Methods of subtracting autofluorescence post-acquisition

Autofluorescence (AF) is the natural emission of light by cells and other cytometric targets. It can contribute substantially to background noise and has the potential to obscure dimly positive populations. It also varies by cell depending on the size and complexity of the cell.

In FlowJo, there are two methods of estimating the contribution of AF per detector, and then subtracting it out: either to assume that the background signal without AF is 0, or to use AutoSpill1 to create a linear regression based model of AF.

Both methods come with the caveat that they produce an estimate of the average AF of all the cells in the population used for the estimate. If the population is homogeneous then the estimate will be more likely to subtract an appropriate amount of AF per cell. An estimate produced on a heterogeneous population will result in an ‘average’ AF across all included cell types, and a larger variance in over and under subtraction.

Zero Fluorescence Assumption

In both approaches, AF is treated as an additional parameter to estimate during compensation, thus requiring both an empty detector and an unstained control to serve as a representative primary channel for an AF parameter and the compensation control, respectively.

To use this approach in the FlowJo Wizard, be sure to include the unstained control in the compensation group. Start the wizard and assign the unstained control to a detector. The wizard will create a clean-up gate on a population of cells that have a consistent FSC vs. SSC signal to attempt to include cells with similar AF; make sure this population is the population you would like to use for AF estimation. We make the assumption that the signal in the unstained control is completely attributable to AF. This is a reasonable assumption in most experiments as the instrument derived background signal is relatively invariant to signal strength2, and thus an equal contributor to all cells regardless of type. The compensation wizard in FlowJo will auto-gate what will typically be a monolithic population on the clean-up gate of unstained cells, deem it the ‘positive’ population, and calculate the median fluorescent intensity (MFI) of this population in the selected channel. From the dropdown menu in the Negative column the user has the option to select <True Zero> to use a value of zero for the MFI of the background or ‘negative’ population.

Compensation will then proceed in the traditional manner by solving a system of linear equations with a target of setting the MFI of the ‘positive’ population equal to the MFI of ‘negative’ for every parameter that a compensation control tube was not stained with.

Recommendation: When selecting a channel to designate as the AF channel, it can be useful to pick the unused channel for which AF is the highest. A nice way to identify this channel is to drag the clean-up gate of the unstained control into the layout editor, right-click on it and choose to make a multigraph overlay of all histograms. The unused channel with the brightest signal is your best choice.


As with the Zero Fluorescence approach, an unstained cells control and an additional detector are required for this approach, as AF will again be treated as a parameter. AutoSpill is a robust linear regression based approach to compensation that involves fitting a best fit line through all of the data with the clean-up gate and creating a spillover matrix such that the slope of the best fit line becomes flat. A nice advantage of this method is that positive and negative exemplar populations are not needed to calculate spillover; all data in the clean-up gate is used. Thus a regression line can be fit to an unstained sample, using the same assumptions to estimate that deviance from a zero slope is attributable to AF alone. It can then be included as a parameter to estimate during compensation.

All caveats about heterogenous vs. homogeneous cell types still apply, and because AutoSpill is an iterative approach that attempts to optimize the complete spillover matrix, if including AF subtraction it is best if all compensation controls are cell based.

Additional Considerations

  • As AF is present in all channels, these approaches work best with spectral systems that will more closely estimate the full impact of AF.
  • Differences between AutoSpill and Zero Fluorescence are expected. The approaches are calculated differently with the AutoSpill approach using a regression approach that will no necessarily use zero as the target for AF.


  1. AutoSpill is a principled framework that simplifies the analysis of multichromatic flow cytometry data
    C. Roca, O. Burton, T. Prezzemolo, C. Whyte, R. Halpert, Ł. Kreft, J. Collier, A. Botzki, J. Spidlen, S.Humblet-Baron, A. Liston. Nature Communications 12, 2890, 2021
  2. Evaluating Flow Cytometric performance with Weighted Quadratic Least Square Analysis of LED and Multi-level Bead Data. D. Parks, F. El Khettabi, E. Chase, R. Hoffman, S. Perfetto, J. Spidlen, J. Wood, W. Moore, R. Brinkman. Cytometry Part A, 91A:232-249, 2017.