From 20-50% of patients with acute myeloid leukemia (AML) are primarily resistant to induction chemotherapy. It has previously been shown that resistance to the first cycle of induction chemotherapy is an independent prognostic factor. We investigated whether resistance to chemotherapy be represented by gene-expression profiles, and which genes are associated with resistance.cDNA microarrays containing approximately 41,000 features were used to compare the gene-expression profile of AML blasts between 33 patients with good or poor response to induction chemotherapy. Data generated by cDNA-arrays were confirmed by quantitative reverse transcription polymerase chain reaction.Using significance analysis of microarrays, we identified a characteristic gene-expression profile which distinguished AML samples from patients with good or poor responses. In hierarchical clustering analysis poor responders clustered together with normal CD34+ cells. Moreover, 13/40 (32.5%) genes highly expressed in poor responders are also overexpressed in hematopoietic stem/progenitor cells. Prediction analysis using 10-fold cross-validation revealed an 80% overall accuracy. Using the treatment-response signature to predict the outcome in an independent test set of 104 AML patients, samples were separated into two subgroups with significantly inferior response rate (43.5% vs. 66.7%, p=0.04), significantly shorter event-free and overall survival (p=0.01 and p=0.03, respectively) in the poor-response compared to in the good-response signature group. In multivariate analysis, the treatment-response signature was an independent prognostic factor (hazard ratio, 2.1, 95% confidence interval 1.2 to 3.6, p=0.006).Resistance to chemotherapy in AML can be identified by gene-expression profiling before treatment and seems to be mediated by a transcriptional program active in hematopoietic stem/progenitor cells.
