Why do we get autofluorescence problems?

Today the aim is to illustrate visually how cellular autofluorescence ends up in our unmixed flow cytometry data, confounding our results.

Autofluorescence is fluorescence that already exists in our cells prior to staining. It can vary depending on the cell type or activation state, meaning there can be some cells with more or less signal in the data and these signals can interfere with our measurements in different ways. In this example of mouse lung cells, there is signal in various channels in the unmixed data without any staining, which limits our ability to use those channels for detecting markers.

That's lot of signal for no staining! Macrophages are fun!

If there is similarity in the profiles of the cellular autofluorescence and the fluorophore spectra being used for unmixing, the autofluorescence can be misassigned as fluorophore signals in the unmixed data.

Below we see similarity in the peaks for the normalized autofluorescence in PBMCs and the signals for BV510, AF532 and BB515.

Normalized AF signature (PMBC lymphocytes)

Spectral signatures for a few fluorophores on the Cytek Aurora, taken from Cytek Cloud

As a result, there is signal even with unstained cells in the AF532 and BB515 channels when unmixed. Extracting autofluorescence as a separate signature provides a better match for the autofluorescence signals, redirecting them and reducing the undesired signals in the fluorophore channels.

Unstained PBMC unmixed with or without the automated AF parameter

Autofluorescence extraction can create problems in the data if not done carefully. If done well, it can improve the resolution in the data and allow us to correct for variable autofluorescence signatures in different cell types.

Here's another example. This is the raw mixed data from unstained mouse brain. This is what we call a spectral ribbon plot, displaying heatmapped data from all the cells across all the detectors.

Again, loads of signal. Yay, microglia!

After unmixing with our fluorophores, we are left with the residuals. This is the part of the data that is not used in the unmixing model. We can nicely see where chunks of it have been converted into unmixed data signals. This is unstained, so all of this is autofluorescence that has been incorporated into the unmixing.

What's happening in this example is that we're only modeling the presence of fluorophores in the data. We're not creating a model that accounts for autofluorescence, so the "best fit" includes assigning all the signals as fluorophores.

Violet Sabrewing, Costa Rica