I have a lot to dig through right now and some of the data summaries say "confidential" on them. I'm trying to get a handle on dealing with a new to me (and VERY large) data type and I've got others stuff ahead of it in the queue. I can, however, drop some first impressions that no one will mind.
What happened: We sent single human hepatocytes from a liver study that we have heavily characterized in multiple ways (held up publishing because we don't get a lot of healthy people donating us livers here, and it's tough to draw comparisons on just livers so bad that people are trading them in for newer ones) on multiple systems to SCIEX in California and they ran these cells on an EvoSep Eno (picture above is dyslexic) using a 40SPD Whisper Zoom on an IonOpticks 15cm column. There was some optimization to get the best parametes for the 8600. (Slowing it down some to get higher sensitivity while still maintaining a full cycle every 1.2 seconds).
The resulting output files are (by my math) achieving true cycle time of 265 Hz which is the fastest acquisition time I've personally ever seen on a commercially available instrument, either in my hands or in data publicly deposited on a repository. This is even more impressive when you consider the chemist running the instrument slowed it down to increase the signal on these cells, as mentioned above.
These files are large. 72 single cells returned nearly 2.5 TERABYTES of data. Karl warned me they were big. They are BIG. Somehow they still process relativly quickly. Even when they're on an old fashioned spinning hard drive (cause my biggest SSDs are 4TB, which might not leave room for scratch/temp files) and not using every resource on an (admittedly lousy) Intel Ultra 9, (no joke, this processor sucks compared to the much older AMDs in the lab) DIA-NN is something in the 40 minutes per file. I'm clearly used to the 4 dimensional data from the TIMSTOFs and that extra dimension adds an extra layer of requirements to any analysis. Same 40SPD single cells from the Ultra2 in DIA-NN are probably 20-25 minutes per file, even though they are 10x smaller.
Preliminary data is getting us into the 2,000 protein groups/cell range, which isn't quite where we are with the Ultra2 but it better than just about anything else I've ever seen. There is a wide range in IDs in the different software and how we employ it. The histogram above from SCP-Viz is showing the cells from the best plate analyzed with the parameters and tools that provided the highest number of IDs. I wouldn't stake my watch and warrant by the mean displayed above. While the very ethical chemist who ran the cells took our cell sorting report and removed any doublets from the single cell report (and set them aside for us for interpretation) these hepatocytes were not separated by ploidy. Boring liver facts -As people age hepatocytes have generally increased DNA content (more cells have more DNA than the normal 2n) and this patient typical age for a sitting US Senator. I suspect that cell with 4889 protein groups will have at least 2x Histone H4 compared to the others. You don't double a cell's DNA content without increasing the protein content of the cell. Doubling the DNA probably doesn't double the protein content, but it does increase it by a lot. Smarter analysis on that here.
Obviously, there is a paper in preparation here and a solid pile of work still to do, so interpret my typing here with a pound of salt, as always. (We've got 13 months of TIMSTOF Ultra2 data that needs submitted in a series of manuscripts and editor/reviewer responses before the end of the year, this may need to come after).
However - if you've seen some single cells ran on the 7600 (including data I presented at iHUPO Cancun and maybe some other places) you might not be thinking ZenoTOF as a viable option for ultra-low load or single cell proteomics. My opinion this morning is that this budget friendly instrument deserves a second look!
A lot has happened since the original 7600. Mine didn't have nanoflow. Well...it did...but it sucked. The source design required a very large post-column dead volume. Which, from this recent paper, you can see what a bad idea that is. A while back IonOpticks fixed that problem and that allows nano and EvoSep coupling.
My 7600 didn't even have ZenoPulse DIA when the flood destroyed it. So I was running ZenoPulse DDA and microflow (5 uL/min) AND I was using the Waters M Class? H Class? The new one. And that HPLC COULD NOT pick up every bit of liquid from the bottom of a plate. The needle has a diagonal cut, so it always leaves some behind. So....here we are a few years later with all those things fixed and the 8600 which not only is higher sensitivity but it has the sliding window quad ramping DIA thing. I forget that it's called on this.
While I'm still keyboard diarrhea here - I've taken 2 of the random data files and drastically reduced their size. My first 2 attempts were to use ProteoWizard to only keep the 500 most intense fragment ions in each spectra. This reduced files from 30GB to around 7GB and only reduced the number of precursors and protein groups identified by about 20%. Early in my work with the TIMSTOFs when I was trying to get used to having SO MUCH INTRASCAN DYNAMIC RANGE I'd break those MS/MS spectra into 100 amu bins and only keep the X most intense fragments in each spectrum. The tools I did that with aren't Windows 11 compatible (lolz) so I might need to fire up a PC I'm not supposed to use anymore and set it chugging on a similar data reduction strategy.
A lot of words, but I do think this new(ish? wait. was ASMS Baltimore last year?) box deserves some hype.
No comments:
Post a Comment