When we perform compensation or unmixing for flow cytometry data, we have the option to use either an internal negative population or a completely independent "Universal Negative". At a training I ran recently, we had a nice example of a case where the separate universal negative didn't work, and that's inspired me to write up this post.
First, let's review what the general requirements are for a universal negative (point out anything I've overlooked).
It should be the same source of cells (or beads) as your positive control (e.g., control mouse spleen for both).
It should generally have the same autofluorescence for both (some exceptions to this exist). This should be covered by point 1.
It should be treated the same way as your positive control (same flow protocol, same buffers...).
It should be acquired on the same cytometer with the same settings (voltages/gains, flow rate...).
It should be acquired at the same time and at the same temperature. This is particularly important for autofluorescence in live cells, which changes in response to external conditions.
That's a fair few conditions. If you have trouble replicating all of those, you might be better off using an internal negative.
Why might we want to use a universal negative? For one, it saves us the time of setting negative gates for each of our controls. Two, if your single color control lacks a clear negative (e.g., CD45, CD11a or CD44 staining of leukocytes), an external negative can give a good indication of what the true negative is. In other cases, you may have simply added too much dye (common with viability stains), resulting in all the cells being more or less positive (hint: this will almost inevitably produce inaccurate unmixing, so try to keep the control staining consistent with your fully stained sample). Finally, there are cases where the positively stained cells are very different from the remaining negative cells in the single stained cell sample. This can lead to spectral extraction errors. An example of this might be for CD45: in a spleen or PBMC population, you might have enough CD45- cells to identify a negative, but that negative will be exclusively comprised of non-leukocytes (probably erythrocytes and/or fibroblasts). This would be even more extreme if you used a control for CD45 on a tissue sample where the CD45- cells would a large collection of non-hematopoietic cells, debris and dead cells. The difference between the positive cells and the negative cells in these settings is no longer just the fluorophore, but also includes the difference in autofluorescence between vastly different cell populations.
With certain compensation beads that I do not recommend using, there is also a difference in background between the positive and negative beads, so you would get somewhat better results by using a universal negative.
That said, in most cases, I have seen relatively little difference between using an internal versus universal negative. The cases where there is a difference, something is going on that requires us to think a bit about what we're doing and why.
Let's look at an example where there is a difference.
Here's a simple 20-color staining on mouse splenocytes run on a 5-laser Aurora.
The plots above were unmixed using internal negatives, using automated autofluorescence extraction.
Here's the comparison:
More plots:
There's an uncorrected autofluorescence spike visible in the plots (hypernegative for Foxp3 in the upper left, positive for CD11b & PD1 in the lower right, hypernegative for both on Foxp3 vs Ki67 and CD62 vs Ki67). Some other minor issues with Ki67 and PD-1.
Why is this happening? Looking at the spectral profiles, nothing appears to be wrong:
If we look at the normalized spectral traces, which you should always do, we can see there's an issue. The normalized traces are the spectra that are used for unmixing--these are the most important part of your workflow.
At first glance, this might look fine to you. If you look closely, you'll see that the scale goes below zero. Negative fluorescence is not possible. The dyes are not emitting anti-photons. Moreover, there is a drop in exactly the same detectors in both profiles. This indicates a mis-match in the basal fluorescence background (autofluorescence) between the Universal Negative and the controls.
The user here has followed the protocol perfectly, though. The cells are the same, the only difference in the treatment is the conjugated antibody at the correct concentration.
The issue here is that the machine clogged! Here's what some of the controls look like when plotted against Time (Time is your friend for identifying acquisition errors with the cytometer).
Frankly, it's amazing the unmixing works with this kind of data. I'm not gating on Time for the unmixing, though, so why does a clog (or a series of clogs) matter? It matters because if the flow cell is dirty the samples won't pass through the lasers the same and the excitation will be slightly different. In this case, the Universal Negative sample was run first, before the instrument clogged, and about half of the controls were acquired after clearing the clog.
Here's what happened:
The samples were put on the machine without filtering and contained visible lumps. This was actually intentional for the purposes of showing what happens when you do that and to illustrate how it can mess up the data. Success!
The cytometer clogged eventually, manifesting as a failure to acquire any events and a very low flow rate (<1ul/min).
SIT flush failed to clear the SIT.
The flow cell was cleaned and then water was run for about half an hour to clear out any residual bleach.
The remaining controls were acquired.
The step that's missing and probably would have helped? Running the QC again before acquiring any new data.
The upshot is that in this situation, the Universal Negative isn't universal anymore. It's only a good negative control for the samples that were run prior to clogging. After clogging, there's a consistent change in the detector sensitivity in certain channels (mainly far red), resulting in systematic errors for dyes in that range.
There were a couple of residual unmixing errors (Ki67-AF700, PD-1-BV711) in the data even using the internal negative. Here's what that looks like on the other Aurora, the one that didn't clog:
New Holland Honeyeater, Tasmania
Comments