Gene-expression profiles of NPMc⁺and NPMc^–AML. (A) Unsupervised hierarchical clustering of the training set. The dendrogram at the top was obtained using a list of 7197 selected genes (see Document S1, available on the Blood website; see the Supplemental Materials link at the top of the online article). The strongest parameter in determining AML clustering is NPM localization. Each of the 39 columns represents an AML sample, and each of the 7197 rows represents a gene (probe set). Genes were clustered according to Pearson correlation (the structure of the gene tree is not shown). (B) Hierarchical clustering of the test set using 369 probe sets obtained from an analysis of variance of the training set (see Document S1). The predictor genes efficiently discriminate AML cases according to NPM localization. Each of the 39 columns represents an AML sample, and each of the 369 rows represents a gene (probe set). Genes were clustered according to Pearson correlation, and the structure of the gene tree is shown. Color scheme used to identify sample characteristics is shown between panels A and B. NK indicates normal karyotype; AK, abnormal karyotype. (C-D) Affymetrix (C) and reverse transcription–quantitative polymerase chain reaction (RT-qPCR) (D) analysis of expression levels of 20 genes (with the highest scores among the 369 predictors) evaluated in 16 patients (8 NPMc⁺ and 8 NPMc^–, identified by numbers). (D) Relative expression levels are calculated as deviation from the median, and expression values for each gene in each sample are calculated as 2^-ΔCT (ΔCT = difference between the mean threshold cycle for each specific gene and for the 18S control ribosomal RNA gene). In the vast majority of cases, RT-qPCR reflects the results expected from microarray analysis. Homeobox gene expression levels appear to be particularly elevated in NPMc⁺ AML. The color bar (D, right) represents the color scheme applied to all parts of the figure.