Simplifying Complex Microarray Data To Derive Gene Expression Profiles Which Identify Childhood Acute Lymphoblastic Leukaemia Patients at Risk of Relapse.

The optimal treatment of patients with childhood acute lymphoblastic leukaemia (ALL) depends on establishing accurate diagnosis. Our investigations seek to strategically develop the application of microarray gene expression profiling to identify ALL patients with clinically homogenous presentations but which may respond differently to established treatment regimens. We have determined the gene expression profiles of ALL bone marrow (BM) samples taken from patients at diagnosis. Data analysis has focussed on the use of a novel and innovative statistical technology, Gene-RaVE. This series of patent protected algorithms builds a multinomial regression model using Bayesian variable selection. Gene-RaVE leads to the generation of a parsimonious and simple diagnostic gene expression signature, but which provides increased predictive ability over current analysis approaches. We describe our analysis of both Affymetrix (HU133A) and cDNA (10.5K) microarray gene expression profiles generated from diagnostic BM from >100 ALL patients covering the broad ALL subtypes including T and B lineage as well as T cell lymphoma leukaemia. Comparison of gene expression data failed to identify clearly distinguishing profiles between patients identified as ‘standard risk’ from ‘medium risk’ according to BFM95 clinical criteria. Gene expression profiles from a cohort of ALL patients, identified as ‘standard risk’ at diagnosis, were compared on the basis of their overall clinical outcome: relapse within 2 yrs vs non-relapse. Using a range of analyses including t-test, Gene-RaVE, discriminant analysis approaches and principle component analysis, we have discovered that small subsets of genes (<10), all of which included Nedd4BP3 and Ribosomal Protein L38 (RPL38), can be used to distinguish the two outcome groups. Subsequent validation using real time PCR supports the increase in Nedd4BP3 expression in standard risk patients which do not respond well to established treatment regimens. The Gene-RaVE algorithm also provides a generic framework for survival analysis. This approach indicates that the expression of these Nedd4BP3, RPL38 and inositol 1, 4, 5-triphosphate receptor, type 2 can be used to build a survival ‘index’ which correlates with the time to a relapse event in standard risk childhood ALL patients. Our results are suggestive of a way forward in the development of an informative, yet efficient diagnostic tool for this childhood malignancy using microarray gene expression analysis technology.