The reason you want to subtract the background is that if you do not know a priori that your treatment acts as a multiplier of the baseline value, you risk having to divide by zero when some of your unstained cells actually show no fluorescence. In any case, you want the fluorescence due to your treatment, so you should subtract non-specific background to get that net fluorescence. Ideally, the signal to noise ratio is large enough that the subtraction is of a small number from a large number, leaving you with a large signal. When you divide by a low number, you artificially increase the signal and make it more dependent on the value of the denominator. If you apply equipartition of variance to your error analysis, you will find that by dividing, you make the means far less precise and much more variable.

If the data are normally distributed, you certainly want to subtract background. If they are log-normally distributed, you should take the logs of the values, determine the means and standard deviations and standard errors of the mean for the log transformed data, and then take the antilogs for your mean values (in this case a geometric mean) and their precision estimates. Hope that this helps.

Use relative values (%), normalized with the control, finally plot only the increases.

Thank you for your answer. I already plotted relative values of stained fluorescence of the treated compared to stained fluorescence of the control. My question is about normalizing the fluorescence of stained cells with the fluorescence of unstained cells to obtain a ratio.

Hello Valentin!

Normalizing MFI of each test sample with an internal control is a good idea. This should help you cancel out the increase in MFI caused due to the buildup of the contaminants like you said. Although you will get a lower ratio for the stained cells in the example that you have given as the degree of increase in the unstained control is more than the stained sample.

But I have a question too. Could the increase in the MFI of the unstained control also be due to the increase in autofluorescence in the algal cells due to the oxidative stress induced by the pesticides?

Normalizing by dividing stained by unstained (background?) values is logically dangerous, since you could potentially end up trying to divide by zero. To correct for differential backgrounds, one should SUBTRACT the backgrounds for each group from the treated and untreated groups, respectively, and then take the difference between net treated and net untreated groups. Is there any mechanistic reason why you expect the treatment to increase fluorescence in a proportional manner rather than turning on a new pathway or mechanism that generates the different fluorescence intensities between the different groups?

Hello Ashutosh, thank you for your answer.

Yes, the increase in the MFI values for unstained cells exposed to the contaminant is most probably due to an effect of the contaminant on the pigments.

Hello Christopher,

I'm not talking about "background" fluorescence bust rather of the "natural" fluorescence of the cells when they are not stained. My real question is: Considering two cells with unstained MFI values of 5000 and 10000 in FL1, after adding the fluorescent dye, will it ADD fluorescence on top of the unstained MFI values, in which case I should just SUBSTRACT the unstained fluorescence ; or will it multiply the unstained MFI values, in which case I should DIVIDE the fluorescence of stained cells by the fluorescence of unstained cells... In my opinion the fluorescence due to the fluorescent dye is not related to the one of unstained cells so I should indeed substract, bu I have a doubt...

The reason you want to subtract the background is that if you do not know a priori that your treatment acts as a multiplier of the baseline value, you risk having to divide by zero when some of your unstained cells actually show no fluorescence. In any case, you want the fluorescence due to your treatment, so you should subtract non-specific background to get that net fluorescence. Ideally, the signal to noise ratio is large enough that the subtraction is of a small number from a large number, leaving you with a large signal. When you divide by a low number, you artificially increase the signal and make it more dependent on the value of the denominator. If you apply equipartition of variance to your error analysis, you will find that by dividing, you make the means far less precise and much more variable.

If the data are normally distributed, you certainly want to subtract background. If they are log-normally distributed, you should take the logs of the values, determine the means and standard deviations and standard errors of the mean for the log transformed data, and then take the antilogs for your mean values (in this case a geometric mean) and their precision estimates. Hope that this helps.

Hi valentin

For normalizing the Mean Fluorescent Intensity (MFI), you should follow this:

" MFIn = MFI of positive population/MFI of negative population" for each marker that you evaluate.

Hope to Help

Regards.

Reza, if your MFI for negative cells varies between 0 and 2, and that for p on positives varies between 100 and 120, by dividing you have made a 110 +/- 10% value become a 110 + infinity or - 50% reading. This makes your results meaningless. By subtracting, you have 109 +/- about 10%. Which set of results do you think will be more informative? Unless you have a prior I information that the staining mechanism is multiplicative, you should go with background subtraction. See my previous comments above.

Hi Christopher

Thank you for your comment

I agree with you, theoretically and in mathematical computations, this method for normalization of MFI maybe not true. but I'm working on immune cells and practically, because of auto-fluorescent of these cells, the MFI of negative population will never be zero.

Let me give you an example: we have two samples and we want to evaluate the MFI of CD3 marker on these. in first sample, MFI of positive population for CD3 marker is 75 and MFI negative population is 25 and in second , the MFI of positive population is 150 and negative population is 50.

so, as i said in above:

First sample: MFIn= 3, Second sample: MFIn=3

Although, MFI of CD3 marker in two sample is different but when we normalize this , we understand that the result is same for both samples. in this method we will have a better comparison.

As I noted in my first note above, if you have evidence that whatever induces the marker acts as a multiplier, and the negatives are ALL non-zero, then by all means, do so. However. I would be concerned about why the two experiments had such large differences in background (two-fold). Also, perhaps the positive scale is not binomial and there is a scale along which one has arbitrarily chosen to select lowest and highest as negative and positive, respectively. That means that one should look at the distribution of marker in the negatives and in the positives, and are there any or no intermediates?

Hi Valentin,

I'm assuming you've already subtracted background as you would wiht any flourscence assay?