Stabilizing your unmixing

Today's post pulls from a lot of previous posts on this blog to show how to effectively eliminate batch effects in your flow from an experimental side. Stabilizing the unmixing is a big part of this. This is going to be a set of posts since it's ended up being a bit long.

We're going to look at how to get the data to look like this--stable over time:

Unmixing using stored controls on the Aurora. tSNE plots with FlowSOM clusters in colored overlay of a very large panel using the same donor's frozen PBMCs thawed and stained using a stored master mix.

I'm a firm believer that making experiments easier gives you better data. An easier experiment is more reproducible. An easier experiment means fewer chances to make mistakes. With spectral cytometers and new dye technology, flow cytometry has reached a point where it's easier than ever to jump in and stain your cells with a lot of colors. A lot of people's introductions to flow now are with 20+ colors. This is great, but it hasn't really changed the fact that running a large panel requires careful attention to the controls to get consistent, correct results. The more complex the panel, the greater the precision required. Getting the controls right and consistently right is actually really difficult, in my opinion. It's especially difficult to do when you're trying to prepare the master mix(es) for your big panel(s) and thawing/prepping all your samples.

So, in this set of posts, let's look at how using the software tools can help us get good results and make our lives easier. I'm going to show some examples of how re-using controls in a lazy way (re-using the data files from a different day) doesn't work (in my opinion) and how storing the controls as spectral references in the software does. My thanks today to Adam Davison at Cytek for sitting down with me last year to go over the best way of doing this. Additional thanks to Kristina Petrović at Sony for her help with the ID7000. Just so we're clear, I'll also point out that I'm not trying to claim I'm the first to do any of this.

The key thing here is that this allows you to separate the preparation and acquisition of controls, doing it potentially on a different day from your big experiment. This is the same idea explored in the stabilization of master mixes in this post. Both of prepping mixes and staining controls add complexity and stress on the day of the big experiment. By separating them out, we can focus on doing each of these things well, reducing our chances of making a mistake. We also give ourselves the chance to re-do things; not everything has to work perfectly on the same day every single time (that's too much to ask, at least for me).

What do I mean by setting up spectral references? On the Aurora and the ID7000, you have the option to save your spectra from your single-stained controls in the software. On the Aurora, you do this by running the controls in a special mode under the QC menu, storing them in the "library" (I'll put up instructions for this in a future video, or you can check out the Cytek webinars). On the ID7000, you can click on your acquired control and save it as a "spectral reference". As I understand it (correct me if I'm wrong here), these spectral profiles are then updated by the software to reflect changes in the instrument as per the QC. This, in theory, means you can use a stored control for unmixing, and it will be accurate.

First, let's consider the theoretical reasons why this would and wouldn't be true.

The Aurora and ID7000 have good QC (in my opinion, but I am not at expert on this stuff) and the sensitivity of the detectors is adjusted to meet certain criteria. This standardization is meant to make the cytometers "see" the fluorophores the same way over time. Additionally, the vials of fluorophore-conjugated antibody, even tandems, are stable if well treated, so the fluorophore signatures (as measured on compensation beads on the same cytometer) are stable over time. With the Aurora, the Cytek Assay Settings will attempt to mathematically correct for any minor variation in detector sensitivity (from the QC) and apply this to your unmixing (again, correct me if I'm wrong here, my understanding of this is more empirical). So, yes, in an ideal world, this should work.

Why wouldn't this work? Biology is messy. We, as humans performing the experiments, are error prone. We do not produce exactly the same result every time. With larger panels, the chance of minor variation increases, and the tolerance for error in the result diminishes. Even the simplest experiment, say GFP measurement, will be subject to variation in unmixing over time due to changes in cellular autofluorescence, which is affected by metabolism (and thus handling of the cells).

Several years ago, I tried to unmix a large panel using stored controls on the Aurora and it was pretty much a disaster after only a week. While I can't say exactly all of the things that you need to get right in order for this to work, these appear to be important factors:

• Run the instrument QC daily

• On the Aurora, use saved settings linked to Cytek Assay Settings

• On the ID7000, use Standardized Mode

• Prepare good quality controls that are accurate for your panel

• Have a well-designed panel. You want to avoid unmixing hotspots, as discussed in OMIP-102.

• Have a stable master mix or a robust protocol for preparing the mix

• Have a stable staining protocol

• Have an effective blocking protocol

• Prevent tandem breakdown

• Use the same lot of antibody as in the reference

  
  

How stable are experiments run on the Aurora, using the Cytek Assay Settings?

Dot plots showing the stability of the unmixing over time.

Note that these are some of the more problematic combinations in this panel: PE-Fire 810 is prone to tandem breakdown, separating BUV395 and Spark UV 387 pushes the limits of the machine, and the KIRAVIA Blue 520 is getting hit by an unmixing hotspot in this iteration of the panel design.

If we don't use the software to track the changes in the instrument, re-using controls for unmixing can create unmixing errors. Here are examples where I've unmixed the fully stained samples using controls acquired on day 0 on the Aurora. This is just adding new samples to an existing experiment, assuming the spectral signatures will be stable and that there's no change in how the cytometer detects these fluorophores. Those are big assumptions, as addressed by Mario Roederer in this paper on compensation (see Compensation Myths). This example I'm showing is with a single donor using a stored master mix, so the aim is to focus on instrument variation.

Changes in detection of PE-Cy7 vs. PerCP-Fire 806

There's drift and variability. In this case, that can be largely fixed by using the stored spectral reference (library) in the software, even out to a month as we see in the lower right panel. To be clear, using the library won't always work--there may be changes in how you've treated the cells that affect the spectra of the fluorophores. There were also a lot of cases in this experiment where re-using the acquired controls worked well in this test. That is likely due to using saved settings, maintaining the instrument well and having limited brightness in the panel so that any errors are smaller (all of that stuff in the bullet points above).

In the following posts, we'll look some more at this drift and how to set up the spectral references.

Cape Sugarbird, South Africa