A Genomic Strategy To Refine Prognosis and Predict Response to Therapy in Chronic Lymphocytic Leukemia.

Chronic Lymphocytic Leukemia (CLL) is notable for variation in aggressiveness of disease. Many patients with low-risk disease at diagnosis can be followed expectantly, while others rapidly require therapy. While a number of factors aid in determining prognosis, they have no role in predicting response to chemotherapy. We collected clinical data from a cohort of 220 patients with CLL followed at Duke University and the Durham VA Medical Centers. We assessed factors such as stage, leukocyte doubling time, IgVH mutational status, CD38 and ZAP-70 expression, cytogenetics, and time to treatment. We isolated RNA from purified CD5+, CD19+ leukemia cell samples from low and intermediate risk CLL patients, assessed the RNA expression with Affymetrix U133 Plus 2.0 GeneChips, and analyzed differential gene expression using previously developed methods of Bayesian binary regression. In a group of seventy-nine CLL patients who either eventually required (n = 45) or did not (n = 34) require therapy, CD38 status, ZAP70 expression, and cytogenetics were not statistically different (p > 0.5) while IgVH mutational status and leukocyte doubling time were different (p < 0.003). Of the patients treated, 69% were treated with chlorambucil, 51% with fludarabine, 31% with cyclophosphamide, and 49% with rituximab. Using genomic expression data, we developed a 100 gene expression signature that correlated with the need for eventual therapy. The accuracy of this signature was 94% using the leave-one-out cross validation method. Further validation using published data is currently being conducted, and data will be presented. Genes associated with the need for treatment include cell cycle regulators (such as E2F2 and CDK6), inhibitors of apoptosis (such as BAG1), and apolipoprotein B. In addition, using techniques described previously (Potti A et al, Nature Medicine, 2006, 12(11):1294), gene expression data coupled with either in vitro drug sensitivity in the NCI-60 panel of cell lines or in vivo drug response data were used to generate genomic models predictive of sensitivity to drugs commonly used in CLL including fludarabine, cyclophosphamide, and rituximab with accuracies on leave-one-out cross validation of 93%, 96%, and 79% respectively. These predictive models are currently being applied to gene expression data from leukemia samples of CLL patients to predict response to therapy (data to be presented). Thus, a genomic approach can be used to determine which CLL patients will require treatment. Importantly, models of chemosensitivity may predict response to therapy, and thus may ultimately help target the appropriate treatment for each patient.