FAQ  •  Register  •  Login

Long CyTOF runs

<<

emallen

Participant

Posts: 18

Joined: Thu Oct 23, 2014 6:10 pm

Post Mon Jun 08, 2015 9:24 pm

Long CyTOF runs

Hello, I apologize if someone has asked this before... I couldn't find any previous posts on this subject.

I'm wondering whether anyone has experience with long runs on the CyTOF... I've heard that the Nolan Lab will sometimes run their CyTOF for 12+ hours at a time, but I begin to lose a problematic amount of signal (especially for dim markers) at around 5 hrs. Is there anything I can do to preserve detector sensitivity for longer runs? Alternatively, could the detector be 'restarted' intermittently to restore sensitivity?

On a related note, does anyone have experience running samples for one experiment over multiple days? Right now we're running one donor per day, and using an internal control donor to compare between days. However we do see variability between each run, so we're looking to 1) determine the best way to process and validate our data generated over multiple days, and 2) find the likeliest sources of variability.

Any help would be greatly appreciated!

Thanks,
Elyse
<<

mleipold

Guru

Posts: 5792

Joined: Fri Nov 01, 2013 5:30 pm

Location: Stanford HIMC, CA, USA

Post Mon Jun 08, 2015 11:44 pm

Re: Long CyTOF runs

Hi Elyse,

There are a few things that affect machine performance over very long run times.

1. Detector gain optimization. Over long runtimes, the detector starts seeing enough ions that its Gain starts to decrease. Remember, there's a finite number of ions that a detector can see before you have to adjust its voltage, and a finite number of ions it can see in total before it has to be replaced.
2. Ion peak TOF drift. Over long runtimes, the TOF of the ion peaks can start to drift, usually to higher TOF. From what I've been told, this is due to slight decreases in vacuum efficiency: lower vacuum means more "other stuff" for the ions to run into, which delays their arrival. Akin to taking longer to cross an empty room vs an increasingly crowded room.
3. Dirt buildup on machine parts. This is particularly on the cones. Your Current Setting may change a bit depending on your sample (again, usually to higher Current).

Cleaning the cones will help with #3, which will help a bit with ion transmission. Minimally, re-tuning the instrument periodically throughout very long runs can help keep you in the optimum setting.

If you re-tune, this will also help with #1, as it would let you do Tb gain calculations and therefore adjust Detector Voltage if necessary to get back in the Tb gain = 4-5 (for a CyTOFv1) range.

Tuning will also allow you to check your ion peak TOF, and adjust the acquisition window (Cs133/Ir193 values in Mass Calibration) if needed. You'll probably need to change it back the next day, as the vacuum will have improved overnight.


My understanding is that if you re-tune a CyTOFv2 using the automatic tuning, all of these things get checked and changed during the process.

So, if you get in the habit of re-tuning every 5-6hr or so, it should help to smooth some of this out.


Mike
<<

kunicki

Contributor

Posts: 38

Joined: Thu Apr 13, 2017 8:46 pm

Post Mon Sep 18, 2017 8:06 pm

Re: Long CyTOF runs

Hello Mike,

You bring up some very good points that affect performance over long runs. I was curious if you could expound on this topic being it's been a couple of years, and whether you've found any other issues that could impact performance for 12+ hour runs.

It's important for us to know what is an optimal procedure depending on the sample type and scarcity of a given target molecule on your cells. You mention tuning every 5-6 hours, but have you seen a large improvement in performance if one were to tune every 2-3 hours instead?


Best,
Matthew
<<

mleipold

Guru

Posts: 5792

Joined: Fri Nov 01, 2013 5:30 pm

Location: Stanford HIMC, CA, USA

Post Tue Sep 19, 2017 3:35 pm

Re: Long CyTOF runs

Hi Matthew,

I'll definitely defer to Adeeb, Zach, and a bunch of other people who often do long barcoding runs, but here are some general guidelines:

1. I personally have not found it necessary to clean the cones during long runs (ie, shut down, remove them, sonicate or use Brasso, rinse, then reinsert). This mostly affects the Current setting, where I haven't seen much drift even over the course of a week as long as samples are well washed in MilliQ to remove residual salt.
2. If I am doing long runs (>6hr), I typically stop, clean with Wash, then rerun Tuning (then Wash to remove Tuning and then MilliQ rinse, of course). The time interval depends a bit: I'd guess every 4 hr is probably sufficient, based on my experience. So, if you're doing an 8hr run, stop at the 4hr mark, retune, and then resume. You can shift that time interval a bit as needed: for example, if you have a 9hr run, I would retune at 4.5hr and be done with it, rather than at 4hr, then 8hr. If you have a 10hr run, I would probably do 3.5hr, 7hr; or just one at 5hr.
-Some of this will depend on the instrument model you have: A Helios drifts a lot less than a v1, and I think less than a v2.
-While you can certainly retune as often as you want, I don't think you'll see a benefit retuning any more frequently than every 4hr or so. And, at ~15min/retune + 10min or so Washing before and after, you're approaching 30min per Retune.
3. It is critical that you redo Mass Calibration, Dual Slopes, and Detector Voltage when you retune. Those are the main things that will drift. On a Helios, there is usually a Quick Tune Protocol that basically covers those things; however, I've seldom been able to get it to work properly.....therefore, I just do the whole Full Protocol which takes about 15min.
4. If you are running a barcoded sample, 4hr may be one sample or multiple samples. Since MilliQ water is so harsh even on fixed cells, samples are usually more stable as cell pellets with minimal overlay (resuspend and filter just before putting on the machine). Therefore, it is often a good idea to split one giant BC sample into a few smaller pellet aliquots. Here's a thread on the topic: viewtopic.php?f=3&t=671&p=2260&hilit=normalization#p2260
- this does not inherently mean you have to run them as separate files (though you may wish to); you can always add in the next aliquot to the sampling tube on the fly.


Mike
<<

CRStevens

Master

Posts: 60

Joined: Thu Jul 17, 2014 5:07 pm

Post Tue Sep 19, 2017 4:38 pm

Re: Long CyTOF runs

Hi Mike,

Thank you for your insight on this and would like to hear more opinions on this.
I have battled with this topic on multiple occasions and have sided on the side of I did not want to re-tune my machine mid-run if I didn't have to. First we all spike in the EQ4 beads, so in a barcoded sample if you are running 1-2million cells per file, then you can track through all of these files what your median EU151 or 153 bead signal is. From multiple runs you can even start to draw a conclusion of how long you can run before you get this signal into the "dangerously" low range. My initial thoughts on not re-tuning mid run was every time you re-tune it is really like resetting the machine. This is essentially like splitting your run between two days because once you re-tune your gases, voltages, current will all change. I've said for many years if we could tube to a median bead value, then get the gases/current to pass under those conditions, then when we went to retune we could try to hit that median again. This is akin to using rainbow beads in flow cytometry to ensure consistent PE signal from day to day and run to run.

My big concern is you have split your barcoded samples into 6 different runs and each technically should be normalized individually (based on your tuning settings) and then each one of those runs would have to be normalized to the first run (to my knowledge this would have to be done manually).

Please tear this apart or confirm, as this is a really important topic as we all are running large barcoded studies and these runs take 8-12h long.

-Chad
<<

mleipold

Guru

Posts: 5792

Joined: Fri Nov 01, 2013 5:30 pm

Location: Stanford HIMC, CA, USA

Post Tue Sep 19, 2017 5:13 pm

Re: Long CyTOF runs

Hi Chad,

There are a few topics here:

1. Overall instrument sensitivity: if you have some signals that are dim (only slightly above the cutoff for background), instrument sensitivity (mass calibration and DV, mostly) will affect your fundamental ability to detect those. If a dim signal drops below this cutoff, then it can't really be properly normalized using the beads: your signal is lost in the weeds, no matter if you then normalize the weeds.
- If you couple this to signal decrease due to sample degradation (see previous link to Bill and Adeeb's comments re fixative and sample aliquotting), it's even worse.

2. The main things that change on same-day Retuning are Mass Calibration and DV (and then the resulting Dual slope coefficients, of course). I generally haven't seen NG, MG, or Current change when I retune (assuming there are no clogs, no salty sample, etc).

3. I'm not sure if I understood what you mean by "each technically should be normalized individually (based on your tuning settings) and then each one of those runs would have to be normalized to the first run (to my knowledge this would have to be done manually)". It's not obvious to me why you couldn't throw all of them into the same normalization. Also, between-days, if you're using the MATLAB normalizer, you can use the Beads files from the first plate as additional information for subsequent day-plates. This is also a feature in Premessa (Federico's R package: viewtopic.php?f=3&t=762&p=2307&hilit=Premessa#p2307), and prevents you from constantly having to renormalize with each new plate.

4. By "running large barcoded studies and these runs take 8-12h long", do you mean that you're doing one sample that lasts for 8-12h, or that's just one day's work (but perhaps 3-6 samples)?


Mike
<<

BjornZ

Contributor

Posts: 43

Joined: Fri Jul 10, 2015 1:04 am

Post Wed Sep 20, 2017 6:16 am

Re: Long CyTOF runs

Mike's suggestions are all great.

For my largest dataset (~5M cells per donor * about 200 donors) I cleaned the clones with Brasso between every donor (two or three donors a day). This only takes a few minutes, short enough that I left the heater on. With auto-tuning (or fast manual tuning), this comes out to only about 20 minutes of downtime between each ~4 hour sample. (This was admittedly probably excessive; we would have been fine cleaning once a day, and just re-tuning the DV, mass calibration, duals and current, but did it for procedure.) The glassware we only cleaned twice a week.

Ahead of time we set acceptance criteria for the instrument (>2M Tb159 dual counts, <2% Ox, dual slope values, etc), and this routine kept us in "spec."

Chad and Mike both mentioned normalization in the later posts -- I normalized all of the files to the beads from the first ~30 donors using the Matlab normalizer.

One of my colleagues used the technique mentioned in the first post, wherein a single donor was included in every barcoded sample for about 42 barcoded tubes. I think this worked well for him, but I forget if he ended up normalizing with it, or just using it to measure technical variation and using that as a threshold for calling out biological variation. Feel free to email me if you want to talk to him and I'll put you in touch (zbbjornson at gmail). It would be an interesting sample to look at on its own -- I'll see if I can coerce him to put some data up in the next few weeks.

-Zach
<<

mleipold

Guru

Posts: 5792

Joined: Fri Nov 01, 2013 5:30 pm

Location: Stanford HIMC, CA, USA

Post Wed Sep 20, 2017 2:49 pm

Re: Long CyTOF runs

Hi Zach,

That would be great if your colleague would comment or post data. Katja Kleinsteuber's spike-in control was used as a QC check (viewtopic.php?f=10&t=543&p=1764&hilit=kleinsteuber#p1764), but not really for normalization.

At CYTO 2016, Jeff Hokanson (then at MD Anderson) spoke about the CLEAN algorithm he had developed to use a spike-in control as a normalization/batch correction method. Unfortunately, as far as I can tell, CLEAN hasn't been published yet, and according to his Github page, Jeff is no longer at MD Anderson, so no idea when/if it will be published.

However, for those that are interested in the CLEAN algorithm, his CYTO 2016 talk is available as a PDF: https://github.com/jeffrey-hokanson/jef ... TO2016.pdf


Batch correction/normalization is definitely something we need......


Mike
<<

bc2zbUVA

Contributor

Posts: 22

Joined: Thu Nov 19, 2015 4:23 pm

Post Wed Sep 20, 2017 5:53 pm

Re: Long CyTOF runs

I have had excellent success dealing with (though not explicitly removing) batch effects using edgeR for estimating differential frequency. If you have barcoded samples with a balanced design across multiple runs, edgeR has no issue with handling batch effects. The bioconductor package cydar puts forth a framework for using edgeR with high dimensional cytometry data, but you don't have to use their hypersphere clustering method, you can use any clustering algorithm you like as long as you can coerce it to a feature table.
<<

vtosevski

Contributor

Posts: 44

Joined: Wed Nov 20, 2013 12:50 pm

Location: Zurich, Switzerland

Post Wed Sep 20, 2017 7:56 pm

Re: Long CyTOF runs

Hi all,

let me just chime in... I can definitely see an argument in favor of re-tuning during the day to keep the instrument within the required spec. However, we've seen what appears to be the change in mass response as we went about this. In each barcoded tube we usually have a calibrator sample. Even though samples before and after tuning were bead-normalized together (and bead channels, thus, looked comparable) we've seen non-bead channels showing up with different intensity on calibrator samples... We're just looking into this, but I thought it's worth mentioning. Anyone else seen something similar? Do you even look for that kind of anomaly?

Vinko
Next

Return to CyTOF general discussion

Who is online

Users browsing this forum: No registered users and 7 guests