FAQ  •  Register  •  Login

Tuning sensitivity and data quality

Forum rules
Please be polite and civil. We know that troubleshooting is vexing...
<<

SteinErik

Participant

Posts: 9

Joined: Fri Sep 30, 2016 6:18 am

Post Tue Feb 23, 2021 10:17 am

Tuning sensitivity and data quality

We have now for the last two years been concerned about lower tuning results on the Helios machine at our core facility. It was installed in late 2015, and for the first years operated quite steadily at around 1.6-1.4 x10^6 Tb dual counts (DCs). Through some up and down periods, it has now settled at around or just below 1.0 x 10^6 Tb DCs. We consider this to be low sensitivity, as we are used to much higher values. Even in a newly washed machine with brand new cones we cannot get any noticeable (or reproducible) increase in tuning sensitivity.

What is the normal level of 159Tb in the tuning of a Helios machine? Does our machine indeed have a sensitivity lower than what can be expected?

We did after long (!) discussions with Fluidigm get our quadropol changed, and the sensitivity increased to around 1.3 x10^6 Tb DCs immediately. Disappointingly, the machine quickly declined in tuning sensitivity and is now back at a steady 1.0 x 10^6 Tb DCs. However, taking advantage of the planned service, I prepared a four-point titration experiment and froze down two aliquoted sets of samples. One sample set was thawed and run on the machine before changing the quadropol, and the other sample set was run right after, when the machine had a substantially increased sensitivity.

From the titration data (see powerpoint), it is clear that an increased tuning sensitivity of the machine directly translate into increased antigen expression (DCs) on single cell data. Likely, a higher sensitivity also decreases the number of antigens needed on each single cell for it to be detectable. This might result in more antibodies to be usable in our panels, as some antigen could be just in the detection limit.

We showed these data to Fluidigm and argued that lower tuning sensitivity provided poorer data with reduced sensitivity. They were quick to reassure us that we should just use their normalization algorithm which will align (increase) the poor data to their reference bead data, and thereby fixing the lower sensitivity issue. Our immediate concern that increasing the values will potentially drag very low values (potential noise) into an area that might be not be obviously noise, may therefore disturb analysis of the samples who’s lower range extend down into this more uncertain range. Just as a note, we have no problem with the normalization of higher values, here, the algorithm works nicely. But as we and many of my collages are interested in signal transduction antigens, who are often weakly expressed in steady-state/baseline primary material, the lower end of the scale is very important! In a nutshell, higher sensitivity allows us to measure (with increased confidence) lower levels of e.g. pSTAT5 in our patient cells. Indeed, variation between patients will inevitably result in some of the data being on the really low end of the spectrum, and here we believe that it is important to get as much out of the cells as possible (have a look at the 1:800 dilution of CD223 in slide 11 of the attached pdf)

I am just very curious about what the mass cytometry community has to comment on our views on tuning sensitivity and data quality.
Attachments
Titration and Tuning_Cytoforum.pdf
(1.77 MiB) Downloaded 333 times
<<

mleipold

Guru

Posts: 5796

Joined: Fri Nov 01, 2013 5:30 pm

Location: Stanford HIMC, CA, USA

Post Tue Feb 23, 2021 9:29 pm

Re: Tuning sensitivity and data quality

HI Stein-Erik,

One initial question: your Experimental Setup slide says that you had 18 antibodies held constant and then 19 other antibodies that were titrated.

I assume from your example that CD4 is one of the Constant antibodies, but could you give lists of the Constant and the Titrated for clarity?


Mike
<<

SteinErik

Participant

Posts: 9

Joined: Fri Sep 30, 2016 6:18 am

Post Wed Feb 24, 2021 12:14 pm

Re: Tuning sensitivity and data quality

Hi,

Thank you for asking! Uploaded a new slide with the panel information added. These antibodies were just a initial selection we had planned to use i a particular project. So not a polished panel at all, many of them turned out were suboptimal on these samples.

I also added a legend describing the different populations that were plotted in the slide showing the titration of CD223.

-SEG
Attachments
Titration and Tuning_Cytoforum_2.pdf
(97.95 KiB) Downloaded 323 times
<<

mleipold

Guru

Posts: 5796

Joined: Fri Nov 01, 2013 5:30 pm

Location: Stanford HIMC, CA, USA

Post Wed Feb 24, 2021 6:03 pm

Re: Tuning sensitivity and data quality

Hi Stein-Erik,

Thanks for the additional info.

Regarding normalization and "boosting" noise: yes, this can happen. It's explicitly shown in Fig 4 Day A CD8 signal in the Finck et al paper (http://dx.doi.org/10.1002/cyto.a.22271). That's one of the reasons I really like that paper: it shows all the upsides and downsides of normalization.

Regarding your question about 89Y and 209Bi normalization (being far from the EQ beads): in my experience, this hasn't been too much of an issue, but yes, the further you are from the data points, the greater the extrapolation error. I think this is one reason why Fluidigm is testing new beads that contain 89Y and 209Bi (viewtopic.php?f=10&t=1868)

Regarding your overall signal intensity issue: I notice a couple things:
1. On Slide 2, you show screenshots of your Tuning results from 02 Sept and 21 Oct.
a) Your Makeup Gas value seems *really* off for a Helios instrument. In particular, the 0.81 for 02 Sept seems extremely high. For comparison, my two Helios instruments run in the 0.46-0.55 range: I know that if they're outside of that, there's a problem such as a mis-assembled spray chamber leaking MUG, a misfiring nebulizer (where a high MUG "shoves" the angled spray back into the center), a partially cracked injector (enough to leak gas, but not enough to be visible or to let in enough room air to keep plasma from igniting), and a few other things.
- all of this can affect your overall sample introduction into the instrument, so how do these MUGs compare to your historical (2015-2019) values?

b) Your Tb and Tm values swap maxima. On 02 Sept, Tm was higher, while on 21 Oct, Tb was higher. I don't want to read *too* much into this because this can just happen from day to day. However, overall ion alignment through the turnblock into the quadrupole is something that your engineer can adjust, and affects this value (as well as can affect your overall signal)


2. How did you do the Premessa norm? By this, I mean did you norm the 02 Sept files together, and separately norm the 21 Oct files together? That's kind of how it looks on Slide 8, since the Premessa and Raw 02 Sept data points look almost identical.
- The reason I ask: it's a bit of an odd comparison, in some ways. Raw files are Raw files, so comparing one day to the other is fine. And since Fluidigm uses exact values for the bead isotopes, you also have a common comparator for the Fluidigm-norm files. But since Premessa (Finck style) uses the bead signals within each set of files, you'd kind of have two different baseline/comparator if you did each Day separately.
- note: this wouldn't necessarily change the overall conclusion of your slides that Premessa and Fluidigm give different results and that 02 Sept is distinctly different from 21 Oct....more of an issue of how much they differ.


All that being said: based on what you've shown, I would agree with you that there appears to be a signal intensity problem in your machine that you need to address with Fluidigm.


Mike
<<

mleipold

Guru

Posts: 5796

Joined: Fri Nov 01, 2013 5:30 pm

Location: Stanford HIMC, CA, USA

Post Wed Feb 24, 2021 6:13 pm

Re: Tuning sensitivity and data quality

One other minor thing to check: on the left side of the cone interface, there's a metal-reinforced translucent rubber tube coming down, that connects to the Interface pump.

I've had 2-3 instances where the top of that tube (where it connects to the cone interface) gets heat fatigue and partially collapses, even with the metal reinforcement. In most cases, it kind of spirals or "irises" shut, leaving a reduced diameter center to the tubing, rather than collapsing like you stepped on it and closing completely.

If this happens, the vacuum level at the cone interface will be weaker than it should be. Since the vacuum behind the sampler helps "suck in" your ions, if that vacuum is weak, you'll have lower signal of both Tuning Solution and Cells/Beads.


It's really rare, but it can happen, and it can be tricky to catch, since the tubing connection is half-hidden behind the left side flange of the cone interface.


Mike
<<

SteinErik

Participant

Posts: 9

Joined: Fri Sep 30, 2016 6:18 am

Post Sat Feb 27, 2021 7:05 am

Re: Tuning sensitivity and data quality

- 89Y and 209Bi normalization.
I think that the data of CD45-89Y in slide 8 shows that, on our machine at least, the normalization of 89Y at least does not work well at all. At least in the way of normalizing the signal to match that of the bead references of Fluidigm. But it could also just be this individual experiment. I will ask around here in Bergen if others have some data we can have a look at.

- 1. On Slide 2, you show screenshots of your Tuning results from 02 Sept and 21 Oct.
a) Your Makeup Gas value seems *really* off for a Helios instrument.

This is WB and CAS and should therefore be ~0.15 higher. (I took a bunch of slides out to make the file size smaller, sorry about these missing details!) But I have to say that it is interesting that these values (0.81 in 02SEP and 0.69 in 21OCT) are on the extremes of what we are used to.

We had a service in November 2019 where the MUG-values went from ~0.7+/-0.2 to ~0.79+/-0.2. When the quadropol got replaced the MUG-values dropped (back) to the values you see in the powerpoint for 21OKT, around ~0.7 +/- 0.2. I will try to make some graphs of this.

- b) Your Tb and Tm values swap maxima.
Tb and Tm have swapped min/max relatively frequently (maybe once or more in a month). Fluidigm did not seem to be worried about this at all. It has stabilized after last service (Tb ~10% higher than Tm) having an adjustment of the TurnBlock. But I think this also ties in with the 89Y normalization issue. The sensitivity across our mass range is seems to me to very unstable. I will try to plot this as well


- 2. How did you do the Premessa norm?
I normalized each day separately. The reason is that I wanted to see how the data was “boosted” when we just normalize progressive loss of sensitivity on each day using premessa. This would then contrast nicely to the Fluidigm approach where we do the same and then align the data to external reference values (the bead passport). I wanted to isolate effects that is from the alignment of the data to the reference values, and not just “generic” normalization to account for machine drift over time.

The reason I wanted to understand this nuance is that (if I understand Fluidigm correctly) we should not be worried about our (perceived) lower sensitivity, because we can just use their normalizer to fix that. Meaning, if we run some samples on a single day, we should not use premessa etc, but Fluidigm to “fix” the data. In our case, clearly boost the values. And this is the core of the issues I guess… Is this “boosting” okey when I want to be able to measure data at around 5 dual counts per cell? Is it impossible to accurately measure data here?

Btw, I guess that doing premessa on both days together would produce something very similar to the Fluidigm result, as it seems like beads on 21 OKT are very similar to the passport (not adjusted much after Fluidigm normalization)?

- About the Metal-reinforced translucent rubber tube
This has already been changed after we saw a mechanical break 10 cm below the cone interface. I could not see any changes around the connection when I had a look now, but we will certainly keep an eye on this!
<<

mleipold

Guru

Posts: 5796

Joined: Fri Nov 01, 2013 5:30 pm

Location: Stanford HIMC, CA, USA

Post Thu Mar 04, 2021 7:10 pm

Re: Tuning sensitivity and data quality

Hi Stein-Erik,

Regarding the larger 89Y variation in Slide 8 relative to other channels: On top of the extrapolation down to 89Y, one point that's sometimes missed with the Fluidigm normalizer is that it's predicated on the idea that *your* machine has the same relative mass response curve/sensitivity as the machine(s) Fluidigm used to create the Passport values.

However, different instruments have different mass response curves. You can see this clearly in Fig 2 of the 2015-Tricot et al-Cytometry A paper ( http://dx.doi.org/10.1002/cyto.a.22648 ) where Lab N had a different maximum sensitivity than Lab C or Lab F. Also, Lab C and F had the same mass of maximum sensitivity, but decreased in signal at different rates at masses above or below that.

Additionally, you can see this in the Biosurf companion website to the 2018-Olsen et al-Cytometry A paper ( https://biosurf.org/cytof_data_scientis ... inck_et_al) ), where the red CyTOF1 mass response curve was distinctly different from the Helios instruments or the CyTOF2 instrument. This affects normalization: in particular, the CD16-149Sm signal changed dramatically between Raw vs Fluidigm vs Finck.

So, yes, if your instrument has a different sensitivity maximum or drops off faster or slower to the low or high mass relative to Fluidigm's, the normalization can affect your data.


Mike
<<

mleipold

Guru

Posts: 5796

Joined: Fri Nov 01, 2013 5:30 pm

Location: Stanford HIMC, CA, USA

Post Wed Jun 01, 2022 11:00 pm

Re: Tuning sensitivity and data quality

Hi Stein-Erik,

The Premessa folk are working on incorporating EQ6 as a normalizer; here's a discussion about a fork (and which I got to work): https://github.com/ParkerICI/premessa/issues/39

Since EQ6 contains 89Y and 209Bi, you could compare EQ4 vs EQ6 normalization on your 89Y and 209Bi probes (assuming you have a sample with EQ6, of course). FYI, there are enough different isotopes between EQ4 and EQ6 that I was able to run them in the same sample.


Mike
<<

mleipold

Guru

Posts: 5796

Joined: Fri Nov 01, 2013 5:30 pm

Location: Stanford HIMC, CA, USA

Post Fri Sep 16, 2022 6:43 pm

Re: Tuning sensitivity and data quality

Hi all,

Looks like the Premessa folk have recently updated it to include EQ6 beads (they call it "XT") as well as the Beta beads (original Finck paper beads).

https://github.com/ParkerICI/premessa/r ... tag/v0.3.4


Mike

Return to CyTOF troubleshooting

Who is online

Users browsing this forum: No registered users and 11 guests