I've been composing this post in my head for a couple of weeks now, but have been too busy to sit down and write it, so it shouldn't really surprise me that it popped up in a dream. However, a new twist was added via my unconscious mind (which I'll get to later). So, the original post was all about how I've pretty much given up on compensating using cells, and if you're not using beads, then you're pretty much setting yourself up for compensation failure (unless of course you're using things like PI, or mCherry, or the like). I mean, the whole point of 'autocomp' is to take the subjectivity out of compensation, and using objective mathematics to correct for fluorescence spillover. However, every single time I've done autocomp using cells, it just doesn't look 'right' and I end up tweaking the values just a little bit. I've come to terms with this fact, and have pretty much settled with this sub-par situation. But, if you're trying to teach someone about compensation, and you introduce this 'autocomp' feature, it makes for a pretty awkward conversation when you then go on to say, "Well, just adjust the values a little bit until it looks right." So, I typically recommend people do their compensation with beads. For many of my users, the thing that prevents them from doing this is cost, or maybe a bit of skepticism in changing the ways they were taught to do their staining. The reasons why compensating using cells doesn't always work are many, but let me just outline a few for you here.
1. Insufficient frequencies of both positive and negative fraction to make a statistically significant regression of means. If in your stained cell sample, you only have a 0.1% positive fraction, the mean of that population in the spillover channel will not reach a high enough statistical significance until you collect millions of cells. No one is going to collect millions of cells on their single stain control. This also holds true when all your cells are positive for your single stain control, and you have a really low negative (or low) population.
2. Poor resolution of the positive fraction. Sometimes you will not have a clear positive population, so making a gate around the positive fraction for performing compensation is difficult. If you end up encircling some of the high autofluorescent cells that you mistakenly call positive, your compensation will surely be off.
3. Non-linearities at the extremes can lead to inaccurate compensation. If you're compensating using an unstained (or negative) fraction that is at the very low-end of the scale, or if your positive fraction is at the very high-end of the scale, you're likely using a data point in the non-linear range of the log scale. Since compensation algorithms are basically relying on the fact that your range of analysis is linear, you're going to run into lots of problems if you're using "unstained" cells as your low-data point, or really bright cells as your high data point. Side bar: Yes, I know, your comp control should be at least as bright as your sample staining, blah, blah, blah. However, the only reason why this is the case is because of non-linearities at the very high end of the scale. If all your staining fell within the linear portion of the scale (let's say 1.0 logs to 3.5 logs), then this isn't necessarily a problem. You can take any two points within that range, and create a regression line that will model the entire scale. No-scale is linear enough, especially at the extremes, so the 'rule' of a maximally bright comp control needs to be adhered to.
4. Mismatched autofluorescence between positive and negative. If I stained my leukocyte prep with a monocyte marker (CD14, for example). All my monocytes will be positive. For this single stained comp control, what should I use as my negative? Many people would simply use the negative lymphocytes or granulocytes, and many people would end up with a poor compensation matrix. For channels where autofluorescence is a factor (mostly the green/yellow detectors off the blue and lower laser lines), the positive fraction's autofluorescence should match the negative fraction's autofluorescence. This is, evidently only necessary when you're using cells for compensation, and you have a mixed cell-type sample.
So, there are certainly lots of pitfalls when using cells for compensation, which is why using beads is a good idea. To solve many of these issues, simply using an antibody capture bead at two fluorescence levels should do the trick. You'll notice I said two fluorescence levels, and not one positive and the 'blank' bead. Using the blank bead can lead us into issue #3 above, so I prefer to use the bead at a saturating level of antibody and maybe 100-fold less, to create a high and low peak. In the end the peaks will fall around the 3.5 decade range and 1.5 decade range. Use these peaks as your 'positive' and 'negative' values in your favorite autocomp program, and voila, perfect compensation. Of course, these beads are run at the appropriate voltage that is set up according to your cell type.
But, what about the twist? The twist is, that you don't need to only use beads as your capture matrix. You could use cells! I know, I know, I just went on and on about NOT using cells, now I'm telling you to use cells, but wait, let me explain. Take a thymus, get all your non-tandem antibodies in CD4, stain them at two concentrations, fix them, and stick them in the fridge. You now have ready-made compensation controls that are much cheaper than buying capture beads. Why thymus? They're the closest thing to beads; pretty much homogeneous, so we don't have to worry about autofluorescence mismatch, they're almost all CD4 positive, so that makes it easier to create two nice peaks, and you can get a boatload of them from a young mouse. On top of all this, we gain the ability to use other things besides antibodies. You could stain them with many of your dyes for a comp control, PI, DAPI, CFSE, etc... Something you can't do with beads. For tandems, I'd stick with capture beads.
Ok, there you have it. If you've made it this far reading through all my gibberish, let me know what you think.