Acute lymphoblastic leukemia (ALL) is the most common childhood malignancy and the second leading cause of deaths from childhood cancer in the United States: 1. Approximately 3000 cases a year, representing 75 percent of all pediatric leukemias, are diagnosed in children in the United States, and the incidence continues to increase steadily 2. Despite dramatic progress in treating this disease, the absolute number of patients who relapse (who are initially defined as 'standard' risk) is greater than the number of children diagnosed with AML, neuroblastoma, or Wilms tumor 2. These patients appear to have a different class of disease, which cannot be differentiated using available clinical/pathological approaches (such as white blood celI count) at the time of diagnosis. The goal of this proposal is to define a diagnostic system, derived from gene expression profiles, which would differentiate between tumors initially classified as 'standard risk' which are amenable to therapy (achieve long-term disease free survival (LTDFS)) and those which are not (relapse within 2 years). Salvage therapy for this poor outcome group is currently ineffectual (LTDFS in less than 10 percent of cases) but there is evidence that intensification of treatment at the time of diagnosis for these cases would be highly effective. Expression profiling and analysis, blinded to clinical data, has previously been shown to easily discriminate between different types of leukemia tumors (ALL vs. AML) 3. We hypothesize that similar techniques will enable us to identify tumor sub-types with increased propensity to rapidly relapse within the 'standard' risk group. Our team consists of the Chairman of the Children's Oncology Group (Dr. Reaman), CNMC (one of 11 sites in the country to receive a $14,000,000 NHLBI PGA grant for profiling clinical samples)(Dr. Stephan), and several of the most experienced groups in the field of expression array statistics and bioinformatics in the community (Drs. Butte, Golub, Trent). In the R21 portion of this proposal we will first develop, as a proof-of-principle, a cDNA array-based system that can, in a single step, provide prognostic fusion transcript information, which is currently obtained by cytogenetics. We have an in-house tissue bank of greater than 3000 leukemia bone marrow aspirates (extremely high tissue homogeneity) which have been flash frozen which will be used as the resource for the entire proposal. Secondly, we will expression profile 50 tumors of the 'standard' risk - no relapse and 50 closely matched tumors of 'standard' risk which relapse rapidly after therapy to define a set of predictor gene candidates using both supervised and unsupervised techniques. In the R33 phase, this set of predictor gene candidates will be validated by profiling an additional 100 ALL samples blinded to relapse status. These samples will simultaneously be genotyped for the known prognostic indicators using the cDNA array to validate our ability to detect translocation status (specificity and sensitivity determined by comparison with the COG database). Finally, both the oligos which are able to detect translocation status from tumor RNA, as well as the validated predictor gene set will be incorporated onto a custom Affymetrix array for use as a rapid, one-step, inexpensive molecular diagnostic tool. The future goal of this proposal is to prospectively diagnose relapse so that therapy can be intensified at the time of presentation.
| English, Sangeeta B; Shih, Shou-Ching; Ramoni, Marco F et al. (2009) Use of Bayesian networks to probabilistically model and improve the likelihood of validation of microarray findings by RT-PCR. J Biomed Inform 42:287-95 |
| Mintz, Michelle B; Sowers, Rebecca; Brown, Kevin M et al. (2005) An expression signature classifies chemotherapy-resistant pediatric osteosarcoma. Cancer Res 65:1748-54 |
| Mitchell, Stephanie A; Brown, Kevin M; Henry, Michael M et al. (2004) Inter-platform comparability of microarrays in acute lymphoblastic leukemia. BMC Genomics 5:71 |
