Abstract
AML is a clonal disorder of immature hematopoietic blasts and has a variable clinical outcome. Current classification of AML is based predominantly on the cytogenetic abnormalities and morphology of the malignant blasts and is not always helpful for optimization of treatment strategy. It is, for instance, very difficult to predict the prognosis of AML patients with a normal karyotype, who constitute ~50% of the AML population. DNA microarray analysis has the potential to provide a novel stratification scheme for AML patients, which is based on gene expression profile, and might help to predict the prognosis of, and optimize the treatment strategy for, each affected individual. However, leukemic blasts derived from bone marrow (BM) of AML-related disorders, are not homogeneous. The blasts may constitute from 20% to almost 100% of mononuclear cells (MNCs) in the marrow. Furthermore, given that many leukemic blasts possess the ability to differentiate to a certain extent, the marrow of AML patients contains not only the immature blasts (leukemic stem clone) but also differentiated blasts. A simple comparison of BM MNCs among heterogeneous AML patients is thus likely to reveal a large number of changes in gene expression that only reflect differences either in the percentage of blasts or in the differentiation ability of the blasts. To minimize such population-shift effects in microarray analyses, we established a large-scale cell depository “Blast Bank” for the storage of CD133 (AC133)-positive hematopoietic stem cell-like fractions from individuals with a wide range of hematopoietic disorders. In the present study, we have used Affymetrix HGU133 A&B microarrays to measure the expression profiles of ~33,000 genes in the Blast Bank specimens of 99 adults with AML-related disorders: 83 individuals with AML and 16 patients in the RAEB stage of MDS. In contrast to the previous microarray analyses of BM MNCs of AML, unsupervised hierarchical clustering of the subjects based on the expression profile did not separate the patients into FAB subtype-matched subgroups. Comparison of gene expression profile between the long-time and short-time survivors has identified a small number of outcome-related genes. Supervised class prediction, based on these genes, with k-nearest neighbor method or Cox proportional hazard model both succeeded to clearly separate individuals into subgroups with statistically distinct prognoses. Our analysis may pave a way toward the expression profile-based novel stratification scheme for AML.
Microarray analysis of gene expression profile of multidrug resistance in pancreatic cancer
