Written by Mike Kissner.
Cell sorters have become more sophisticated to rival the multicolor capabilities of analytical cytometers, offering up to seven lasers and a score or more of detectors.
Likewise, cell sorting experiments have also become more complicated in terms of the number of markers and fluorophores utilized to identify target populations, and more complicated in terms of understanding an increasing canon of flow cytometry and cell sorting definitions.
These days, sorts of up to 12 or more colors are not uncommon.
While multicolor sorts are very feasible and can yield excellent results, success is always a product of very careful planning and optimization.
Attempting a 12-color sort of precious samples without trial runs will almost always result in failure and require a return to the drawing board to perform the optimization steps that were initially foregone.
Additionally, all setup steps, especially compensation, must be performed on-the-spot rather than during offline data analysis, as is commonly done in analytical experiments.
Satisfactory results require that all controls and settings must be perfected during the setup phase immediately before the sort.
The following 4 core technique will help you through the planning, optimization, and trial process to give you successful multicolor sorting experiments.
1. Choose your fluorophores more intelligently.
One of first steps to designing a multicolor experiment is to determine which fluorophores to use to detect target markers.
When going about this task, keep in mind that compensation in and of itself is not something to be feared.
Compensation is unavoidable when venturing into the world of multicolor experiments, and the methodology for choosing fluorophores is to do so in a way that maximizes signal detection above background.
First, obtain the optical configuration – the lasers, their power, and the filter sets – of the instrument that you will be using to sort.
Once you have determined which spectral bands can be detected by your instrument, utilize an online spectral analyzer tool to find optimal fluorophores for that instrument.
There are several options, and they are all useful. Here are some to try…
Use these tools in combination with a reagent search on the manufacturers’ websites to find conjugates that are optimal for the instrument.
When choosing reagents, the general rule is to match fluorophore brightness with antigen density.
Keep in mind that all fluorophores are not created equal.
Some are much brighter than others: specifically, they emit more photons and have higher probabilities of detection than others.
For example, Brilliant Violet 421, PE, and APC/Alexa 647 are brighter than PE-Cy5 and FITC, which in turn are brighter than Pacific Blue.
With the advent of polymer dyes like the Brilliant Violets, there are more choices for bright fluorophores than there were in the past.
Reserve bright fluorophores, like APC or Brilliant Violet 421, for low-expression markers like CD25 or CD56.
Dimmer markers, like FITC, can adequately be used to detect bright antigens like CD45 or CD3.
Additionally, choose fluorophores to minimize the amount of overlap between them as much as possible, while not neglecting the importance of the relationship of brightness and antigen density.
On multilaser instruments with multiple beam spots and detection paths for each laser, it can often be very helpful to choose fluorophores so that detection is spread across the lasers and detection paths rather than saturating a single laser with several fluorophores.
On these kinds of instruments, fluorophores are excited and detected at different points in time as the cells pass through the interrogation points, which minimizes overlap and permits detection of fluorophores with the similar emission peaks but different excitation maxima (e.g. PE-Cy7 and APC-Cy7).
Online spectral analyzer tools permit visualization of fluorescence spectra based on laser line and will also overlay typical filter sets to get a sense of how much overlap there may be between detectors.
Keep in mind that overlap depends on laser power, filter sets, and reagent quality (especially in the case of tandem dyes) so the actual overlap on your instrument may be quite different.
The topic of choosing dyes in relation to compensation is a very tricky and difficult one.
In essence, compensation itself is not a problem.
Instrument acquisition software using the proper controls does an excellent job of calculating and correcting for spectral spillover between detectors: the overlap alone is not a problem that cannot be handled.
The big problem, which can result in severe loss of signal-to-noise resolution between stained and unstained populations, is something called spillover spreading.
Spillover spreading is a visual consequence of the process of compensation.
Compensation does not cause it but rather, reveals it.
The detection of every fluorophore is associated with a degree of error that is inherent to the measurement.
In other words, if the same cell stained with a particular dye was passed through the instrument 1,000 times, the fluorescence from the dye would be detected slightly differently during each measurement, which gives rise to a distribution or population.
The amount of variation in this distribution depends on many factors, including the power of the laser, the spectral region being measured (red fluorophores are lower energy, which gives rise to more variation in how many photons are detected at the photocathode during the measurement), the filter set used to detect the dye being measured, and the dye itself.
As with all measurements, dyes detected in channels other than their primary channels (spillover) are subject to the same variation.
During the process of compensation, the population measured (the FITC+ population in the example below) in the spillover channel (the 585/40 channel in the example below; typically used to measure a dye like PE) is shifted down on the log scale so that its median matches that of the negative population.
This process causes the FITC population to appear wider or more “spread out” than it appeared to be before compensation.
This phenomenon is essentially a visualization artifact of logarithmic scaling.
The portion of the scale in which the data resided before compensation is more compact (contains more data “bins” to which measurement values can be assigned) than the portion of the scale that results after compensation, so the population appears to be wider compensated compared to uncompensated.
The critical point about this phenomenon is that the process of compensation has absolutely no impact on the actual variation of the population being compensation; all it does is reveal this error.
Spillover spreading can have significant impact on the resolution between positive and negative populations and this specifically is the reason why avoiding spillover, when possible and reasonable, is important.
This topic can become very complicated very quickly, but the main takeaway is to be mindful of channels that receive spillover from other channels.
The degree to which spillover spreading occurs depends on many things – laser power, filter set, and the fluorophore itself – but, generally, the more spillover from other fluorophores that occurs in a particular channel and the more compensation required for that detector, the more spreading there will be.
Interestingly, a channel used for PE detection can receive significant spillover from other channels and can be problematic in this regard.
So even though PE is a terrific fluorophore in terms of brightness, it may be wise to be wary of using this channel.
While this is not necessarily reason to avoid using PE, it is something to be aware of and can help design smart and well-crafted panels.
For example, utilizing markers that are not co-expressed with the marker that PE is used to detect will help mitigate the extent to which spillover spreading is a problem.
There are some excellent and thorough resources out there that can assist in both better understanding and dealing with spillover spreading. Two of these are listed below:
2. Titrate your antibodies correctly.
This step of setting up a panel should be performed regardless of whether the experiment is a sorting experiment or a solely analytical one.
In flow cytometry, it is always important to keep in mind that almost everything we do is geared towards maximizing signal-to-noise resolution to provide the best possible chance to distinguish stained and unstained cells.
The better that we can distinguish positives from negatives, the more robust our identification of target cells and the better the sorting results will be.
Titration is a critical first step in ensuring that signal-to-noise and staining index are as high as they can be.
Although understandably daunting and arduous, with panels that include large numbers of fluorophores, the initial time investment will be well worth it.
When a suboptimal amount of antibody is used, the ultimate result is the same regardless of whether the concentration was too high or too low – compromised population identification.
When the amount of antibody is too high, non-specific binding of the antibody to cells that do not express the target antigen will occur.
When this happens, the negative population may both expand and increase in fluorescence intensity as compared to a non-stained sample.
Both of these effects influence how well the positive population can be discerned from the negative population and are components of the staining index, which is a an excellent measurement of separation.
Staining Index (SI) = MedianPositive-MedianNegative2 x Standard DeviationNegative
When too little antibody is used, the fluorescence intensity of the positive will decrease, yielding compromised separation and determination of positivity.
Be aware that the manufacturer’s recommended antibody volume “per test” is often much higher than what is needed.
The correct volume may be orders of magnitude lower than the recommended volume.
There are many nuanced methods for performing titrations, but they all rely on serial dilutions of a starting amount of antibody, often the manufacturer’s recommendation or slightly more.
However you titrate, make sure to choose the amount that generated staining with the highest staining index.
Keep in mind that only the Staining Index takes any widening of the negative into account compared to the S/N (MedianPositive – MedianNegative), so SI can, but not necessarily, be a better choice for calculating the right titer.
3. Test and optimize the instrument being used.
After the markers, fluorophores, and titrations have been worked out, be sure to run a few pilot experiments before the actual sort.
There is always a chance that some unforeseen issue may crop up even after everything has been optimized.
Even more critically, make sure to test the stain on the instrument that you will be using to sort.
It is very tempting, given the typical price difference in core facilities between time on an analytical cytometer compared with time on a cell sorter, to use an analyzer to test the stain, however, there is significant danger in this approach.
The robustness of a stain and the cardinal ability to distinguish stained populations from unstained populations can be highly dependent on the fundamentals of a cytometer like laser power, filter set, laser geometry and even the detectors themselves.
Sometimes, two cytometers are configured to detect the same fluorophore with different laser paths and excitation sources.
For example, PE can be detected either under 488 nm excitation conditions or 561 nm excitation conditions, and the stain may look significantly different under each of these conditions.
Additionally, the stain may look very different under the typical, more sensitive cuvette-driven interrogation points in analytical cytometers than under the jet-in-air interrogation points in many cell sorters.
4. Use more and better controls.
This can not be emphasized enough, especially when sorting, as all of the setup must be performed immediately before the sort.
There are two primary kinds of controls to that are essential: compensation controls and controls that determine positivity in a channel.
Proper compensation can only be calculated properly by using high-quality controls.
In order for a control to be acceptable, a clear positive and negative population must be discernible, and there must be sufficient cells in each population to collect a file with at least a few thousand events in both the positive and negative gates.
Correctly calculating compensation is not dependent on signal intensity.
The stain will be compensated correctly as long as the control is brighter or as bright as the staining intensity of the actual stain.
Both stained cells and beads can generate high-quality compensation controls.
If you anticipate that one of your stains will be dim or contain a low frequency of positive (or negative cells), beads are a better option.
As far as beads are concerned, be sure to use antibody capture beads, which are available from a variety of sources (BD™ Comp Beads, eBioscienceOneCompeBeads, Thermo Fisher Flow Cytometry Compensation Beads, among others), and not hard-dyed beads.
Antibody capture beads contain both a population of beads that bind antibodies so they can be stained with your experimental fluorophore as well a population of negative beads that will not bind antibodies.
Be sure to choose the product appropriate for the antibodies you are using in your experiment – some are species and/or isotype-specific and may not bind every antibody in your panel.
What about the autofluorescence component of compensation?
What happens if the autofluorescence of cells being used for compensation is different than the autofluorescence of the cells being measured in the experimental conditions?
It turns out that autofluorescence does not affect the compensation as long as the autofluorescence of the positive and the negative in each control is the same.
If this is the case, autofluorescence will not factor into the math of compensation, so it has no impact on compensation.
This can be tricky when compensating for a marker like CD14.
CD14+ cells, primarily monocytes, display a different autofluorescence profile than most of the CD14- cells (e.g. lymphocytes), so autofluorescence may impact the compensation calculation. In this case, antibody capture beads are the better choice.
It is perfectly acceptable if some compensation controls are generated with beads and others with cells.
Be sure that each control contains a negative population so that autofluorescence of the positive population and the negative population is the same.
“Universal negatives” are not appropriate in this case, as autofluorescence will be different among the compensation control samples.
What about compensating for fluorescent proteins?
Because it is critical that the fluorophore used in a compensation control is the same fluorophore being used in the full stained sample, a sample of FITC-stained cells is not a good compensation control for a GFP signal.
Clontech offers beads that are spectrally matched to GFP or mCherry.
For other fluorescent proteins, it is optimal to utilize a cell line that expresses each singly.
This can be arduous, especially if a transfection must be performed, but it really is the best way to properly compensate.
Controls that help determine what is positive and what is not positive are trickier.
We don’t have an extensive arsenal of controls at our disposal to do this, but the ones we do have can be very powerful.
Unstained cells, while often used, are not the best choice.
They may be passable for smaller panels, but they don’t tell us anything about background staining or spillover spreading, so they can often result in false positives, which can be devastating for sorting.
Isotype controls, while perhaps helpful to determine whether there is a general non-specific binding problem with the sample, can display different non-specific binding patterns than the antibody being used in the full stain, so they may not be terribly relevant either.
When staining is bright with a particular marker in a particular channel, we really don’t need (or have) controls to determine what is positive and negative.
When the separation is good, it is clear which populations are stained and which are unstained.
Keep in mind that staining above background itself does not necessarily mean that a population is positive in that it expresses the target marker.
Non-specific binding of antibodies to dead cells, for example, may result in a signal that appears positive for a marker but in reality is not.
What is more critical are controls that help us to determine what is stained and what is not stained when fluorescence is dim and there isn’t a clear bimodal distribution in the channels being measured.
These controls are called fluorescence minus one, or FMO controls, and are generated by staining the experimental sample with every antibody conjugate in the panel except for one.
For example, the FITC FMO control will be stained with every other marker except FITC.
FMO controls help us to determine the effect that our old friend spillover spreading has on the stain, which is an effect that no other type of control can reveal.
Below is a famous figure from a 2004 paper from the NIH that very clearly shows why FMO controls are necessary.
This figure shows, with real data, that bounds of positive staining are different depending on whether an unstained sample is used or an FMO control is used.
The unstained sample overestimates the amount of positive staining because it fails to take into account the background expansion as a result of spillover spreading.
In other words, the brighter fluorescence of the FMO control, compared to the unstained sample, in the PE detector in the figure above is due to the combined effect of FITC, Cy5PE, and Cy7PE.
The latter two are not represented on these plots but, regardless, have an impact on the fluorescence measured in the PE detector.
While it is often not necessary to prepare FMO controls for every single channel, it is wise to do so during optimization steps and the first time that an experiment is executed on the sorter to get a sense of the extent that spillover spreading is a problem.
Channels that often do not require FMO controls are those in which there is significant separation between stained and unstained populations and those which received little spillover from other channels.
These properties of a stain can be predicted to a certain degree, but even though it can certainly be arduous, be sure to cover all bases by preparing FMO controls for all channels the first few times the experiment is run in-full.
Choosing fluorophores to maximize signal detection and utilizing compensation controls to minimize spillover spreading are important first steps in planning your execution for this kind of cytometry experiment. Titrating your antibodies and optimizing your equipment while maintaining proper controls that you’ve set up and performed in advance are critical first steps to performing accurate multicolor sorting experiments.
To learn more about getting your flow cytometry data published and to get access to all of our advanced materials including 20 training videos, presentations, workbooks, and private group membership, get on the Flow Cytometry Mastery Class wait list.