Notes on Auto Analysis of B-Cells Video

C. Bruce Bagwell, 30 Nov 2010

Video Title: Unattended analysis of bone marrow B-cells, identification of very rare events, and evaluation of the efficacy of reagent panels.

This video was constructed to show the general approach we intend to explore for analyzing bone marrow B-cells for minimum residual disease.  The first part of the video shows how GemStone can largely automate the process of modeling normal B-cells.  It’s important to realize that this modeling method adjusts to slight changes in marker intensities and therefore can be generalized to any patient’s bone marrow sample.  Once the system has identified CD19 dim to positive events with a lymphocyte intensity of side-scatter, it models the down-regulation of CD10.  These are the only assumptions made in the model.

At first, CD10 is assumed to be just a simple step-down type of parameter profile; however, as soon as the system models CD38 and CD34 as step-downs, it realizes that it can more exactly model CD10 as a three-level parameter profile.  It also refines its analysis of CD38 in the same way.  Finally, it models CD45 as a multi-level increasing parameter profile.  The system is designed to workout the progressions of CD38, CD34, and CD45 given only the starting information that CD19 is dim to positive, SSC is in the lymphocyte range, and CD10 down-regulates.

Once the system has finished its modeling of normal B-cell progression, it automatically stages the sample and presents all the measurement correlations into an easy-to-understand overlay and set of information rich bivariates.  In the overlay, the x-axis represents B-cell progression and the y-axis represents relative measurement intensity.  Using an animation technique and model vectors, the progression through high-dimensional space can be viewed through any set of bivariates.

The second part of the video shows how the system can find very rare populations given its knowledge of the normal B-cell probability distribution in high-dimensional space.  The heat map displays the degree of difference from the model on the y-axis and the position along the lineage progression on the x-axis.  One important observation to make about this plot is that these events are defined probabilistically and thus are free from the typical subjectivity that normally accompanies minimum residual disease (MRD) types of analyses.

The video shows one aberrant population that is due to some APC-Cy7 tandem aggregates contaminating the CD45 APC-Cy7 reagent tube.  The video shows that the heat map easily identifies a population that has a frequency of occurrence of around 2x10^-5.  With more events in the file, there is no reason this sensitivity could not be enhanced to a 1x10^-6 level. 

The video then goes on to show the normal contamination of the marrow with plasma cells.  The heat map shows two distinct clusters that represent plasma cells with low and intermediate CD45 levels respectively.  Finally the video uses a region around a kappa+ lambda+ population to show how simple aggregation affects the ultimate sensitivity in finding very small residual populations.  Elimination or reduction of these aggregates will have a profound affect on the identification of real aberrant populations.

The third part of the video shows how the model can be used to quantify how well a set of reagents, fluorochromes, and instrument setup are working together to stratify events along the progression axis.  The %Fidelity Probability is a single objective measurement that is a function of measurement line-spread, compensation, and reagent synergism that quantifies how well the panel is working together on a specific instrument.  At the end, the video shows a complete Fidelity Analysis that shows how specific combinations of reagents work together.  This type of analysis is valuable in determining the minimum set of reagents that can stratify B-cells along their lineage pathway.