Association of NK cell states and markers with TKI response. (A) Expression heatmap of the top 10 marker genes for each NK cell subpopulation. The expression of literature-derived NK cell marker genes (CD56 [NCAM1], FCGR3A [CD16], CD57 [B3GAT1], NKG2C [KLRC2], and ZBTB16]) are plotted for comparison. Annotations were aided by Module Score computations for the 7 NK cell gene signatures compiled from literature (supplemental Table 21A; supplemental Figure 8C; “Methods”). Module scores were calculated for each NK cell type; note that adaptive NK cell signatures 1 to 3 are gene sets that are upregulated, whereas signature 4 is a gene set that is downregulated in the adaptive NK population. (B) Left: UMAP showing the 6 NK cell subpopulations identified through NK cell subclustering. Right: UMAP colored by control or CML labels. (C) NK subpopulations are plotted as a proportion of total NK in control and patients with CML. (D) The proportion of adaptive-like NK cells across control and prognostic groups is shown. Each dot represents 1 patient. (E) Expression heatmap of the top 15 genes upregulated and downregulated in adaptive-like NK cells, relative to other NK subpopulations. (F) Cell-to-cell communication analysis: NATMI was performed to identify potential interactions between ligands on HSPCs and receptors on NK cells. Ligand-receptor pairs predicted to mediate differential HSPC-NK interactions between control and group A (left), or groups C and A (right). x-axis: log2-transformed fold change of edge expression weight inferred by NATMI. The interactions marked by red asterisks represent NK cell–inhibitory interactions. (G) Expression of (left) HLA-E in HSPCs, and (right) NKG2A (KLRC1) in NK cells using pseudobulked populations. Each dot represents 1 patient. (H) Schematic summarizing prognostic NK cell features across groups. Statistical tests: Figure 4D,G; 2-tailed t test. Supplemental Figures 8 and 9 and supplemental Table 21 are linked with the data shown in Figure 4.