You've designed your big flow panel--a great accomplishment! But wait, there's more work to do before you use it 😟
Panel instability is a big factor in producing batch effects. Are you sure you're going to get the same results every time you run it on control samples? Let's look at a few ways we can make flow panels more robust.
If you're planning to use your flow panel for longitudinal or multi-center studies, these steps are more important.
Standardize sample collection and cell numbers. Your flow panel is essentially a chemical reaction. Your antibodies are reacting against their antigens on the cells with the aim of producing a set product: optimal staining. If you use variable amounts of cells and fixed antibody amounts, you will get variable product. Understandably, with valuable human samples, we want to stain everything, no matter how much we have. If you decide to go with this approach rather than counting and using a specific cell number, you can mitigate the variation by stress testing your panel first. How well does it perform with different cell numbers? Test it on the minimum and maximum number you envisage staining. Are there certain antigens that stain too brightly with few cells or too dimly with many cells? Can you fix this by changing the staining volume proportionally to the approximate cell number? Should you change the dilution factor for any antibodies?
Human PBMCs stained with the same panel, but using differing numbers of input cells. Resolution of CD45RA is lost with higher numbers of cells, indicating that the antibody is limiting (and probably needs to be increased in concentration). Spread and unmixing issues may become more apparent at low cell numbers, as in this example.
What are the extremes of biology your panel will face? If you've done all your testing with PBMCs and now want to assess tumour-infiltrating CAR T cells, you're going to want to run that panel on something approximating the tumour sample. If you can't get any extra sample of exactly the right thing, use something else that might replicate the biology somewhat: another type of tumour, CAR T cells activated in vitro or perhaps tonsil.
It's important not to "overfit" your panel design to the cells you use for testing it. If you're always using splenocytes or buffy coats for the testing, and then go to using various tissues or healthy versus patient samples, well, your panel may only work perfectly for the control samples. That's not really what you want.
Taking the example below, if we were only ever to have unstimulated healthy PBMCs, a decent solution for CD137 in the panel design would be to put this on a marker with extreme brightness without any regard to spread from that fluorophore. Why? Spread is proportional to signal strength, so if we only ever have tiny amounts of signal, we'll never get much spread from this marker. However, as we can see from the other conditions, that could be disastrous. Markers that may vary highly in expression need to be assigned to fluorophores with minimal spread to avoid artefacts when the expression changes.
Human PBMCs stained with the same panel after different treatments to invoke differential expression patterns.
Fix any variables possible. Time, temperature, blocking, washes and all other aspects of the protocol.
Develop an SOP for the controls and unmixing/compensation. This is as important if not more important than the SOP for the panel itself. A topic for another post, perhaps.
Assess titration data for robustness of dilution.
An unstable titration. Every dilution gives different results with a noticeable impact on brightness and background binding. This means that changing other variables such as incubation time or cell number will have a noticeable impact on staining, resulting in batch or sample-to-sample variation. Human PBMCs stained for 1hr at room temperature, CD4 is fixed, CD40 is used at the indicated dilution factors.
This is what a stable titration looks like. The staining profile is similar or indistinguishable over an order of magnitude of difference in antibody concentration. What's the difference between this one and the one above? The anti-CD40 antibodies for human are provided at ridiculously low concentrations, something like 1-3ug/ml--far too low to achieve reliable staining with short staining times at the recommended test volumes.
Repeat testing using known samples on different days. Can you overlay the data from different days? Do the batches segregate on a tSNE, or is everything lining up nicely?
A 50+ colour panel stained and acquired a month apart to check stability. tSNE with FlowSOM clusters in coloured overlay. No adjustments or batch corrections applied.
Finally, write up a detailed protocol and hand it off to someone else to run. If they're less experienced than you, even better. Did their data look like yours? If not, my rule is that the fault lies in the protocol or the panel, not the person. My panel, my problem to fix.