top of page
Writer's pictureolivertburton

Autofluorescence in conventional flow cytometry

The autofluorescence saga resumes!


I've been a bit slow with writing these posts because autofluorescence (AF) touches on many aspects of flow, making it hard to know where best to start.


Before we get into all the new ways of dealing with AF in spectral flow cytometry, let's take a step back and look at how AF is handled in conventional flow. I hope that by looking at this "simpler" cytometry, we can understand a bit more easily what's going on.


You'd be forgiven for thinking that we don't do anything about AF in conventional flow. In actuality, AF background subtraction is such an integral part of the conventional flow cytometry workflow that we don't even think of it as such. We call it "setting the voltage" or "adjusting the negative".


On a conventional flow cytometer, you need to adjust the voltages (or gains) on your fluorescence detectors using your unstained cells to position them at the lower end of the plot, near zero (hopefully above the electronic noise!). In doing this, we "zeroing out" the main AF signal, setting our background to near zero. Because we're doing this differently in each detector, we are differentially subtracting signals, i.e., subtracting an AF signature. We're just doing it manually and crudely.


Obviously, this doesn't do anything about spikes of AF, different types of AF from different cell types, and it certainly doesn't help with AF interfering in single stained controls for compensation. What it does do is make the plots more interpretable because our "negative" is now where we want it to be.


We can see that AF background still occurs by looking at cells that have been treated with fix/perm reagents versus those that haven't, as in the example in this post where we saw the spectral signatures on the Aurora.


Increased signal in the BV480 with fix/perm treatment. Mouse splenocytes run on the FACSymphony using the same voltages. The baseline (background) signal in the BV480 detector is higher after fix/perm treatment due to an increase in AF.


Spectral signatures of splenocytes from those two conditions on the Aurora. The bandpass for the BV480 detector on the Symphony roughly corresponds to V5-V7, right in the middle of the biggest hump.


Why does setting the voltage matter? The difference in MFI between the positive and negatives in CD25 above may well be very similar (different samples from different mice and protocols, so some differences are expected). So, the numbers might give us the same results. The scaling matters, though. We view the data on a biexponentially transformed logarithmic scale, which means that the tick marks on the axes aren't really the same. In the Fresh Cells stained with CD25, the negative is centered at around 100 (10^2), and we can see a CD25-low Foxp3- population creeping up to 1000 (10^3). We can see that because each of those ticks between 10^2 and 10^3 represents an increase of 100 (10^2). In the Fix/perm cells, the negative is almost reaches 10^3, and each tick above that represents an increase of 1000 (10^3). It's very hard to see the CD25-low cells in that part of the plot because they're only a couple hundred MFI units above the negative, so all within a single tick above 10^3. That's important for us, visually, to be able to set gates, and it's important for dimensionality reduction and clustering, which use the scales we define in the data. Setting the voltage is part of setting the scale.





This isn't, though, the only way of handling AF in conventional flow. There are a couple other main approaches, which are geared towards dealing with the spikes and secondary signatures:

  1. Compensate it. This is comparable to AF extraction in spectral flow.

  2. Dump it. Design your panel to exclude it, usually through gating.

  3. Ignore it (often done in not-so-blissful ignorance).


Compensation is probably the most straight forward approach. For this, we can determine a detector that receives a strong signal from the main spike of AF (often the channels for BUV496 or PerCP-Cy5.5), and instead of placing a fluorescently conjugated marker there, instead designate that channel to receive the AF. We then compensate this channel using the unstained cell control. (Note: as with spectral flow, this can be done with the unstained cell control of your choice, say the most autofluorescent tissue you've got.)


Unstained mouse splenocytes exhibiting a spike of autofluorescence in the channels used to detect PerCP-Cy5.5 and BV510. We could dedicate either channel to receiving the AF signal and compensate it out of the other channel (and all the rest while we're at it).


After applying some compensation manually. There's still spread, but we now have some more space to work with in the BV480/BV510 channel. Also, look how messy that AF "fluorophore" is--the negative is now really skewed.


AutoSpill employs this compensation method for extracting AF. This works pretty well on conventional instruments for simple lymphocyte populations such as human PBMCs or mouse spleen where you are dealing with somewhat homogeneously autofluorescent cells (they're not really, but on a conventional instrument, this approximation is often good enough).


Here's my unstained sample showing the AF with and without AutoSpill compensation. In the lefthand plot, I've compensated just the fluorophores manually in Diva, ignoring the AF signals. On the right, I've compensated using AutoSpill in FlowJo, designating the empty channel BV750 to collect the AF signal on the unstained cell control.



Cool, that kind of helps, but with conventional instruments, we're more limited on detectors, and we don't have them in all the right places to catch the AF signatures well. So, this compensation method works somewhat, but you'd usually only do one signature. Most people are reluctant to do even that because they want more markers. Plus, it's still messy because there's a lot of variability in that AF spike.



A dump channel is the next option. There are many ways of doing this, because this is essentially designing your panel around the AF. Say you're working on T cells: you can create an exclusion gate for all the non-T cell populations (macrophages, monocytes, B cells), even including the viability marker in this mix if you want. You pick a channel that receives a lot of AF (as above, perhaps a violet-excitable ~500nm emission dye), add all the other markers you want to exclude from data in this channel, and--hey presto!--the AF gets gated out along with the unwanted cells. The compensation for this channel is inexact because you've got multiple fluorescent conjugates (hopefully at least all the same fluorophore) and AF, but if you just want to cut out everything falling into that dimension of the data, spillover from that data isn't so big of a problem.


I don't have any data to demonstrate this with multiple markers because I don't use this technique that way. I worry too much about excluding things I really wanted in a channel that doesn't have a clear meaning. In some cases, though, I've done this with a single marker that will catch the autofluorescent cells and put them away from my target cells. For instance, here's an example with CD11b catching the AF and carrying it away from CD3.


To start, I've got an AF spike appearing in both the channels that I want to use for CD3 and CD11b:

Unstained splenocytes on the FACSymphony.


If we just go straight into gating our CD3+CD4+ T cells, we'll end up with this AF spike in our gate. The size of that spike will vary depending on how frequent the cells are that cause it in the particular sample. In this case, those cells are macrophages, so a sample with few macrophages like the lymph node will have minimal AF spike contamination, but what about a tissue sample with lots and lots of macrophages?


Congrats! We've discovered a novel CD25+ NKT cell population unique to the brain!


Hmm, couldn't be those pesky autofluorescent cells, could it?


I guess we can hold off on talking to the editor at Cell.



So, what if we design our gating strategy to exclude the AF spike?

Fully stained samples using CD3 in BV650 and CD11b in BV750.



If we define our T cells as CD3+CD11b- (note, not always true in non-homeostatic conditions), then we can gate cleanly on T cells, excluding the AF-spike containing cells. This is similar to just compensating the AF spike into the BV750 channel, but we're overwriting it with CD11b staining, on the premise that the autofluorescent cells are CD11b+. Is this really correct? Probably not, but you can get away with cheats like this in conventional flow. In this case, it's probably better than the alternative.



Finally, we've got the option to incorporate the AF signature into the marker. If you use a cell-based single colour control, the spillover signals are a composite of both the fluorophore on your antibody (or other reagent) and the inherent AF signature of the specific cell type that is staining positively for that marker. If you don't use AutoSpill, the spillover signals will also include spikes of AF intruding into that channel. That means, we'll be trying to compensate for two signals, like so:


Applying compensation to equalize the MFIs for BV650 in both the PE-Cy7+ and PE-Cy7- population. This splits the autofluorescent and non-autofluorescent data around zero as spread.


The problem with this approach is that AF doesn't typically just appear in one channel, and it's not particularly similar to the spectra of most fluorophores we use.

Single color cell controls from mouse spleen run on a FACSymphony showing a spike of AF (macrophages) that co-stains for CD24 and CD172a. We can see this as a spike because these cells haven't been taken into account when calculating the compensation.


If we compensate traditionally using these controls, we incorporate that AF spike into the spillover for each fluorophore where the AF spike intrudes. We do that over and over for each one. Since that's a lot of markers and fluorophores in this panel, which focuses on myeloid cells, that creates a big mess:



That might look familiar to users of spectral flow. And that's the point of today's post--we face similar problems with AF in spectral flow that we did with conventional flow. Ignoring the problem doesn't make it go away. Fixing it badly also doesn't help.







Cactus Wren, Arizona, USA

50 views1 comment
bottom of page