In the previous Model Building tutorials, we learned how to build a model for a single-tube B-cell assay. Those tutorials are pre-requisites for this one.
Much of the time we use panels with several tubes - and therefore several listmode files. That presents a different set of problems for analysis. This tutorial will focus on the differences in how you approach model building for panels with more than one tube.
Concept and Strategy
The model that we build needs to do everything that the single-tube model does. It needs to help us understand the coordinated expression and transitions of the markers in the panel. However, GemStone is going to read the files in our panel one at a time. So, how can we build a model that correlates all markers if they are spread across several tubes?
GemStone helps us with this task by retaining the results for the markers it has analyzed. Here’s a simple example. Let’s say that tube 1 has CD10 and some other markers, and tube 2 has CD34 among its markers. GemStone will store the means and standard deviations for every state of the CD10 in tube 1 and allow us to correlate it with CD34 in tube 2. After GemStone has analyzed all of the tubes in the panel, the results for the entire panel are available to us. We can display Parameter Overlay plots and view statistics for any of the markers in the panel. We can even look at dot plots for markers that are actually in separate tubes.
Our model-building strategy is similar to the one we used to build our single-tube model. The difference is that we are more dependent on a good set of common parameters. The term “common parameters” refers to the set of markers that are common to all tubes in the panel. Using these parameters, we will build the skeleton of a model that will span the several tubes in our panel. The common parameters provide the backbone that allows us to make correlations among all of the markers.
Model Building Tip: Common Parameters
Common parameters are the parameters that are included in all tubes of a panel. Good common parameters make it possible to create a framework for comparing markers that are not in the same tube.
Choosing Good Common Parameters
A good set of common parameters is the key to success. For most cytometers, the light scatter parameters are available in every file. Unfortunately, they do not help us correlate markers in most cases. Ideally, the markers we choose to be common will have simple, well-defined transitions in the progression.
For our tutorial, we will set up a model for a 3-tube bone marrow panel, each tube having 7 colors, for a total of 11 fluorescent markers. The tubes have 5 common parameters: CD19 is required in all tubes so that we can identify the B-cells; CD34 will help us identify early B-cells and is a simple, step-down parameter; CD20 is up-regulated later in B-cell maturation; CD10 starts high and is lost in two later stages; and CD45 will also help identify transitions in the B-cell progression.
Along with these common parameters, tube 1 has CD58 and CD38. Tube 2 contains CD33 and CD13. And finally, tube 3 includes CD9 and CD22, rounding out this panel.
Let’s get started.
Tube 1
Click the Select FCS Files button on the main toolbar. Navigate to GemStone’s Sample Files folder, and select 3 files in the list: BCell_BM_Tube_001.fcs, BCell_BM_Tube_002.fcs, and BCell_BM_Tube_003.fcs. Click Open. The files are added to the File Database and the first file is read into GemStone.
Model Building Tip: Renaming Parameters
The parameter names for the files in this tutorial may not be especially helpful the first time you read them into GemStone. This can be easily corrected by selecting the Parameter Database command in the Edit menu. In this dialog, you can rename parameters (and do lots of other things, too). For additional details, see Edit Parameter Database.
We will build the model by first setting up the selection parameters: CD19 and SSC. Click the Choose Parameter button in the Workspace toolbar, and select CD19 in the list.
Let’s use the context menu in the CD19 parameter plot to select a Constant profile. Right-click inside the CD19 plot and from the menu, select Choose Parameter Profile. In the popup menu, select Constant.
With the markers in this panel, we will not be using the three auto-adjustment buttons as we have in other tutorials. Those routines are helpful in many cases, but they are not smart enough to handle the data in these files. We will size and position the definition points manually, and then use the Estimate X-Position button to fine-tune the placement.
Position the profile on the band of high-intensity CD19 cells. Use the Line Spread slider in the Parameter Profile panel to increase the line spread to about 7.0, and then click the Classify Data button on the main toolbar to analyze the events.
Next, we’ll repeat this process for SSC and identify cells with low SSC values.
Click the Choose Parameter button in the Workspace toolbar, and select SSC in the list. Right-click inside the SSC plot and from the menu, select Choose Parameter Profile. In the popup menu, select Constant. Position the profile on the band of low-intensity SSC cells. Click the Classify Data button to analyze the events.
Now before we go on, let’s see if we’ve really captured the cells of interest by using a conventional 2P dot plot display of SSC vs CD19. Click the Create 2P button in the Workspace toolbar.
Find an empty space on the Cell Type Canvas, click and drag to define a 2P plot. Select SSC on the X-axis and CD19 on the Y-axis.
If we look carefully at the plot, we can see that we have cut off some of the B-cells with low CD19 and some with higher SSC. At this stage in the model-building process, it is very important not to eliminate cells that may be of interest. Since we’re about to enrich for the cells we’ve selected, it is better to be a little generous on the edges at this step. So, we need to increase the line-spread settings for both CD19 and SSC before going on.
Click the CD19 parameter profile plot to activate it, and then increase the line-spread setting to approximately 8.40. Increase the line-spread for SSC slightly. Click the Classify Data button (or use the short-cut, Ctrl-A on the PC or Command-A on the Mac). You should now see more dark gray events in the 2P plot.
Let’s enrich for B-cells. Click the Enrich button on the main toolbar.
Now we’re ready to really define the progression. We will analyze the common parameters from the simplest to the hardest: CD34, CD20, CD10, and CD45.
Click the Choose Parameter button in the Workspace toolbar, and select CD34 in the list. The CD34-negative population is very broad and dim, and the positives at approximately 150 on the Y-axis. We will use a Step-Down profile for it. Right-click inside the CD34 plot and from the menu, select Choose Parameter Profile. Choose the Step Down profile in the list.
Position the upper step at ~ 150 and the lower step at ~0 on the Y-axis. Next, change the Variance Source in the Parameter Profile panel to Manual Point Entry. This allows us to adjust the line-spread for the upper and lower steps separately.
Click the first point in the lower step and adjust the Line-spread Value to 10 or 11. Then, click Estimate X-Positions to activate the profile and allow GemStone to determine the best location horizontally.
Model Building Tip: Tweaking the Profile
When you are manually adjusting definition points and not using the 3 adjuster buttons, you can tweak the positions of the points and re-analyze to find optimal positions and line-spread settings. Try moving points up, down, and sideways just a little and reanalyzing. Then, use the Estimate X-Positions button to fine-tune it.
Let’s move on to CD20. Click the Choose Parameter button in the Workspace toolbar, and select CD20 in the list. We want to consider everything less than ~150 as negative, and the events centered at ~800 as positive. Right-click inside the CD20 plot and from the menu, select Choose Parameter Profile. Choose the Step Up profile in the list.
Position and size the definition points approximately as shown in the picture below.
Trigger the Classify Data command from the menu, from the toolbar, or using the shortcut keys. Then, click the Estimate X-Positions button to allow GemStone to tweak the horizontal placement.
Click the Choose Parameter button in the Workspace toolbar, and select CD10 in the list. With CD34 and CD20 defined, the expression characteristics of CD10 become more obvious. We can easily see 3 levels of expression that we will model.
Right-click inside the CD10 plot and from the menu, select Choose Parameter Profile. Choose the Three Levels profile in the list. Position the levels to match the transitions that appear in the distribution.
Model Building Tip: 3-Levels or 4-Levels?
We know that when B-cells become mature, they lose CD10. In this example, there are actually a few cells that have become fully mature and show virtually no CD10 expression. We could model this with a Four Levels profile, and set the last level very low at the end of the progression to capture those mature B-cells. For this tutorial, we’ll keep it simple with Three Levels.
Click the Estimate X-Positions button to allow GemStone to tweak the CD10 definition points.
Now let’s add CD45 to the model. This marker starts low on early B-cells and elevates twice in the progression. Click the Choose Parameter button in the Workspace toolbar, and select CD45 in the list. Right-click inside the CD45 plot and from the menu, select Choose Parameter Profile. Choose the Three Levels profile in the list. For the initial placement of the definition points, use the graphic below as a guide.
Model Building Tip: Positioning Definition Points
The horizontal spacing of Definition Points is based on the number of events that the points relate to. If there are many events that relate to a pair of points, there is more space between them on the X axis. If there are few events between the pair of points, the points are close together on the X axis.
Now use the Estimate X-Positions tool to fine-tune the positions and classify the data.
The adjusted profile will look similar to the image shown below.
At this point, our model is defined. We have modeled 6 parameters common to all tubes. Two parameters (CD19 and SSC) were used to select the B-cells. The other four parameters (CD34, CD20, CD10, and CD45) were used to define a scaffold for the progression.
We can now use Auto Analysis to optimize the fit. Click the Auto Analysis button.
We can now add the remaining parameters in tube 1. However, we will not need to do any modeling for these markers because the model is already well-defined. This is one of the benefits to using GemStone for the analysis: with a well-defined model, you can explore how other markers - even those with unknown expression characteristics.
Add CD58 and CD38 to the model.
Let’s add a Parameter Overlay plot to the model, and then save the document.
Click the Parameter Overlay tool on the Workspace toolbar, hold the Shift key down and click the mouse on the canvas. The plot will be displayed.
Because the CD34 negatives are so dim, much of the CD34 ribbon in the plot is off scale. To adjust for this, we need to change the Y-axis scaling for the plot.
Double-click the axis tics for the Y-axis to display the Edit Properties for Y-Axis Area dialog. Change the Axis Scaling Mode to Fixed Scale, and set the Axis Min Scale to -10. Click OK.
On the main toolbar, click the Save Document button, and save the model document as “BCell-Multi-Tube.gs”.
Analyzing Tube 2
We’re ready to open the next file in the panel. Click the Next File button on the batch toolbar to read it into the model.
The first things you will notice when the second file is loaded are that some of the plots now show no dots and that the Parameter Overlay plot did not appear to change. Here’s why. The parameters that are common between tube 1 and 2 have been reanalyzed. The parameters that were in tube 1 and not in tube 2 still show the means and confidence limits for the analysis - but no dots. Those parameters can still be plotted on the Parameter Overlay plot based on the analysis from tube 1.
Let’s add the parameters from tube 2 all at once. Click the Choose Parameter button in the Workspace toolbar, and select Choose… in the list. The Edit Parameters dialog is displayed. Add CD13 and CD33 by double-clicking each one in the list on the left. Click OK to close the dialog.
We do not need to add Parameter Profiles for these markers because the model is already well defined. We can see the expected transitions for the markers, which come from the parts of the model that we did define. If the markers had characteristics that were not shown at this point, we probably would try to model them here.
Model Building Tip: Watch the RCS when opening a new data file.
The RCS (Reduced Chi-Square) is a measurement of how well the model matches the data. With a multi-tube panel, it can also be used as a quality control tool. If the RCS changes much when a new tube is read into the model, there’s a good chance that one of the common parameters has changed from the previous tube. If this occurs, you should stop and determine what has changed before moving on. GemStone has tools that can normalize parameter intensities from one file to the next.
Let’s move on to tube 3.
Adding Tube 3
Click the Next File button on the batch toolbar to read it into the model. Using any technique you prefer, add CD22, CD9, and FSC-A to the model.
Here again, if we needed to define any of these additional parameters, we could. It is not necessary in this case.
Our parameter overlay now shows us the correlated expression for all of the markers in the panel.
We would typically want to create some analysis zones for the model. That is left to the reader as an additional exercise, since it has been covered in earlier tutorials.
Let’s save our model, now that it represents the entire panel of tubes. Click the Save Document button on the main toolbar. Type a name for the model and click Save.
Summary
One of the most important aspects about multi-tube panels is the selection of common parameters that will be included in all tubes. These parameters are used by GemStone to create a backbone for the analysis, allowing correlations of markers in separate tubes to be explored as if they were in the same tube.
When creating a model for a panel of files, the common parameters are heavily relied upon. Markers that are found in only one tube are modeled when they provide key information to the progression. Once the model is fully defined with common parameters, GemStone can display and correlate parameters from different tubes.