MicroRNA expression profiling in DLBCL, training and validation sets. (A) Unsupervised hierarchical clustering analysis of 21 DLBCLs (training set) defined 3 unique subsets of tumors based on the differential expression of 98 miRNA genes (MiRNA Groups A-C). (B,C) Two rounds of one-way ANOVA testing identified subsets of 38 and 16 miRNAs whose expression could effectively discriminate DLBCLs into these 3 subsets. (D) Validation of the miRNA-defined molecular substructure in an extended cohort of DLBCLs. Expression of 17 mature miRNAs was defined by stem-loop real-time RT-PCR, and unsupervised hierarchical clustering of 42 DLBCLs was performed with the use of dChip. All tumors reanalyzed by RT-PCR clustered into their originally assigned groups. In each heat-map, a column represents a DLBCL sample and a row represents a miRNA. Tumor features listed at the top of the figure include the following: site, GC or non-GC origin (IHC), extent of T-cell infiltrate, and copy number of the miR-17-92 cluster for samples in which both array CGH and expression analyses were performed; G indicates gain; D, diploid, L, loss. Original MG cluster was determined by microarray analyses.