Odd results? Trouble Analyzing.

[ndawson]

Hi everyone,

So I've been trying to analyze some data that we just got off the CyTOF and am having some difficulties. I am trailblazing these experiments at our university so you guys are my support! Thank you in advance.

We are looking at healthy donor human PBMCs thawed from LN2. Basically, my main problem is I can't recapitulate the data that I've seen on here and in the Bendall Science paper. I have the scale set to arcsinh as in the Bendall paper. My concern is the lack of "clouds" in my data. Everything is on the 0 when I look at two parameters at once. If I do NC1/2 vs a parameter, I can easily pick out a positive population, but I want to be to look at more than one maker at a time on biaxial plots. The only thing different from the paper (in flowjo settings) is that the Arcsinh "Augment" value isn't in FlowJo X (at least thats what the rep told me) and they did their analysis on cytobank but I don't think its that. I feel like I'm missing something obvious!

Also, how do they have events below the 0 axis? I thought we were dealing with positive integers only.

Any suggestions/comments you guys have are much appreciated.

Thanks!

The plots on the top are mine and the ones below are from the Bendall paper.

troubleshooting cytoforum1

[mleipold]

Hi Nick,

I think there are a few things.

1. The CyTOF software takes all the zeroes in your data and scatters them randomly between 0 and -1. This is partly to bring down the height of the "0" peak so you can see small positive populations even in the presence of mostly-negative cells. So, this is why, if you gate on your negative cells, the median value can be something like -0.3.

2. I think it would also help you if you shrank your horizontal scale down to 1e3 in CD4 and CD8a dimensions (rather than 1e4): this would zoom in on your working range and probably show you what is now a very thin streak near or at zero. If you're using Mac FlowJo-v10, you can just adjust the scaling on those parameters.


-Mike

[ianfrank]

Nick,

It looks like the CD45RA high population (above the 2nd decade) could actually be background or maybe beads. The DVS Calibration beads stick to the cells, and they have to be manually gated out. As for the background, a lot of my issues were solved by just spinning down the antibody cocktail to pellet any aggregates. It really made a big difference. Mike has also suggested filtering the cocktail with a 0.1um spin filter.

Another reason the Bendall data may look different is that they may have just collected more events. Also keep in mind that the CD8 negative data in the Bendall diagram looks like it includes spillover from the n-1 or n-16 channels. This creates a more rounded population, but in cytof data it's actually less desirable. Notice the CD4 negative data is piled up on the axis, and not in a cloud shape. This represents less background noise from other channels.

ian

[AdeebR]

Hi Nick,

I should also point out that the visualization type that you choose can influence how your data looks. I've attached a simple CD4 vs CD8 plot that I drew up in Cytobank showing the same data presented as a density dot plot vs. a shaded contour plot.

You can see that the shaded contour produces more rounded populations vs. the more abrupt lower limit on the dot plots.

Adeeb

[ndawson]

Thanks for your comments, everyone.

Mike, is this randomization something that is not default on the CyTOF software that you have to activate in the settings?

Ian, this sample contains no calibration beads because those channels are taken up by other targets we are looking at. Why do you think it is background above the 2nd decade?

Nick

[mleipold]

Hi Nick,

There are two types of randomization that the CyTOF software can do. One is a default, the other is something you have to select.

1. Scatter the zeros between 0 and -1 is a default. As mentioned before, this helps decrease the height of the "0" peak so smaller populations can be more easily seen (in histogram layout, for instance).

2. The selectable randomization is under the "Convert to FCS" button. You can select the "Gaussian negative half-zero randomization" (GNHZR) box. This takes all the zeroes and scatters them around zero according to the Gaussian parameters you input. This has 2 effects: first, it further brings down the height of the "0" peak, thereby comparatively increasing the height of all other peaks. Second, it makes the negative data spread +/- zero in a way that more closely resembles fluorescent flow data.

I personally hate the GHNZR. To me, it's injecting noise into your data for no reason other than aesthetics. And on a more practical level, your "negatives" can start spilling into your "dims", especially when you have a lot of zeroes or you choose Gaussian parameters that give you a wide peak.


Mike

[ianfrank]

Nick,

When I look at the plots, I imagine the boxes as the CD45RA positive populations and anything above that as background. That's just what it looks like to me. It's rare to have events that "bright" on the cytof.

Ian

[ndawson]

Thanks for your comments.

Mike, I agree with your opinion of GNZHR. I went and I tried the GNZHR - didn't like it. Changed the shape of my plots and I didn't get the look I was hoping. I think I will plug away at my analysis using contour plots like Adeeb suggested as they seem to get the point across without manipulating the data.

Ian, I see what you mean by background. Though I'm not quite sure how to discriminate between real/background - these plots already have backgates on singlets, nuclear stain gates, viability and CD3+. Also, I'm not sure I completely agree with your gating at anything above 10^2 is background - shouldn't points up to 10^3 potentially be real? So I guess my real question is - is there a control or criteria I can follow that is convention so that I can say that something is/isn't background?

[mleipold]

Hi Nick,

To help decide whether something is "true"/real vs background, I would advise looking at other markers simultaneously.

For instance, use CD27 or CCR7 vs CD45RA: that will help you figure out whether your CD45RA-hi cells correspond to the usual Naive/CM/EM/Eff subsets. See attached example plots.

You can also try gating on the CD45RA-ultra-hi (above CD45RA+CCR7+) and then look at other markers like CD20 that would not be expected to be co-expressed. In other words, if the ultra-hi cells have marker combinations that don't make sense in terms of lineages, then they're likely background.