Study design and characterization of the gene expression–based subgroups of myelodysplasia. (A) Schematic depicting the study design. Venn diagrams show sources of RNA. The left and right halves indicate outlines of the analyses for bone marrow CD34⁺ cells and BMMNCs, respectively. Unsupervised clustering was performed on gene expression data of bone marrow CD34⁺ cells from 100 patients (training set of CD34⁺ cell samples), revealing 2 distinct subgroups. A regression model was constructed from the training set, followed by validation in an independent cohort (lower left). A regression model to predict the subgroups using BMMNC samples was also constructed from gene expression data of 51 patients with both CD34⁺ cell and BMMNC samples (training set of BMMNC samples). Prognostic significance of the model was tested in 114 patients with only BMMNC samples. (B) A heatmap shows expression levels of 3141 genes with high variability in 100 CD34⁺ cell samples. Each row represents 1 gene, and each column represents 1 sample. Gene expression–based subgroups, WHO subtypes, genetic lesions, and patients’ prognosis are shown below the heatmap. AML-MDS, AML with myelodysplasia-related changes; CMML, chronic myelomonocytic leukemia; MDS-EB, MDS with excess blasts; MDS-MLD, MDS with multilineage dysplasia; MDS/MPN-RS-T, MDS/MPN with ring sideroblasts and thrombocytosis; MDS/MPN-U, MDS/MPN, unclassifiable; MDS-RS-SLD, MDS with ring sideroblasts with single lineage dysplasia; MDS-SLD, MDS with single-lineage dysplasia; MDS-RS-MLD, MDS with ring sideroblasts with multilineage dysplasia. (C) A heatmap of expression levels of 7 genes of known prognostic significance in 100 CD34⁺ cell samples. (D) Expression levels of genes related to specific hematopoietic lineages. The left panel is a heatmap of gene expression levels in 100 CD34⁺ cell samples. Rows represent genes sorted according to hematopoietic lineages in which they are specifically expressed. Columns represent samples along with their gene expression–based subgroups and WHO subtypes. The middle panel represents mean z scores for each hematopoietic lineage. CLP, common lymphoid progenitor; CMP, common myeloid progenitor; EB, erythroblast; GMP, granulocyte monocyte progenitor; HSC, hematopoietic stem cell; MPP, multipotent progenitor; MEP, megakaryocyte/erythrocyte progenitor; MK, megakaryocyte.