I just put this question on Purdue's flow cytometry lists, and wonder if anyone here has an opinion on this. I am looking at shifts in various surface marker expression in different human peripheral blood leukocyte populations (e.g. T cells, monocytes, neutrophils) following certain treatments. Since doing some quantitative assessment of prion protein expression years ago (Brit J Haematol 1999; 107: 804) I have a thing about not using "percent positive" by comparing the degree of non-overlap of stained cells with unstained control (which my students often try to do inappropriately) since we found subsequently that although there is only a very small shift (large overlap with control) when red cells are stained with anti-PrP (and only certain anti-PrP antibodies can pick up PrP on red cells because of different epitope exposure), blotting shows that human red cells do indeed express PrP, but at low density. Thus 100% of red cells express PrP (while looking at non-overlap you would say that there are very few that exceed the negative control gate).
So - if I see a clear single peak, and it doesn't coincide with the unstained control, I regard it as 100% positive for that cell population - then take the shift from baseline (control) as an indication of the density of marker on those cells. Our original prion work used unlabelled primary (mouse monoclonal) and used a "standard" fluorescent conjugated secondary anti-mouse-IgG to compare the amount of primary antibody on different cells and on beads (Quifikit) coated with different known amounts of mouse IgG to give a standard curve against which we could compare the cells (all geometric mean fluorescence index). Having done a lot of ELISA I am comfortable to relate linear concentration to log of brightness. We calculated the amount of IgG on different cells and apparent IgG on unstained cells (without primary anti-PrP, secondary anti-IgG only), then subtracted the control (IgG value) from the value for the cells after the calculation, giving net IgG (primary, mouse) bound. We were comfortable relating that to molecules of PrP since there was a single epitope on each PrP molecule that the primary antibody bound to.
In the current situation I am using different fluorescence-conjugated primary antibodies to different markers (and other antibodies to define the leukocyte populations, in a multicolour system) - and because of differences in specific activities and affinities I obviously can't compare one anti-marker antibody to another. However for the same antibody/marker (same batch, lot, acquired using same instrument settings, etc) I should be able to compare different individuals and/or different times (in relation to treatments). We note that different individuals, different cells, different channels, different times can influence the MFI of the unstained control, so they can move around slightly. Where we get homogeneous single peak staining for a marker (what I regard as 100% positive - it is more complicated when there are two peaks implying only a subset of that leukocyte population is positive for that marker, where a different approach needs to be taken to the effect of a treatment) you obviously get an MFI value for that peak (I use geometric means). Fluorescence is plotted on a log axis. My instinct is that the log(MFI) values (stain - control) should be subtracted to get the net shift, then that can be converted back to a linear value, rather than subtracting the linear MFI control value from the linear MFI stained value. The reason I hesitate is that subtracting one log value from another is the same as dividing linear values - so the result is actually giving the net stained value as a multiple (fold) value of the unstained value.
Many years ago, looking at tritiated thymidine incorporation by stimulated proliferating lymphocytes, we held the view that so-called stimulation indices (the stimulated response divided by the unstimulated background) were spurious and that responses should be measured linearly, net of background. What are readers' opinions on MFI in flow cytometry in this context - subtract log values or subtract liner values for a particular marker net of background?
It is clear that different cell populations with a sample, of say PBMC, have different background fluorescene values. This probably occurs due to properties of the cells themselves (i.e. how many/much autlfuorecent protein they contain) as well as bleedover from the antibodies used in your panel to define that cell population. For instance, basophils have slightly higher autofluorescence than T cells bc they are more granular. If i were to identify basophils w/in a PBMC mixture using a FITC antibody, then they may gain slightly more background in the PE channel due to incomplete compensation, since it is never perfect.
Do you use an FMO control or a fully unstained sample? This would be your entire panel of antibodies used to identify cell types within the mixture, only it would leave out the antibody that you are trying to quantify. This way, you can gate on a distinct cell type within the mixture (T cells, B cells, basophils, etc.) to get the background fluorescence of that cell type. Then, you would subtract this background value from the geometric MFI of your sample containing the antibody of interest to get a net change. I'm not sure if it matters if you do linear or log - can't help you with that one.
I think this method is a little more accurate, but still not perfect. Don't get stuck trying to turn a SEMI-quantitative method into a fully-quantitative one. If you run enough replicates, you should be fine.
I would subtract linear values. It seems to me that subtracting log values would make your data far more sensitive to small shifts in the background where the detector sensitivity is lower.
Also, you didn't ask an explicit question about this, but you noted that "different individuals, different cells, different channels, different times can influence the MFI of the unstained control, so they can move around slightly." This is very true, and I think this informs what the appropriate control should be when comparing MFIs in a multicolor panel: an FMO using the same source of cells as your stained samples and collected at the same time. This would control for all the factors you noted.
Because background staining and autofluorescence will vary in different populations in a heterogeneous mixture, using a general unstained sample may not appropriate as a normalization for a marker on a specific population (e.g., if you are looking at the expression of a particular activation marker on CD4 T cells in whole blood, you should compare this to background values for the same CD4 T cell population (compensated the same way), not unstained whole blood, in which other populations with different autofluoresence, such as neutrophils, will contribute to your background).
Hope that helps.
Hello, I do not conduct semi-quantitative Flow cytometric measurements via MFI or calculated values. In a multicolor assay I see a problem in spillover from other channels etc. The 'stimulation indix'is an interesting idea, and why not look for a combination of markers where you would be able to create such an index? In other word: calculate a ratio of the fluorescene in the "antibody-channel of importance" and the fluorescence of a "housekeeping-marker-channel". I guess, there might be combinations that give sense. In order to exclude differences in fluorescence because of different antibody-mixes, one would need a master mix for all assay, that will be compared with each other (see the discussion about master mix-storage in the Purdue`s platform).
Anyhow, thanks for the discussion you opened, because it is an interesting issue,
Greetings, Th. Adler
I am completely familiar with using appropriate FMO's on the individual cell types that I am looking at (I have been doing flow cytometry for nearly 20 years). I am well aware that different cells in a sample have different autofluorescence (for example eosinophils are very bright - but easily recognised by their very high side-scatter) and when comparing different populations I would always only compare the background of a specific population with the stained sample for that specific population (using FMO when staining is required to identify that population). I am not asking for advice on these basic requirements. I am also aware that there are different opinions held by different "flow cytometrists". In this case I am seeking the opinion of what should be used when comparing a specific subpopulation (using all relevant FMO controls) in different individuals, or the same individual on different days. I can clearly see (from a large panel of markers being studied) that some markers are not expressed, some are expressed but do not change (pre/post treatment), some consistently decrease and some consistently increase pre/post treatment across a group of individual patients. This is largely (but not entirely) reflected across different cell types (I am analysing paired samples pre/post within a cell type - only looking at trends for consistency across different cell types since that information is available, but not analytically comparing different cells). However, I can clearly see (on histogram plots of stained v unstained of gated cells) that the "distance" on the logarithmic plot of fluorescence intensity between stained and unstained within a particular group (for example, monocytes pre-treatment) can be approximately equal, but there can be variation in the brightness of the unstained sample (reflected in the brighness of the stained sample). For example, say an unstained sample has a (log) value of 1.17 and a stained value of 2.56 (net log difference 1.39), while a second sample has (log) unstained 1.22 and stained 2.62 (same net log difference 1.39). If you use the linear values (first sample unstained 14.79, stained 363.08, difference 348.29; second sample unstained 16.59, stained 407.38, difference 390.79), the second mathematical approach (subtract the linear values) gives the second as having a greater net shift than the first. This doesn't "look" right - nor does it follow other experience (from ELISA or where log "light" (absorption or, in luminescence, emission) is linearly related to concentration. I am not trying to work out any absolute values here, just a statistical comparison of relative MFI's.
It's not entirely clear to me why you think the values derived from the linear subtraction don't "look right" in your example. It seems to me that a shift in background from 14.79 to 16.59 is very small and could simply reflect variation due to detector noise. In contrast, the shift from 363.08 to 407.38 is larger and is less likely to simply be due to detector noise. So in your two examples, I think that the second scenario does in fact reflect a greater net shift than the first, and the linear subtraction accurately reflects this whereas the log subtraction does not.
Really later to this question (am only a recent member of ResearchGate), so don't know if you found a solution/made a decision.
I'm in the linear camp. I don't think it's fair to compare flow to ELISA or bioluminescence, since they are enzyme-driven processes. Fluorescence emission from a set number of fluorochromes should be linearly related to the number of fluorochormes present, provided you don't have something funny going on (quenching, FRET etc.) and of course provided you are detecting your signals within the linear range of the PMT.
When we look at CFSE dilution peaks, where we know that fluorescence is halved with each cell division, the linear MFI of each peak is pretty much bang on half of the next brightest peak. And although autofluorescence doesn't impact on the high values, the MFI of later peaks keeps to the 1/2(previous peak) formula with a linear correction for autofluorescence.
- Badges
- Science method