GM-pre are the closest normal counterparts to CD34⁻ AML LSCs by gene expression and disorganized transcriptional landscape. (A, left) Three-dimensional PCA separating normal HSPC/precursor/mature and CD34⁺ AML and CD34⁻ AML LSC populations using the ANOVA 500 gene set (adjusted p-value ≤1.79 × 10⁻²⁸; Table S4 B). The percentage of variance for each component is indicated. The relationship between NPM1-mutated (mut) CD34⁺ and CD34⁻ LSCs and NPM1 wild-type (wt) CD34⁻ LSCs (all in CD34⁻ AML) is indicated. (Right) The loadings plot annotated with a selection of most variant genes is shown. (B) Bootstrap hierarchical clustering of normal and LSC populations using the same ANOVA 500 gene set using 1,000 bootstrap (bs) permutations. (C) Heat map showing hierarchical clustering of ANOVA 500 genes and their expression by normal BM HSPC/precursor/mature, CD34⁺ AML, and CD34⁻ AML LSC populations. Expression values are normalized per gene (by row) and are shown as a minimum–maximum scale (color bar). The blue block indicates genes with a scattered expression profile across normal, multiple normal, and LSC populations. The open rectangles highlight clusters of genes, with their gene names, which show preferential expression in GM-pre, GM-mat, and CD34⁻ AML LSC populations and HSPC and CD34⁺ progenitor LSC populations. (D, right) Expression of the top 75 differentially expressed genes (rows = genes) in normal LMPPs and GM-pre populations in normal populations, CD34⁻ AML, and CD34⁺ AML progenitor-like LSCs (indicated by colored horizontal bars; each column represents a different normal or LSC sample). Expression values are normalized per gene (by row) and are shown as a minimum–maximum scale (color bar). (Left) Mean expression of all 75 genes in each normal and LSC population. Triple asterisks indicate highly significant differences in mean expression of 75 genes between CD34⁻ AML and CD34⁺ progenitor-like AML (P < 0.001, corrected for multiple testing) in a test of 10⁵ permutations of 75 randomly selected genes (of 16,284 expressed genes). Error bars represent SD. (E) GSEA of normal BM HSC/MPP, LMPP, GMP, and GM-pre signatures in preranked profiles (based on mean fold change), comparing CD34⁻ LSCs with normal GM-pre and CD34⁻ LSCs with normal HSCs/MPPs (all others). Normalized enrichment scores (NES) and FDR q-values (FDRq) are shown. All comparisons were statistically significant. (F) PCA using a curated list of 547 TFs adapted from Novershtern et al., 2011, showing the position of normal and CD34⁻ AML LSC populations, with the associated loading plot. (G, top) Venn diagram of the overlap in differentially up-regulated genes encoding TFs in normal HSCs/MPPs, GM lineage–committed populations (GMP, GM-pre, and GM-mat), and CD34⁻ AML LSCs. 28 TFs (of which 27 have HGNC annotation) are up-regulated in both HSCs/MPPs and CD34⁻ AML LSCs and are indicated. (Bottom) Combined analysis of the top Venn diagram with overlap in differentially up-regulated genes encoding TFs in normal HSCs/MPPs, GM lineage–committed populations, and CD34⁺ progenitor AML LSCs. Seven TFs are coexpressed in CD34⁺ progenitor AML LSCs and HSCs/MPPs. Of these, five (shown) are also expressed in CD34⁻ AML LSCs. CMP, common myeloid progenitor; Ery-Pre, erythroid precursor; MEP, megakaryocyte-erythroid progenitor; prog, progenitor.