Flow cytometry compensation

Posted on by kmatreyek

So I tend to use fluorescent protein combinations that are not spectrally overlapping (eg. BFP, GFP, mCherry, miRFP670), so that circumvents the need for any compensation (at least on the flow cytometer configurations that we normally use). That being said, I apparently started using mScarlet-I in some of our vectors, and there is some bleedover into the green channel if it’s bright enough.

These cells only express mScarlet-I, and yet green signal is seen when red MFI values > 10^4… booo……

Well, that’s annoying, but the concept of compensation seems pretty straightforward to me, so I figure we can also do it post hoc if necessary. The idea here is to take the value of red fluorescence, multiply that value to a fraction (< 1) constant value representing the amount of bleedover that is happening into the second channel, and then subtracting this product from the original amount of green fluorescence to make the compensated green measurement.

To actually work with the data, I exported the above cell measurements shown in the Flowjo plot above as a csv and imported it into R. Very easy to execute the above formula, but how does one figure out the relevant constant value that should be used for this particular type of bleedover? Well, I wrote a for-loop testing values from 0.0010 to 0.1, and saw whether the adjusted values now resulted in a straight horizontal line with ~ zero slope (since then, regardless of red fluorescence, green fluorescence would be unchanged).

Now as that value becomes too large, then more will be subtracted than should be, resulting in an inverse relationship between red and green fluorescence. To make my life easier to find the best value, I took the absolute value of the resulting slope in the points, which pointed me to a value of 0.0046 as the minima, for mScarlet-I red fluorescence bleeding over into my green channel on this particular flow cytometer with these particular settings.

Great, so what does the data actually look like once I compensate for this bleedover? Well, with this control data, this is the before and after (on a random subset of 1000 datapoints)

Hurrah. Crisis averted. Assuming we now have sample with both actual green and red fluorescence (previously confounded by the red to green bleedover from mScarlet), we can presumably now analyze that data in peace.

Just for fun, here’s a couple of additional samples and their before and after this compensation is performed.

First, here are cells that express both EGFP and mScarlet-I at high levels. You can see that the compensation does almost nothing. This makes sense, since the bleedover is contributing such a small total percentage to the total green signal (EGFP itself is contributing most of the signal), that removing that small portion is almost imperceptible.
Here’s a sample that’s a far better example. Here, there’s a bunch of mScarlet-I positive cells (as well as some intermediates), and a smattering of lightly EGFP positive cells throughout. But aside from the shape of the mScarlet-I positive, GFP negative population changing from a 45-degree line to a circular cloud, the overall effects aren’t huge. Still, even that is useful though, b/c if one didn’t look at this scatterplot (and know about the concept of bleedover and compensation), one might interpret that slight uptick in green fluorescence in that aforementioned population as a real biologically meaningful difference.