Compensation in Flow Cytometry

Mario Roederer1

1 Vaccine Research Center NALID, NIH, Bethesda, Maryland
Publication Name:  Current Protocols in Cytometry
Unit Number:  Unit 1.14
DOI:  10.1002/0471142956.cy0114s22
Online Posting Date:  December, 2002
GO TO THE FULL TEXT: PDF or HTML at Wiley Online Library


Many people using multi‐color flow cytometry do not understand what compensation is, when it is needed, or how it should be applied. The author provides a clear explanation of the nature of compensation, the factors that affect compensation values, and the effect of compensation on data visualization. Both hardware and software compensation are covered. Although the subject is treated from a mathematical perspective, the manner of presentation is very straightforward. The unit discusses appropriate controls and protocols to check compensation settings, as well as introducing the concept of fluorescence‐minus‐one gates. Many example figures are supplied to give the reader a clear indication of how the author believes compensation should be handled. Keywords: flow cytometry; spectral compensation; fluorescence spillover; autofluorescence; fluorescence minus one

PDF or HTML at Wiley Online Library

Table of Contents

  • Basic Compensation
  • Pairwise Compensation on Flow Cytometers: Three‐Color Compensation
  • Complete Compensation
  • Autofluorescence
  • Resonance Energy Transfer (Tandem) Dyes
  • Compensation: Effect on Visualization of Data
  • Gating and Analysis Controls: FMO Controls
  • Compensation Controls: Cells
  • Compensation Controls: Beads
  • Factors Affecting Compensation Values
  • Compansation Myths
  • The Protocol: How to Set Compensation
  • Literature Cited
  • Figures
  • Tables
PDF or HTML at Wiley Online Library


PDF or HTML at Wiley Online Library


  •   FigureFigure 1.14.1 Can you determine which sample is appropriately compensated? In this hypothetical experiment, PBMC were stained with FITC‐conjugated anti‐CD3 and PE‐conjugated isotype control. In both (A) and (B), the top graph shows the uncompensated data. Each of the numbered sections of the lower graphs represents the data as the compensation setting (between the FITC and PE detectors) is increased. In panel B, the data are shown for the same sample, with a much lower PMT voltage for the PE detector used during the collection. Answer three questions: (1) Which sample in panel A (no. 1, 2, 3, or 4) is properly compensated? (2) On what basis did you make this decision? (3) Which sample in panel B is properly compensated? See the introductory paragraphs of the text for the answers.
  •   FigureFigure 1.14.2 Incorrectly compensated data can often still be analyzed. These graphs represent data collections of PBMC stained with CD3‐FITC and either CD4‐PE or a PE‐isotype control. The top graphs are uncompensated; the FITC into PE compensation setting is increased for each successive panel. The quadrant lines (dashed lines) represent what would be set based on an unstained sample (or a sample stained with isotype controls for both colors). The solid line represents a gate set based on the singly‐stained isotype control from the left panels. Note that the computation of the percentage of CD3 cells that express CD4 (CD3+CD4+) can be correctly performed on any of the panels irrespective of the compensation setting—but only if the gate is set based on the singly‐stained sample (left). The quadrant setting based on a complete isotype control stain (dashed lines) would result in incorrect frequencies. Only the correctly compensated sample (middle panels) shows the correct amount of CD4 fluorescence on the CD3+CD4 cells (i.e., no different than autofluorescence); the undercompensated samples have apparent CD4 fluorescence and the overcompensated samples have apparent negative fluorescence. Thus, antigen density measurements are particularly sensitive to improper compensation settings.
  •   FigureFigure 1.14.3 Fluorescence emission spectra for FITC and PE. The emission spectrum (the wavelengths of light generated by excitation of these molecules) is shown for an excitation at 488 nm (the same as the argon‐ion laser line). FITC emission is maximal at ∼515 nm; typically, a filter centered on 530 nm is used to collect the emitted light (shaded region). The emission of is farther red, with a maximum at ∼575 nm; typically, a filter centered on this emission maximum is used to collect. Note that PE has some emission in the wavelength bands used to collect PE fluorescence (B); typically, the amount of light in the 575‐nm band is ∼15% of that in the 530‐nm band (A). The PE has very little emission in the 530‐nm band (C), usually less than 2% of the emission in the 575‐nm band (D).
  •   FigureFigure 1.14.4 Emission spectra as a function of concentration (or amount) of dye. The FITC emission spectrum is shown for two samples; the right panel has twice the concentration of FITC as the left. The ratio of emissions in the two detector bands, r = B/ A, is the same irrespective of the amount of dye. After this ratio is determined for one sample, the signal B in a different sample can be determined simply by measuring A and multiplying by r.
  •   FigureFigure 1.14.5 Measurement errors lead to spreading of properly compensated distributions. This example shows data collected for PBMC stained with APC‐conjugated anti‐CD57. This antigen shows a “smear” of events from negative to very bright, leading to an excellent visualization of compensation. (A). Uncompensated distribution of the APC signal versus the APC‐Cy5.5 signal. There is considerable spectral overlap between these two channels; in addition, as they are in the far red, the number of actual photons counted is relatively small. The goal of compensation is to remove the contribution of the primary fluorochrome from the spillover channel (arrow). Note that compensation is a linear process, in which the amount to subtract from any given cell is based on the primary fluorescence value. Thus, every cell on the line of the arrow will have the same amount of fluorescence removed from it, an amount that is proportional to the primary fluorescence value (i.e., proportional to ∼500). Since the same amount has to be subtracted from every cell on the line, the vertical width of the uncompensated distribution cannot change in absolute amounts. The bottom of this distribution is at ∼700 fluorescence units, the top at ∼1100. Therefore, after compensation, the vertical width of this distribution must still be 400 units. (B) The same data, after correct compensation. Note that the vertical width of the distribution at the line is still 400 units, centered at ∼5 ± 200. All events below a value of 1 are forced onto the axis; hence the accumulation of a large number of events at the very bottom. The distribution must extend all the way up to 200. The width of this distribution is determined principally by photon‐counting statistics (see text).
  •   FigureFigure 1.14.6 Fluorescence Minus One (FMO) gates are an accurate way to identify positive versus negative events. See text for full discussion. (A) The distribution as shown in Figure is reproduced here. In these two graphs, cells were stained only with CD57‐APC. The isotype gate is that defined by an unstained (or fully isotype‐stained) sample. The 1‐D FMO gate is defined by the CD57‐APC‐stained sample, examining only the distribution of APC‐Cy5.5 and setting the threshold above all events. The 2‐D FMO gate is defined by the limit of APC‐Cy5.5 distribution when viewing this two‐dimensional graph. (B) Human PBMC were stained and gated for CD3 (not shown), CD8‐APC‐Cy7, with (right) and without (left) CCR5‐APC. The expression of CCR5 is dim, and accurate discrimination of positive and negative events is necessary. Percentages show the fraction of CD8 T cells within each “positive” gate: the isotype gate (lower dotted line), the 1‐D FMO gate (upper dotted line), and the 2‐D FMO gate (solid polygon). (C) A four‐color (single‐laser‐excited) staining combination is used to illustrate the utility of FMO gates. The goal in this illustration is only to identify the CD4 T cell population accurately. The stains used for each sample are listed in the table above the graphs. In the upper 3 graphs, the data are properly compensated. The lower 3 graphs represent exactly the same data, except that the PE‐Cy7 into PE‐Cy5 compensation setting is off by 20%. Note that the fact that this compensation was incorrect is not evident, as neither of these channels is viewed. The effect on the PE channel of this incorrect compensation is due to the interaction of compensation settings across channels (were compensation set by spillover values instead of compensation values, this would not occur; hence, the recent trend by manufacturers to provide control over compensation via the spillover domain). In this sample, the CD8 T cells (PE‐Cy5+) have uncorrected fluorescence in the PE channel and show up as a separate, dull population. The “isotype” gate is that defined by the fully isotype‐stained sample (far left); the FMO gate is that defined by cells stained with everything except CD4‐PE. Without the FMO gate, it would be nearly impossible to know where to set the discriminating gate.
  •   FigureFigure 1.14.7 The most accurate compensation is achieved with the brightest compensation controls. This graph shows three populations of cells stained only with FITC: negatives, a dim population, and a bright population. Linear fluorescences illustrate the principle; the same principles hold true for logarithmic amplification. The correct spillover coefficient is the ratio of the Δorange signal to the Δgreen signal (for either the dim or the bright cells). In an ideal world, this ratio is the same for dim or for bright cells. However, the Δorange and Δgreen values will have an inherently greater proportionate error for the dim cells than for the bright cells; therefore, the spillover coefficient will be less accurately determined on the basis of the dim population than on the bright population. For logarithmic scaling, this problem is exacerbated because dim populations can have orders of magnitude less fluorescence than bright populations (remember, the error of the measurement varies with the square root of the absolute value of the measurement).
  •   FigureFigure 1.14.8 Proper compensation is set when the centers of positive and negative populations align. (A) In a hypothetical experiment, cells stained with CD3‐FITC and PE isotype control were collected at different compensation settings. The horizontal line is drawn through the median of the population. The boxes indicate the analysis gates used when the median fluorescences were computed. Proper compensation is achieved when these centers align; note that the properly compensated positive population extends above the top of the negative population (i.e., above where an isotype gate would be set based on the negative population). (B) An example of using antibody capture beads to set compensation. On the top left panel, a forward‐ and side‐scatter gate is drawn tightly around the main population to select only singlet events. The bottom panels show the distributions in the FITC and PE channels for beads labeled with FITC and PE antibodies, respectively; the beads represent a mixture of capture beads with identical but noncapturing beads as the blank. Gates are drawn around the positive and negative bead populations. Software can be used to automatically align the populations, or manually gate settings to align the medians. The top right panel shows a mixture of unlabeled, FITC‐, and PE‐captured beads. Note that the distribution of the beads is very small, allowing for precise determination of the compensation required. Also note that the distribution in the compensated channel is visually much larger (and similar to that of the blank beads). As shown in Figure , this apparent widening of the distribution is simply a visual artifact of moving the distribution from the bright to the dim area of a logarithmically scaled graph. In this case, because there are many photons being measured for each bead, photon‐counting statistics has not widened the distribution beyond that of the background distribution. Numbers indicate the percentage of events displayed within each gate.
  •   FigureFigure 1.14.9 Setting appropriate compensation. This is a peripheral blood mononuclear cell (PBMC) sample stained with a PE‐conjugated reagent and collected at several different compensation values (into FITC). The top left panel is uncompensated, the top right is undercompensated, the bottom right is overcompensated, and the bottom left is properly compensated. The bold gray lines have been added for emphasis only; no quantitative relationship is implied by the shapes. In this example, the cells have been gated for lymphocytes, so the highly autofluorescent monocytes do not interfere with the setting. Note in particular that while the main PE+ population appears reasonably well compensated in all graphs where some compensation is set, the brightest cells clearly indicate the incorrect compensation level. This is a clear example of why only the brightest stain should be used to estimate correct compensation!


Literature Cited

Literature Cited
   Alberti, S., Parks, D.R., and Herzenberg, L.A. 1987. A single laser method for subtraction of cell autofluorescence in flow cytometry. Cytometry 8:114‐119.
   Bagwell, C.B. and Adams, E.G. 1993. Fluorescence spectral overlap compensation for any number of flow cytometry parameters. Ann. N.Y. Acad. Sci. 677:167‐184.
   Kantor, A. and Roederer, M., 1996. FACS analysis of leukocytes. In Handbook of Experimental Immunology, 5 th ed. (L.A. Herzenberg, D.M. Weir, L.A. Herzenberg, and C. Blackwell, eds.) pp. 49.1‐49.13. Blackwell Scientific, Cambridge.
   Loken, M.R., Parks, D.R., and Herzenberg, L.A. 1977. Two‐color immunofluorescence using a fluorescence‐activated cell sorter. J. Histochem. Cytochem. 25:899‐907.
   Roederer, M. and Murphy, R.F. 1986. Cell‐by‐cell autofluorescence correction for low signal‐to‐noise systems: Application to epidermal growth factor endocytosis by 3T3 fibroblasts. Cytometry 7:558‐565.
   Roederer, M. 2001. Spectral compensation for flow cytometry: Visualization artifacts, limitations, and caveats. Cytometry 45:194‐205.
PDF or HTML at Wiley Online Library