FAQ  •  Register  •  Login

Intensity of Normalization Beads

Forum rules
Please be as geeky as possible. Reference, reference, reference.
Also, please note that this is a mixed bag of math-gurus and mathematically challenged, so choose your words wisely :-)
<<

FlowjoVA

Participant

Posts: 13

Joined: Tue May 24, 2016 4:49 pm

Post Tue Feb 28, 2017 6:35 pm

Intensity of Normalization Beads

Has anyone looked into the possibility of using Eq beads of lower intensity for normalization. I wonder whether the brightness of the beads is not effective for normalizing data in the low end of intensity. I have attached a printout of data which was collected over a relatively long period of time, showing the raw data for the Eq beads and then the normalized data using both the Matlab method (using the average) and the Fluidigm method. While both normalization approaches seem to do a fine job for the Eq beads, it seems less than normalized on the actual data, and seems proportionally less normalized the lower the signal intensity. Is this something others see? Is this a visualization artifact because the signal is lower on the log scale or is this an inaccurate normalization due to intensity differences between the normalization bead and the sample intensity?
Thanks for any input/feedback.
Joanne
Attachments
CYTOF-NORM1.pdf
Middle panel is raw data; left is Matlab norm, right is Fluidigm morm
(323.44 KiB) Downloaded 500 times
<<

mleipold

Guru

Posts: 5792

Joined: Fri Nov 01, 2013 5:30 pm

Location: Stanford HIMC, CA, USA

Post Tue Feb 28, 2017 8:24 pm

Re: Intensity of Normalization Beads

Hi Joanne,

Thanks for the data.....I haven't explored this closely.

A few questions or comments:
1. While I certainly agree that the cell signal you're showing is of *lowER* intensity than the main EQ isotopes, I still wouldn't call Dual ~100 (your CD27) a *low* intensity signal. If there were problems, I would expect it to only kick in closer to Dual ~10-30.

Have you tried gating on you EQ beads, then looking at the 142 and 176 channels, and maybe the 156 (Ce140+16) as well? Those are pretty low-intensity signals that are contained in the EQ beads themselved but, if I recall, *aren't* actually pegged to specific values during Fluidigm normalization (definitely not used in the MATLAB one). If the norm'ing were correct, then you would expect that the 142/176/156 would be a flat line like the other EQ isotopes.

2. Did you have noise reduction on during this acquisition? If so, could there be a baselining issue like previously discussed? (eg, viewtopic.php?f=4&t=582) Of course, that's mainly an issue if you had streaking in your sample.


Yes, in short, I do believe that the CyTOF community needs all the kind of bead standards that the fluorescent flow community does ("empty" background beads, multi-level or intensity beads perhaps even for Tuning/Calibration, etc). In addition to Fluidigm, I've been talking with some companies like Spherotech and Bangs about this; some are aware that it's a problem and they're talking with Fluidigm and other people about it. Others seemed surprised that CyTOF even existed, and didn't understand the potential market.

The more of us that yell for the standards, hopefully the sooner we'll get them......


Mike
<<

FlowjoVA

Participant

Posts: 13

Joined: Tue May 24, 2016 4:49 pm

Post Tue Feb 28, 2017 10:15 pm

Re: Intensity of Normalization Beads

Hi Mike:
Thanks as always for your feedback, I have attached the 142 and 176 channels of the Eq beads, both the raw and normalized. It does seem to do a better job with the beads than the cells.
Thanks,
Joanne
Attachments
142-176[1].pdf
Eq beads channel 142 176
(154.81 KiB) Downloaded 422 times
<<

GregBehbehani

Master

Posts: 85

Joined: Tue Apr 12, 2016 10:17 pm

Location: The Ohio State University, Columbus, Ohio

Post Tue Feb 28, 2017 10:36 pm

Re: Intensity of Normalization Beads

Hi Joanne,

I think your problem is with the normalization and not with the beads. I think the normalization is failing in regions where you have fewer bead events and at the time borders that the Fluidigm normalizer creates.

Before doing anything else, I would try Rachel Finck's Matlab normalizer. I suspect this will fix most of your problems.

Also, your run time also looks pretty long. Are you diluting your entire cell sample and running it all at once? It's possible that your antibodies (or antigen-antibody complexes) are falling off while the cells sit in diluted water. If so, you may try just diluting a few mL of your sample at a time and leave the remainder concentrated and on ice.

As to your original questions, the original normalization beads were made as a collaboration between the Nolan lab (particuarly Rachel Finck, Sean Bendall, and Erin Simonds) and DVS. They had several low intensity metals in them (La was the dimmest from what I recall) and they didn't work any better than the current commercial beads (in fact they were significantly worse in my opinion). When the new beads came out, Rachel (and possibly others in the lab) tested the two head-to-head and didn't see any downside to having the higher intensity bead signals. Importantly, the new beads are much easier to identify and remove from your data. I believe Rachel also looked to see if the intensity changes were significantly different across different parts of the mass range, but she found that the signal decrease was pretty uniform across the mass range. Now that we have 89Y and 209Bi it's probably worth revisiting this, but I have noticed any major issues.

I would try the Matlab normalizer and let us know if that fixes your problem. In the future, if you run a higher concentration of beads the Fluidigm software might work better for you, but I find it easiest to just use Rachel's software.

best,

Greg
<<

antonio

Contributor

Posts: 20

Joined: Wed Dec 04, 2013 3:05 pm

Post Wed Mar 01, 2017 7:26 am

Re: Intensity of Normalization Beads

Dear Joanne,
looking your second PDF, It seems to me that the TCRa-b signal is well normalized whereas the CD123 not. In my opinion, the CD123 is not well normalized since for a weak signal when the intensity reach zero it cannot be normalized anymore (any coefficient multiplied by zero will always result to be zero). As a consequence any normalization algorithm will work only until the signal is >0.
I have no idea about your first PDF, but we use the MATLAB normalizer for the moment.
Best
Antonio
<<

AdeebR

Grand master

Posts: 169

Joined: Thu Mar 13, 2014 5:58 pm

Location: NYC

Post Wed Mar 01, 2017 3:34 pm

Re: Intensity of Normalization Beads

Hi Joanne,

It's a little difficult to tell from your PDFs, but could confirm whether the rate of signal decrease for your cell markers is the same as the rate of decrease for the beads?

The metal signal in the beads is intrinsic to the beads and consequently very stable, so decreasing signal intensity of the beads almost exclusively reflects instrument drift. In contrast, the antibody/metal labels on the cell are less stable, so cell associated signal decreases due to both instrument drift as well gradual sample degradation with prolonged exposure to water. Bead will help to normalize the former but can't address the latter.

Best,

Adeeb
Adeeb Rahman
Icahn School of Medicine at Mount Sinai, NYC
<<

vtosevski

Contributor

Posts: 44

Joined: Wed Nov 20, 2013 12:50 pm

Location: Zurich, Switzerland

Post Thu Mar 02, 2017 11:04 am

Re: Intensity of Normalization Beads

Hi Joanne,

Looking at your plots, I see not the normalization problem but the signal decay/variation problem partly discussed in the previous post Mike also mentioned above. Have a look at my attachment here - looks similar, no?

In my experience, no normalization can account for this, you will see it exclusively on the Ab signals (not beads, not Ir and Pt, not Pd from barcoding) and, funny enough, only on positive populations, which will go down (negative populations have steady signals). The only way we found so far to fight this is similar to what Greg suggested. By no means should you let the entire sample sit diluted during the long acquisition. Keep the pellet in a small volume of water after the final wash and only take as much as you need for 10 min acquisition, while keeping the rest in the fridge. We switched to SuperSampler for any sample that takes more than 8 minutes to acquire (250uL), as it allows us to keep the sample tube on ice, which seem to help.

Best,
Vinko
Attachments
for Joanne.pptx
(44.25 KiB) Downloaded 358 times
<<

mleipold

Guru

Posts: 5792

Joined: Fri Nov 01, 2013 5:30 pm

Location: Stanford HIMC, CA, USA

Post Thu Mar 02, 2017 5:19 pm

Re: Intensity of Normalization Beads

Hi Joanne,

Do you still have the relevant IMD file(s)? If so, try reprocessing it with Noise Reduction turned "off".

If background is causing a baseline subtraction issue, then I think turning Noise Reduction off should remove/reduce the issue. Please let all of us know whether this makes a difference!


Mike
<<

FlowjoVA

Participant

Posts: 13

Joined: Tue May 24, 2016 4:49 pm

Post Mon Mar 06, 2017 10:22 pm

Re: Intensity of Normalization Beads

Thanks for everyone's feedback. Here are some answers to the questions/comments:
1) the normalization was done with BOTH Matlab and the Fluidigm normalization, with no difference in the outcome.
2) We did save the IMD files and reprocessed them with the Noise reduction turned off, no change in the normalization.
3) With regard to antibodies falling off, this sample was run over a period of 2 hrs using the supersampler with a cooling block. If antibodies are falling off under those conditions that is quite concerning. I have been working with fluorescence flow for more than 30 years and I have never seen such a phenomena with fluorescent abs, with a rare exception of a poor quality low affinity antibody. There is definitely a difference in the slope of the data between the beads and the cells indicating a difference in the rate of drop in intensity over time, with the cells having a steeper slope.
Just curious if anyone else has looked at their data this way, especially if you collect over a longer period of time.
Best,
Joanne
<<

AdeebR

Grand master

Posts: 169

Joined: Thu Mar 13, 2014 5:58 pm

Location: NYC

Post Tue Mar 07, 2017 12:57 am

Re: Intensity of Normalization Beads

Hi Joanne,

Yes, I've seen some evidence of reduced cell intensity (relative to beads) in long acquisitions, and it varies somewhat for different kinds of samples with certain cell lines seeming to be more problematic. It was more of an issue when we were relying on the standard DVS/Fluidigm Fix/perm to fix the cells, and has been somewhat improved by always adding a post-fix with fresh formaldehyde. I haven't tested it rigorously, but I think the osmium step that we've recently started including improves things even more since it provide even further fixation.

Even then, I think prolonged exposure to water can still be an issue (which is obviously not a concern with conventional flow). As noted by Greg and Vinko, when doing long SuperSampler acquisitions, we typically try keep our samples on ice in CSM and only prepare ~3mL aliquots (sufficient for ~1hr acquisition) at a time in diH20. This does mean we have to check and refresh the sample every hour, but I think this provide a reasonable compromise between convenience and reducing signal degradation over long runs.

Adeeb
Adeeb Rahman
Icahn School of Medicine at Mount Sinai, NYC
Next

Return to CyTOF data analysis

Who is online

Users browsing this forum: No registered users and 10 guests