"Predicting the Outcome of B-cell Chronic Lymphocytic Leukemia Using FTIR Microspectroscopy"

Rajendra Damle, North Shore University Hospital

B-cell chronic lymphocytic leukemia (B-CLL) is an incurable disease that can follow two different clinical courses. B-CLL patients whose leukemic clone expresses an unmutated IgV_H gene or higher numbers of ZAP-70+ or CD38+ cells exhibit a progressive form of the disease with a median survival of about 8 years, whereas patients with clones expressing a mutated IgV_H gene or with lower numbers of ZAP-70 or CD38 expressing cells have a more benign outcome with an approximately 25-year median survival. However, these relationships between V-gene mutations and ZAP-70 or CD38 expression do not always correlate, and the determination of mutation status is costly and time-consuming. Currently, there is no other clinical test that reliably predicts the course of the disease in B-CLL patients that are diagnosed in early stages. In this work, we are assessing the feasibility of using infrared micro-spectroscopy to differentiate between the different forms of B-CLL. B-cells from 37 patients were isolated and cytospun on IR reflective slides. 100 IR spectra were collected for each patient using a Perkin Elmer Spectrum Spotlight FTIR microscope with an aperture size of 25 µm² in reflection mode. The spectra from each patient were averaged and cluster analysis was performed. Cluster analysis showed that, in the lipid region of the spectrum (2800-3100 cm^-1), patients segregated into two distinct sub-clusters that were strongly correlated with the IgV_H mutation status. For the protein (1470-1740 cm^-1) and nucleic acid (1000-1300 cm^-1) regions, two distinct, but non-identical clusters were also observed. However, unlike the lipid region, the clusters did not correlate with the mutation status of the IgV_H gene. To be able to predict the IgV_H mutation status, we trained a feed-forward neural network in the lipid region, using 18 samples in the training set and 8 samples in the internal test set. The remaining 11 samples were then used to test the trained neural network. For all samples, the IgV_H mutation status was known. Results showed that, from the 11 patients in the test set, 10 were predicted correctly. Interestingly, the patient that was predicted incorrectly has followed a clinical course opposite from that predicted by the IgV_H mutation state. These findings suggest that lipid content in B-CLL cells and possibly their membranes may be linked to the pathology of the disease, which is supported by studies suggesting that lipid rafts may be involved in B-CLL activation. Ongoing studies of lipid composition from isolated B-CLL cell membranes will be used to further test this hypothesis. In the future, additional samples will be added to the neural network in order to improve its performance. Also, data from the clinical course of each patient will be used, in addition to IgV_H mutation status, to train the neural network in order to improve the robustness of the neural network and accuracy of the prediction.