used for its determination. When the flow cytometer is set to be triggered by forward angle light scatter, it will also collect data in all other channels when a particle passes through the laser beam. If the particle is nonfluorescent, then the readings in the fluorescence channels represent only the noise of that particular channel. Noise may arise from both optical (fluorescent or scattered light from the laser, mirrors, filters, lenses, etc.) and electronic sources (PMTs,
614
A. Schwartz and E. Fernandez-Repollet
104
0"
! -
'
1200 n '
I
102:
1600 ' " I
I
.:
"
'
100
L
I
CDCPE
.
" I 1400
+
. '
I I I
-I I
102
104
Fluorescence 1 Fig. 6 Illustration of the relationships between unstained cells, FITC- and PE-labeled cells, and
FITUPE-labeled microbeads after positioning the labeled microbeads in their respective target channels in the window of analysis. Instruments using this method of setup will have a common window of analysis with cells appearing in comparable positions.
amplifiers, power supply, etc.). Specific sources of noise may be isolated by feeding electronic signals into the circuits, but the aggregate noise signal can only be determined by analyzing nonfluorescent particles which combines all noise sources, including scatter, that would be present when analyzing actual samples. In qualitative terms, knowing the position of the noise level with respect to the unlabeled cells being assayed is extremely important since noise can mask the detection of low-level antigen expression. As long as the fluorescence level of the unlabeled cells is higher than the noise level, then the noise cannot interfere with the assay. On the contrary, when the noise level of the instrument is equal or greater than the autofluorescence of the cells, then the autofluorescence of the cells will be masked by the noise and they will appear in the same location as the nonfluorescent particles. Thus, the noise level needs to be determined in a quantitative manner and compared to the known values for that particular cell, as described below. It is best to use unlabeled cells for this "sensitivity check" rather than isotypic controls since the lowest fluorescence intensity that a cell can exhibit will be evident before exposure to any fluorescent antibody. Furthermore, if it is shown
35. Standardization for Flow Cytometry
615
that the instrument noise level is below the fluorescence of the unlabeled cell, then it follows that the instrument can detect the real autofluorescence signal of the cell, and, in turn, an approximation of the degree of nonspecific binding from the isotypic controls can be made. To ensure that the noise level evaluation is meaningful, it is essential that the nonfluorescent particles used in the test are known to have a lower fluorescence than the unlabeled cells in all the fluorescent channels being used. Polymeric microbeads are ideal particles for this application; however, many such “unlabeled” microbeads contain impurities or are made of materials which are autofluorescent, and thus are not suitable for this application. Therefore, the fluorescent levels of the unlabeled microbeads must be “certified” to be lower than the autofluorescence of the specific unstained cells being analyzed. The noise level of a flow cytometer has an additional significance with respect to the compensation circuits. These circuits are designed to subtract a percentage of the signal in the secondary channels as determined from the intensity in the primary channel. For example, fluorescein signals are subtracted from the FL2 channel as per the expression FL2-%FLl. A current practice in setting the compensation circuits is to place the unstained cell population in the lower left-hand corner of the window of analysis and use cells labeled with antibodies conjugated with specific fluorochromes, e.g., FITC and PE, to adjust these populations such that they are orthogonal to each other with respect to the unstained cell population. However, often populations which are dimmer than the population used to adjust the circuit are found to be overcompensated, whereas populations which are brighter are found to be undercompensated. This indicates that the instrument is not correctly compensated because the noise level is not being properly considered. This is represented graphically in Fig. 7. In this representation, the line of compensation described by expressions like FL2-%FLl can be described by a line that pivots around the point of zero fluorescence, which can be approximated by the noise level. When the circuit is properly adjusted, the compensation line will be parallel to the axes of the window of analysis. This yields a correct compensation across the intensity range. Since cells are autofluorescent, the position of the cells will be offset, respectively, when the instrument is adjusted such that the noise level is on scale. This setup has the additional advantage of knowing when overcompensation is present since the populations are not “crammed” against the axes. This noise check has an additional application in determining whether the offset voltages of the amplifiers are properly adjusted. For this test, the compensation circuit is turned off to ensure that the unstained cells have a higher fluorescence than the nonfluorescent particles. Then, the compensation circuits are turn on and the particles and cells are reanalyzed. If the relationship remains such that the cells have a higher fluorescence than the particles, the pivot point is in the correct position and the instrument is performing properly. However, if the particles now appear more fluorescent than the cells, then the pivot point
616
A. Schwartz and E. Fernandez-Repollet
Blank Microbeads
Fig. 7 Illustration of how compensation is set incorrectly by placing the unstained cell population in the corner of the window of analysis. Note that although the compensation appears to be correct for population (B),relative to the unstained cells, the other populations are not compensated correctly. If the blank microbeads were place on scale and the compensation readjusted. then all populations would be parallel to the axis and uniformly compensated across the intensity range.
is not in the correct position, indicating that the offset voltages of the amplifiers require adjustment. In summary, detection of the noise level can be used to determine the limit of detection with respect to an assay, as well as to set correct compensation. d . Instrument Setup and the Use of Appropriate Fluorescent Conjugates One of the major notions working against standardization for flow cytometry is the perception that individual instrument setups are needed for each type of assay. This is fostered by the common practice of placing the unlabeled or isotype controls in the corner of the window of analysis allowing the autofluorescence and/or nonspecific binding of isotype controls to dictate the instrument settings. This approach to instrument setup is responsible for the loss of important information, namely, determination of the level of autofluorescence and the level of nonspecific antibody binding. However, if the instrument is using a log amplifier and it is a setup such that the noise level is just on scale, then this setup will be applicable for any assay. It is thought by some that the ability to detect a fluorescent signal (i.e., sensitivity) can be enhanced by raising the PMT voltages. However, with log amplifiers, the noise level is, for the most part, as stated before, also fixed in sample space and only the window of analysis is shifted by PMT adjustment. In other words, since fluorescence levels cannot be measured below the noise level, changing the PMT will not improve the measurements.
35. Standardization for Flow Cytometry
617
Objections are raised that if the noise level is on scale, then the cells with high antigen expression will be off scale. This problem can be solved easily by the proper selection of the fluorochrome conjugated to a particular marker antibody. For example, CD8-PE-labeled lymphocytes do fall into the fourth decade of many instruments due to the large number of CD8 binding sites and the intense brightness of PE (i.e., high extinction coefficient). In contrast, CD4FITC-labeled cells barely fall into the third decade because there are fewer CD4 binding sites and FITC has a lower extinction coefficient. These positions are “normalized” by labeling with CD4-PE and CD8-FITC (Stewart, 1990). By using such labeling strategies, a single common window of analysis may be obtained for all instruments where the noise level, as well as the labeled populations, remains on scale. e. Practical Applications of the Window of Analysis
In general terms, the hardware of most commercial flow cytometers is very similar. For example, most instruments use PMTs and linear and log amplifiers to detect and process signals from samples. The raw data (list-mode files) are stored and analyzed as histogram channels. However, differences in data begin to surface depending on the resolution of the scalar histogram channels used, i.e., 64,256, 1024, or 4096, and how many decades the log amplifier covers. Significant efforts from the flow cytometer manufacturers and third-party sources have led to the development of software that analyzes and presents data in a more meaningful manner. Unfortunately, these efforts have not been conducted in a coordinated manner among the manufacturers, resulting in a significant problem with respect to standardizing the presentation of flow cytometry analysis. Data processed from software programs may be displayed as histogram channels (HC), usually 64,256, or 1024 channels on a linear scale, or they may be presented as relative linear channels (RLC), which are displayed on a log scale if the log amplifier covers 3, 3.5, or 4 decades, which is specific to the make and model of the flow cytometer. It becomes clear that comparison of data using unspecified channels numbers, even if all instruments performed exactly the same, would be extremely difficult when using different software. The development of software is a major task and is usually approached by a particular philosophy which in turn has particular tradeoffs. For example, some manufacturers want the program to be as flexible as possible, which requires greater understanding by the user of what the program does and how the results are presented. For instance, the user can have various options as to the output scale of histograms, e.g., 256 or 1024 histogram channels (linear scale) or 10,000 relative linear channels (log scale), as is the case with much of Becton-Dickinson Immunocytometry System acquisition and analysis software. Such flexibility makes standardization within even the same instrument model more difficult and complicated.
0;
w
B 0
k
0
5 0
n
a3
b t
o m
o
u
, JaqLlJnN
c
r
4
’
f
J
U
35. Standardization for Flow Cytometry
619
Another approach to software is to have all output scales the same, e.g., 10,000 RLC, but this is still not free of confusion because the output scales may start at different points. For example, Becton-Dickinson Immunocytometry System, Inc., and Ortho Diagnostic Systems, Inc., start their relative linear scales at channel 1 (i.e., the log to the base 10 of channel O), whereas Coulter Electronics, Inc., starts their RLC scales at 0.1. Knowledge of these differences and their effects on flow cytometry data analysis is essential when data comparison is required. It is evident that significant progress in standardization of flow cytometry interpretation can be made if just this one area is addressed. To examine how the concept of a common window of analysis can be applied in comparing flow cytometry data obtained with different instruments and acquisition software, the following experiment was conducted by G. Stelzer from Cytometry Associates. Four different models of flow cytometers manufactured by three different companies were used in this study. Attempts were made to limit sources of variables other than the instrumentation, i.e., the same reference standard, labeled cell samples, and list-mode analysis software were used with each instrument. Normal human peripheral whole blood was prepared by labeling with CD8-FITC and CDCPE monoclonal antibodies, followed by lysing and washing. With the compensation circuits off, QC3 reference microbeads (Flow Cytometry Standards Corp.) labeled with FITC and PE were placed in initial target channels (normalized to the acquisition output scale of each instrument). The labeled cells were then run on each instrument and the compensation circuits adjusted. All the list-mode files from each instrument were analyzed with the same analysis program (WinList from Verity Software, Inc.). Figure 8 shows where each of the labeled cell populations appears in the common window of analysis. As can be seen, the positions of both the labeled and unlabeled populations in the one-dimensional window of analysis are quite comparable even though the output scales and amplifiers differ among the instruments.
111. Considerations for Quantitative Standardization A. Quantitative Standardization of Light Scatter Data
Of the two measurements derived from light scatter signals, determination of cell size can be accomplished in a straightforward manner in the FALS channel with the proper use of matched standards. A set of particles of different Fig. 8 Histograms showing immunophenotyping data obtained from four different flow cytometers (Scan, BDIS FACScan; Profile, Coulter Profile 11; Cytoron, Ortho Cytoron Absolute; and Star, BDIS FACStar plus). Note that each of the cell populations labeled with CD8-FITC and CD4-PE, respectively, fall in the same positions when a common window of analysis is used for the different instruments.
620
A. Schwartz and E. Fernandez-Repollet
sizes, which are made from the same material, may serve as light scatter sizing standards. Size-calibrated populations of microbeads synthesized from the same polymer which have the same refractive index and high uniformity may serve as such standards. These standards should be used to calibrate and determine the linearity of the response of the FALS channel. It may be noted that polymeric microbeads, e.g., 5 p m in diameter, appear to have the same FALS signal as cells which are larger, e.g., lymphocytes which are 8 pm in diameter. This is because the refractive index of cells is different from that of the microbeads (Shapiro, 1988). Correction factors are required to obtain accurate size measurements with FALS. Such corrections may be achieved by shifting the calibration plot while maintaining the slope generated by a cell population of known size (e.g., determined by scanning electron microscopy), until it falls on the proper position of the calibration plot. Quantitative standardization and interpretation of RALS data are very difficult and complex subjects which are beyond the scope of this discussion. At present, quantitative granularity standards for such calibrations are unavailable. It is hoped that future efforts will provide such standardization tools. B. Quantitative Fluorescence Intensity Quantitative fluorescence measurements have the ability to go beyond determining the percentage of positively labeled cells in a particular population. In fact, such measurements are based on relative fluorescence intensities to determine labeled and unlabeled cells. The linear signal to channel response of flow cytometers in the fluorescence channels provides the means to make direct intensity measurements. This requires a unit of fluorescence intensity which is related to the particular fluorochrome and independent of the instrument.
1. MESF Definition and Applications Expressing fluorescence intensity directly in terms of fluorochrome molecules would seem to be a logical approach, but it has a number of pitfalls. They include reduction of intensity as a result of quenching, e.g., energy transfer between close molecules, as well as changes in extinction coefficient and quantum efficiency due to binding and microenvironment effects. These factors can exert even a greater influence when determining the fluorescence intensity of particles, such as biological cells or microbeads. A more consistent determination of the fluorescence intensity of particles can be made by comparing their intensity to that of the soluble fluorochrome under the same environment conditions which can be readily controlled and reproduced in the laboratory. These comparisons will have a constant quantitative relationship as long as both the excitation and emission spectra of the solution and labeled particle are the same. Therefore, a set of particles labeled with a specific fluorochrome whose spectra match a solution of the fluorochrome
35. Standardization for Flow Cytometry
623
can serve as calibrators for cells labeled with the same fluorochrome when suspended in the same medium. A convenient way of expressing a unit of fluorescence intensity is in molecules of equivalent soluble fluorochrome (MESF). When applying this unit, the specific fluorochrome must be indicated since, for example, 50,000 MESF of fluorescein is not equivalent to 50,000 MESF of phycoerythrin.
2. Calibration of Fluorescence Intensity Calibrating the fluorescence channels of a flow cytometer involves determining the instrument response to specific fluorescent signals across the entire fluorescence range. Therefore, when analyzing FITC- or PE-labeled cells, the calibrations must be expressed in units, e.g., MESF units, of the respective fluorochromes. Because of the high uniformity, sets of fluorescent microbeads labeled with these fluorochromes and having preassigned MESF values can serve as a convenient calibrator for flow cytometers, as well as fluorescent microscopes. By plotting the peak channels of these microbeads against their MESF values (or performing a linear regression) a calibration line is obtained. With such calibration lines, the intensity of labeled cell populations can be determined by finding the corresponding MESF values for the peak channel of the cell distribution, as shown in Fig. 9. Many laboratories have incorporated this methodology in their daily calibration procedure to assure reproducibility in instrument performance (Fay et al., 1991; Schols et af., 1990; Bohmer et af., 1992).
3. Evaluation of Instrument Performance There is far more to evaluating the performance of a flow cytometer than merely minimizing the coefficient of variation (CV) for a particular population of microbeads. This will only assess the alignment and focus of the optical components. A more complete instrument evaluation will require information regarding the linearity, resolution, and noise level. This cannot be determined with a single population of microbeads but requires a series of quantitative microbead standards. After plotting a calibration line of MESF values vs peak channels of the quantitative microbead standards, the linearity of the fluorescence response is determined by calculating the coefficient of determination (r’). Care should be taken in interpreting this value since the coefficient of determination is very insensitive to log data and most fluorescence data are obtained using a log amplifier. For example, a r2 value of 0.95 may be quite acceptable with linear data, whereas, it takes a value of 0.99 to be acceptable for log data. A better measure of linearity for fluorescence log data is the average residual percent (AvRes%) which is the root mean square difference of where the points fall relative to the regression line. One set of suggested criteria for AvRes% of flow
622
A. Schwartz and E. Fernandez-Repollet 200
400
600
1000
800
1
I I
I
I Non-specific
100
10'
102
103
104
Fluorescence 1 Fig. 9 Illustration of the relationship between the quantitative microbead standards and the intensities of cells as determined from the calibration line. Note that the MESF or ABC value of the noise level (a) of the instrument can be determined from the position of the blank microbead on the calibration line. In addition, both the level of autofluorescencehonspecificbinding (b) and specific binding (c) can be determined from the calibration line.
35. Standardization for Flow Cytometry
623
cytometers currently under investigation is the following: acceptable 3%. These criteria appear to have held up in a number of instrument performance surveys (Ehman, 1992; Vogt et al., 1991). Resolution is a term used to describe the smallest difference detectable between two signals of the same properties, e.g., two cells labeled with FITC. It is not accurate to describe resolution between an autofluorescent cell and a cell labeled with FITC. Resolution must be evaluated in several portions of the window of analysis when using a log amplifier. In the lower decade, a difference of two histogram channels may only represent 50 MESF units, whereas, in the fourth decade, two histogram channels may represent 5000 MESF units. A qualitative assessment of resolution may be made by examining the separation among the populations of the quantitative microbead standards. The quantitative noise level of the fluorescence channels is also assessed with the calibration line by determining the peak channel of a nonfluorescent reference material and finding its corresponding MESF value (Fig. 9). This value allows direct comparison of instrument noise levels. In regard to the assay, the noise level should be lower than the autofluorescence MESF values of the unstained cells. The calibration line also provides important information with respect to the window of analysis, e.g., the response coefficients (slopes) of the calibration lines are directly related to the size and shape of the window of analysis (Fig. 3). By comparing the slopes of specific fluorescence channels among instruments, size comparisons of their respective windows of analysis can be made directly. Therefore, a comprehensive evaluation of instrument performance not only requires a low CV on alignment microbeads to ensure the instrument is in alignment, but also an evaluation of the linearity, resolution, and noise level of each fluorescence channel. C. Quantitative Antibody Binding Capacity
The ultimate goal for flow cytometry is the determination of the number of antibodies binding to specific cell populations. The key to measuring the number of binding antigens on a cell using fluorescent antibodies resides in quantitating the fluorescence intensity of the particular cell and translating that intensity into the number of bound antibodies. Quantitation of fluorescence intensity is an indirect way of making this measurement since the conjugated antibodies have an average fluorescence. If this average antibody fluorescence could be measured in MESF units, using the calibration plots previously described, then the number of antibodies could be determined by dividing the total MESF intensity of the cell population of the average MESF intensity per antibody. Note that the MESF intensity, or “effective FIP,” of an antibody is not the same as the FIP ratio which is the actual number of fluorochromes bound to
624
A. Schwartz and E. Fernandez-Repollet
an antibody. The FIP ratio does not directly translate into the fluorescent intensity associated with the cell because of environmental conditions such as dye ionizability, pH, and quenching. The effective FIP ratio is determined by measuring the fluorescence of the antibody, whereas the FIP ratio is measured by absorbance (Schwartz, 1988; Hoffman et al., 1992).
1. Indirect Quantitation of Binding Antibodies An easy method to determine the effective FIP ratio of antibodies, or the average fluorescence intensity per antibody molecule, consists of saturating a microbead population which has a calibrated number of binding sites and determine its MESF intensity on a flow cytometer precalibrated with the MESF calibration standards. This methodology has been used successfully with reticulocytes (Schimenti et al., 1992), epidermal growth factor receptor expression (Lopez et al., 1992), nucleoside transporter sites (Jamieson et al., 1993), and the development of anti-OKT3 antibodies (Lim et al., 1989). 2. Direct Quantitation of Binding Antibodies A new methodology has been developed which allows the direct determination of antibody binding capacity by flow cytometry. Rather than calibrating and evaluating the instrument in terms of fluorescent units, this can be accomplished in terms of the specific antibody being measured. Calibration would be expressed in terms of binding antibodies per histogram channel. The noise level would represent the lowest detectable number of the specific antibody. This would mean that the instrument would have to be calibrated for each antibody used; however, such a calibration would take into consideration all the correction factors related to fluorescence measurements, e.g., quenching, changes in extinction coefficient, and quantum efficiency. This methodology requires a series of particle populations which are able to bind calibrated numbers of antibodies and which would have to be saturated with the antibody of interest. The peak channels of each particle population would be plotted against their respective binding capacities. Different antibodies would generate different calibration lines depending on a number of factors, e.g., the effective FIP ratio and quenching. Finally, the number of antibodies binding to a specific cell population can be directly read from the calibration line for that antibody (Figure 9). This method has the advantage of being independent not only of the instrument, but also of the conjugated fluorochrome.
IV. Conclusions The ability of flow cytometry instrumentation has far surpassed the simple determination of percentage of positive cells. These instruments have the capabilities of determining fluorescence intensities in quantitative units and even
35. Standardization for Flow Cytometry
625
measuring the number of antibodies binding to specific subpopulations of cells. However, proper standardization and calibration are required for full realization of these potential applications. Properly designed standards can provide comparable data on instrument performance, as well as assay analysis which is instrument independent and comparable over time and among all instrument makes and models. This will certainly propel clinical immunophenotyping into exciting new areas. Acknowledgments The authors thank Francis Mandy, Kathy Muirhead. Alan Landay, Robert Vogt, and Howard Shapiro for many valuable comments. Mrs. Marinelly Velilla provided excellent technical assistance. The development of the quantitative antibody binding capacity standards was supported in part by the National Cancer Institute of the NIH (SBIR Grant No. R44-CA 48570).
References Bagwell, C. B. (1989). “ModFit Program.” Verity Software House, Inc., Topsham, Maine. Bohmer, R. H., Trinkle, L . S., and Staneck, J. L. (1992). Cytometry 13, 525-531. Carter, P. H., Resto-Ruiz, S., Washington, G. C.. Ethridge, S., Palini, A., Voght, R., Waxdal, M., Fleisher, T., Noguchi, P. D., and Marti, G. E. (1992). Cytometry 3, 68-74. Ehman. D. (1992). Flow Cytometry Stand. Forum 4(1),1-5. Fay, S. P.. Posner, R. G., Swann. W. N.. and Sklar, L . A. (1991). Biochemistry 30, 50665075. Hoffman, R. A., Recktenwald. D. J., and Vogt, R. F. (1992). In “Clinical Flow Cytometry: Principles and Applications” (K. D. Bauer. R. E. Duque, and T. V. Shankey. eds.), pp. 469-477. Williams & Wilkins, Baltimore, MD. Horan, P. K.. Muirhead, K. A., and Slezak, S. E. (1990). In “Flow Cytometry and Sorting” (M. R. Melamed, T. Lindmo, and M. L . Mendelsohn, eds.), pp. 397-414. Wiley-Liss, New York. Jamieson, G . P., Brocklebank, A. M.. Snook, M. B., Sawyer, W. H., Buolamwini. J. K., Paterson, A. R. P., and Wiley, J. S. (1993). Cytometry 14,32-38. Lakowicz, J. R. (1983). “Principles of Fluorescence Microscopy.” Plenum, New York. Lim, V. L., Gumbert, M., and Garovoy, M. R. (1989). J . Immunol. Methods U1, 197-201. Loken, M. R., Civin, C. I., Shah, V. O., Fackler, M. J., Segers-Nolten, I., and Terstappen, L. W. M. M. (1991). In “Flow Cytometry in Hematology” (0. D. Laerum and R. Bjerknes, eds.), pp. 31-42. Academic Press, San Diego. Lopez, J. M., Chew, S. J., Thompson, H. W., Maker, J. S., Insler, M. S., and Beuerman, R. W. (1992). Invest. Ophthalmol. Visual Sci. 33, 2053-2062. Muirhead, K. A. (1992). In “Clinical Flow Cytometry: Principles and Applications” (K. D. Bauer, R. E. Duque, and T. V. Shankey, eds.), pp. 177-199. Williams and Wilkins, Baltimore, MD. National Committee for Clinical Laboratory Standards (1992). “Clinical Applications of Flow Cytometry: Quality Assurance and Immunophenotyping of Peripheral Blood Lymphocytes: Proposed Guideline,” NCCLS Doc. H42-T, No. 12, p. 6. Vol. 12 No. 6, Villanova, PA. Schimenti. K. J., Lacerna, K., and Wambler, A. (1992). Cytometry 13, 853-862. Schols, D., Pauwels, R., Desmyter. J., and De Clercq, E. (1990). Cytometry 11, 736-743. Schwartz, A. (1988). “Monograph: Fluorescent Microbead Standards.” Flow Cytometry Standards Corporation, Research Triangle Park, North Carolina. Schwartz, A., and Ferniindez-Repollet. E. (1991). Flow Cytometry Stand. Forum 3(2) 7-8. Shapiro, H. (1988). “Practical Flow Cytometry.” Liss, New York.
626
A. Schwartz and E. Fernandez-Repollet
Stewart, C. C. (1990). In “Methods in Cell Biology” (2. Darzynkiewicz and H. A. Crissrnan, eds.), Vol. 33, pp. 427-450. Academic Press, San Diego. Terstappen, L. W. M. M.,Shah, V. O., Civin, C. I., Hurwitz, C. A., and Loken, M. R. (1988). Cytometry 9, 471-484. Terstappen, L. W. M. M., Hollander, Z., Meiners, H., and Loken, M. R. (1990). J . Leukocyte B i d . 48, 138-148. Vogt, R. F., Cross, G. D., Phillips, D. L., Henderson, 0.. and Hannon, W. H. (1991). Cytometry 12, 525-536