Fri Sep 16, 2016 5:49 pm
Reagents wishlist?
Hi all,
At the Fluidigm UGM and CYTO 2016 in Seattle in June, there were a few talks and posters about standardization, QC/QA, and such matters. There were also a *lot* of hallway discussions about it, especially as more and more sites get started, and as projects get larger. I kept hearing a lot of the same things from people from all over the world, and in all settings (academic, industry, etc).
As such, I thought I'd try to spark a discussion on here about what CyTOF users think needs to be developed to further standardize the field and therefore increase robustness and reliability (and ultimately, confidence in the data we're generating). I would encourage people to list anything: reagents, hardware, software.....basically, anything.
In particular, everyone on the board who have experience with fluorescence flow cytometry standardization over the years, please speak up: we'd like to learn from your experience (good and bad!)
To help kick this off:
1. I think bead-based tuning would be a better indicator of machine performance than the current solution-based tuning. I showed at the Fluidigm meeting (and in discussions with several people) that the Tuning Solution day-to-day and between-machine differences are usually larger than the differences in signals from the EQ beads, which are themselves usually larger than signal intensity variances on cells (based on running separate aliquots of the same lot of prestained cells on multiple machines). Since beads seem to be more reflective of how cells are measured, I think it makes more sense to tune the instrument on them.
2. As a corollary to that: I think having multilevel beads (similar in concept to flow standards like Rainbow beads that have different levels of signal) would be helpful. I know from the Abdelrahman et al papers that the EQ beads can be made at varying levels of metals (eg, DOI: 10.1021/ja9052009), so there's not necessarily a technical reason why we couldn't have them.
3. Along with the idea of multilevel beads, having a CyTOF version of the Bangs multilevel capture beads (eg, http://www.bangslabs.com/products/flow- ... antitation) would also be useful. I have used the "815A" Simply Cellular beads, and they do work in CyTOF. The issue, though is Event Length consistency: since the beads themselves don't contain a metal signal, there is no "no-ligand"/zero bead. This also means that as you go down in capture ligand levels, the metal signal from your bound antibody also goes down which then causes your Event Length to go down (since Event Length is driven by ion cloud size which is driven by total metal content). Therefore, using default parameters, you "miss" a bunch of your bead events. Even if you play with the Event Length and Lower Convolution Threshold settings, it's still not as good as it could be. I and other people have had similar issues with BD's kappa capture beads, losing the lower signal intensity events.
Many of the things I've listed above are driven by wanting to understand how low you can go on epitope density (number of copies present on your cell) and still have confidence in your signal/data. I know a few comparisons have been published (eg, http://dx.doi.org/10.1016/j.jim.2010.07.002) but it would be nice to have the ability to see how individual machines are performing. Similarly, it would allow better investigation of linearity of detection at the low end: I know that I've seen a few cases where the ratio of spillover (say, M+1/M) changes if my M is very large (2000) vs moderate (200). Granted, in most cases if your M signal is in the 100s range your spillovers will be below 10 (background, in most cases, even without taking into account experimental artifacts with antibody cocktails, etc), but it's still something I'd like to look at further.
4. The above discussion about linearity of signal intensity across the range also has implications for compensation of CyTOF data (M+1, M+16, impurities, etc). If a spillover like M+1/M ratio is changing depending on your signal intensity, then compensation will probably be more difficult.
In terms of software standardizations of data, above and beyond EQ bead-based normalization, there were a few posters and talks at CYTO that were getting more deeply into this:
1. Katja Kleinstuber's poster in control spike-ins to each sample (recently published as viewtopic.php?f=10&t=543 )
2. From MD Anderson, Jeffrey Hokanson's talk on the CLEAN algorithm using spike-in control samples in barcoded samples to further normalize data: talk 129, "CLEAN Corrects Variations in CyTOF Sample Preparation" (not yet published, as far as I can tell)
3. From GSK, Scott Davis' multimedia poster on their CyTOF Studio suite software: poster 141/B4 "CyTOF Studio: Automated Classification and Analysis of High-Dimensional Mass Cytometry Data" (unpublished, as far as I can tell)
I think I'll stop there for now.
Mike
At the Fluidigm UGM and CYTO 2016 in Seattle in June, there were a few talks and posters about standardization, QC/QA, and such matters. There were also a *lot* of hallway discussions about it, especially as more and more sites get started, and as projects get larger. I kept hearing a lot of the same things from people from all over the world, and in all settings (academic, industry, etc).
As such, I thought I'd try to spark a discussion on here about what CyTOF users think needs to be developed to further standardize the field and therefore increase robustness and reliability (and ultimately, confidence in the data we're generating). I would encourage people to list anything: reagents, hardware, software.....basically, anything.
In particular, everyone on the board who have experience with fluorescence flow cytometry standardization over the years, please speak up: we'd like to learn from your experience (good and bad!)
To help kick this off:
1. I think bead-based tuning would be a better indicator of machine performance than the current solution-based tuning. I showed at the Fluidigm meeting (and in discussions with several people) that the Tuning Solution day-to-day and between-machine differences are usually larger than the differences in signals from the EQ beads, which are themselves usually larger than signal intensity variances on cells (based on running separate aliquots of the same lot of prestained cells on multiple machines). Since beads seem to be more reflective of how cells are measured, I think it makes more sense to tune the instrument on them.
2. As a corollary to that: I think having multilevel beads (similar in concept to flow standards like Rainbow beads that have different levels of signal) would be helpful. I know from the Abdelrahman et al papers that the EQ beads can be made at varying levels of metals (eg, DOI: 10.1021/ja9052009), so there's not necessarily a technical reason why we couldn't have them.
3. Along with the idea of multilevel beads, having a CyTOF version of the Bangs multilevel capture beads (eg, http://www.bangslabs.com/products/flow- ... antitation) would also be useful. I have used the "815A" Simply Cellular beads, and they do work in CyTOF. The issue, though is Event Length consistency: since the beads themselves don't contain a metal signal, there is no "no-ligand"/zero bead. This also means that as you go down in capture ligand levels, the metal signal from your bound antibody also goes down which then causes your Event Length to go down (since Event Length is driven by ion cloud size which is driven by total metal content). Therefore, using default parameters, you "miss" a bunch of your bead events. Even if you play with the Event Length and Lower Convolution Threshold settings, it's still not as good as it could be. I and other people have had similar issues with BD's kappa capture beads, losing the lower signal intensity events.
Many of the things I've listed above are driven by wanting to understand how low you can go on epitope density (number of copies present on your cell) and still have confidence in your signal/data. I know a few comparisons have been published (eg, http://dx.doi.org/10.1016/j.jim.2010.07.002) but it would be nice to have the ability to see how individual machines are performing. Similarly, it would allow better investigation of linearity of detection at the low end: I know that I've seen a few cases where the ratio of spillover (say, M+1/M) changes if my M is very large (2000) vs moderate (200). Granted, in most cases if your M signal is in the 100s range your spillovers will be below 10 (background, in most cases, even without taking into account experimental artifacts with antibody cocktails, etc), but it's still something I'd like to look at further.
4. The above discussion about linearity of signal intensity across the range also has implications for compensation of CyTOF data (M+1, M+16, impurities, etc). If a spillover like M+1/M ratio is changing depending on your signal intensity, then compensation will probably be more difficult.
In terms of software standardizations of data, above and beyond EQ bead-based normalization, there were a few posters and talks at CYTO that were getting more deeply into this:
1. Katja Kleinstuber's poster in control spike-ins to each sample (recently published as viewtopic.php?f=10&t=543 )
2. From MD Anderson, Jeffrey Hokanson's talk on the CLEAN algorithm using spike-in control samples in barcoded samples to further normalize data: talk 129, "CLEAN Corrects Variations in CyTOF Sample Preparation" (not yet published, as far as I can tell)
3. From GSK, Scott Davis' multimedia poster on their CyTOF Studio suite software: poster 141/B4 "CyTOF Studio: Automated Classification and Analysis of High-Dimensional Mass Cytometry Data" (unpublished, as far as I can tell)
I think I'll stop there for now.
Mike
