JC -- thanks for bringing up this topic and for reading our papers so thoroughly
Event Length, as it is now called, used to be named Cell_length in the CyTOF software -- this name contributed to the notion that it is a measure of cell size or volume, but that is not what it is meant to be. In a couple of the quotes you posted, we are simply referring to the parameter by name, not intentionally propagating the idea that it correlates with cell size. I believe the name "Cell Length" was originally intended to convey that this parameter measures the
length of time that the ions from a cell were seen by the TOF. "Event Duration" would have been a better name, in my opinion.
As Adeeb explained, Event Length nominally should not correlate with cell size. The 'cloud' of ions exiting the plasma should be the same diameter (2mm if I recall correctly), regardless of the size/volume of the cell inside, and should expand at the same rate as they move through the ion optics. So you should not rely on Event Length as an indicator of cell size, as you would with Forward Scatter. It sounds like Adeeb has a better indicator for cell size in the works, and that would certainly be nice addition to the CyTOF toolbox!
But there's more to the story about Event Length, which really gets into the nitty gritty of the CyTOF detector and software. In many datasets, Event Length
is correlated with cell size, but not in an intuitive or terribly informative way. When analyzing the data for the 2011 Science paper you referenced, we noticed that the platelets had significantly lower Cell_length than the larger nucleated cells, such as monocytes. This is pretty clear even if you just look at DNA x Cell_length (see attached image). To orient you, the platelets in this image are CD61+DNAlow, and the monocytes are CD33+DNAhi. You can see that the platelets have an Cell_length geometric mean of 25.5 versus 37.7 for the monocytes. So based on this observation, and probably an incomplete understanding of what causes it, that's why we wrote " DNA content, and relative cell size", as you quoted. Although that paper was published just 4 years ago, we were basically still in the Dark Ages of CyTOF -- we have learned so much since then!
But if the ion clouds should be the same size independent of cell volume, as we said above, then why did our platelets have lower cell_length in that dataset?
To answer this question, I probably need to explain a bit about the internals of the CyTOF software and the TOF. My sincere apologies to Dmitry/Vlad/Scott if I bungle this, despite many lessons on the subject
The TOF is refreshing at a rate of about 75000Hz, so roughly every 13 microseconds. Event Length is simply the number of refreshes (aka "pushes" or "scans") on the TOF that were summed for a given ion cloud, aka "cell event". When a cell hits the TOF, it comes through like a wave, with fewer ions in the first few scans, the majority of ions in the middle scans, and then a tail of fewer ions in the last few scans. It's basically a little Gaussian distribution that occurs over about 400 microseconds (30 scans). Since there is always some background noise due to free antibody and electronic noise, the CyTOF software uses the total number of ions
across all channels in each scan to determine if that scan is
above background or not. (Note: That's the default behavior -- it can be changed with the 'custom expression' feature in the software). When you have a bunch of above-background scans in a row, and they fall within the minimum and maximum allowed number of scans in a row, then they are grouped as part of a "cell event" and the sum of their intensities (per-channel) is recorded in the FCS file. The
width of the peak -- aka the number of scans -- is the Event Length.
OK, so why is Event Length sometimes influenced by cell size, as it was in the Science 2011 dataset? Well, if you have a cell that is loaded up with large amounts of metal (i.e. DNA intercalator and many positive antibody channels), the ion cloud contains a lot of metal ions that the detector can see. That's the situation with the monocytes in that example. Platelets, on the other hand, were only positive for one marker (CD61) and very dim on the DNA channel. As a result, the scans in the leading and trailing edges of a monocyte cloud will have stronger signals than a platelet cloud, and will thus be more easily differentiated from the background signal. So as the "wave" of ions from a big, metal-saturated monocyte hits the detector, the software will identify the beginning of the cell earlier and the end of the cell later. This means there are more above-background scans and thus a larger Event Length. All nucleated cell types will all take up large amounts of DNA intercalator, so they will all be discriminated from background at roughly the same efficiency. As a result, you will get very similar Event Lengths, despite slightly different cell sizes, which is exactly what Adeeb showed. Non-nucleated cell types and debris (as long as they are not stained with many other barcodes or antibodies) will be very dim and difficult to discriminate from background. As a result, those will have lower Event Lengths, like what we saw in the Science 2011 dataset (again, see attached image).
I should note that the iridium intercalator itself is not perfectly specific for DNA, and is also somewhat correlated with cell volume, because it sticks to other biomacromolecules. The intensity of metal cell barcoding is also associated with cell volume, because bigger cells have more protein that can take up the barcode dyes. We've noticed that barcode intensities are considerably higher in some cancer cell lines than PBMCs, for example. Unfortunately, although total barcode intensity and DNA are correlated with cell size, they are not really sensitive enough to be useful. Frankly, this phenomenon is more of a challenge than a benefit with barcoding, because ideally everything should barcode at the same intensity! This challenge is what inspired the single-cell debarcoding algorithm in Zunder et al.
Nature Protocols 2015 (
http://www.nature.com/nprot/journal/v10/n2/full/nprot.2015.020.html).
Anyway, I hope that was informative, and I hope I didn't bungle it too badly. And Adeeb, I look forward to your new solution for approximating cell size!
- Erin