The Hex^lowSM^high and Hex^highSM^low AML subtypes differ in gene expression and survival outcome. (A-B) Kaplan-Meier plots of (EFS; n_purple = 69, n_green = 72) (A) and (OS; n_purple = 70, n_green = 72) (B) for patients with AML grouped into Hex^lowSM^high (purple) and Hex^highSM^low (green) subtypes. The study cohort comprised patients from 3 institutions (University of Virginia [UVA], Memorial Sloan Kettering [MSK], and Pennsylvania State University [PSU]). Patients who received intensive induction chemotherapy treatment were included in the analyses. Within each plot are the corresponding risk tables for the 2 groups. (C) Volcano plot of differentially expressed genes between the Hex^lowSM^high and Hex^highSM^low subtypes. Purple genes are upregulated in the Hex^lowSM^high cluster whereas green genes are upregulated in the Hex^highSM^low cluster (−log₁₀P_adj >1). (D) The Hex^lowSM^high subtype is enriched for the LSC program. Gene set enrichment analysis score plot for an LSC signature of 104 genes.¹⁴ The y-axis is the running enrichment score along the ranked gene list. The enrichment score is the maximum deviation from 0 encountered along the list and represents the degree to which a gene set is overrepresented at the top or the bottom of the ranked gene list. The normalized enrichment score (NES) is the enrichment score normalized for variation in gene set sizes. The adjusted P-value (P_adj) for the NES is shown. (E-F) The deconvolved Hex^lowSM^high subtype is enriched for LSPCs and depleted in differentiated myeloid cell types. The CIBERSORTx log₂-transformed absolute abundance score of total LSPCs (sum of quiescent, primed, and cycling LSPCs) (E) and total myeloid-lineage cells (sum of promono-like, mono-like, and monocyte) (F) is shown for the Hex^lowSM^high (n = 217) and Hex^highSM^low (n = 199) subtypes colored by study. Full deconvolution of leukemic cell types including the different LSPC and myeloid cell types is shown in supplemental Figure 3I and J. Differences between the 2 sphingolipid subtypes were assessed by a 2-sided Wilcoxon rank-sum test. (G-H) Frequency of CD34⁺CD38⁻ cells between Hex^lowSM^high and Hex^highSM^low subtypes. Log₁₀-transformed proportions of CD34⁺CD38⁻ cells (G) and CD34⁺CD38⁻CD96⁺ cells (H) among 100 000 viable cells are shown for the Hex^lowSM^high (n = 11) and Hex^highSM^low (n = 12) subtypes. Differences between the 2 sphingolipid subtypes were assessed by a 1-sided Wilcoxon rank-sum test. (I) An RNASeq-based classifier accurately distinguishes sphingolipidomic subtypes. Receiver operating characteristics curve for a 284-gene support vector machine classifier applied to test data that includes both primary AML samples and cell lines with paired RNASeq and sphingolipidomic data. The area under the curve (AUC), its 95% confidence interval in brackets, and the 1-sided binomial test P-value (P_binom) of the classifier are shown. (J-L) Kaplan-Meier plots for patients with AML inferred to be Hex^lowSM^high (purple) or Hex^highSM^low (green) in BeatAML (n_purple = 102, n_green = 72) (J), TCGA-AML (n_purple = 67, n_green = 84) (K), or the molecularly–defined intermediate-risk group combined for both BeatAML and TCGA-AML (n_purple = 80, n_green = 58) (L). Only patients with AML who received standard intensive induction chemotherapy were included in the analyses for both datasets. Log-rank P-values, Hazard ratio and 95% confidence interval in brackets are shown. The bottom of each plot shows risk tables for the 2 subtypes.