AML: clustering genes to predict outcome

Acute myeloid leukemia (AML) is a heterogeneous group of hematopoietic malignancies with diverse genetic abnormalities and phenotypes. Currently, treatment decisions are based on the French-American-British (FAB) classification scheme, which uses largely morphologic characteristics, as well as immunophenotyping and cytogenetic analyses to identify different subtypes of the disease associated with better or worse prognosis. In this issue, Yagi and colleagues (page 1849) have used microarray-based assays to identify gene expression patterns that correlate with prognosis in a collection of pediatric AML patients. The authors assayed the expression of more than 12 000 genes in bone marrow and blood samples and used various data analysis methods to identify groups, or clusters, of patients with distinct phenotypes. Although the study was performed with only 54 patients divided amongst several FAB subgroups, the results have several important implications for the development of new prognostic tests and for the analysis of microarray data in patient samples. First, the simplistic approach of hierarchical clustering was unable to distinguish groups of genes that could predict outcome. However, by using more powerful statistical approaches, the researchers were able to identify a set of 35 genes that were highly predictive for good or bad prognosis. The list includes regulators of cell cycle and apoptosis that could be targets for novel therapeutic agents.

Another interesting finding reported by Yagi and colleagues concerns the relationship between the standard FAB classifications and the gene expression data. Although the FAB subtypes are relatively good predictors of prognosis, when gene sets that correlated with the FAB subtypes were identified, the resulting gene lists were poor predictors of outcome, suggesting that the FAB subtypes and the gene expression profiles measure fundamentally distinct features of the leukemic cells that are difficult to compare. The findings raise interesting questions about the relationship of genetic and morphologic indicators and suggest that microarray-based approaches will open new avenues in the treatment of AML.