AFFILIATIONS
1 Hematologic Malignancies Translational Program, Tampa General Hospital, Tampa, Florida, United States, University of Minnesota Medical School, Minneapolis, Minnesota, United States
2 Hematologic Malignancies Translational Program, Tampa General Hospital, Tampa, Florida, United States, USF Health Morsani College of Medicine, Tampa, Florida, United States
3 Florida State University, Tallahassee, Florida, United States, Hematologic Malignancies Translational Program, Tampa General Hospital, Tampa, Florida, United States
4 Hematologic Malignancies Translational Program, Tampa General Hospital, Tampa, Florida, United States, University of Florida, Gainesville, Florida, United States
5 Des Moines University College of Medicine, Des Moines, Iowa, United States, Hematologic Malignancies Translational Program, Tampa General Hospital, Tampa, Florida, United States
6 Hematologic Malignancies Translational Program, Tampa General Hospital, Tampa, Florida, United States, University of Florida, Gainesville, Florida, United States
7 Hematologic Malignancies Translational Program, Tampa General Hospital, Tampa, Florida, United States, University of Florida College of Medicine, Gainesville, Florida, United States
8 Taneja College of Pharmacy, Tampa, Florida, United States
9 Instituto Nacional de Cancerología, Programa de Leucemia, Ciudad de México, Mexico
10 Instituto Nacional de Cancerología, Programa de Leucemia, Ciudad de México, Mexico
11 Taneja College of Pharmacy, Tampa, Florida, United States
12 Hematologic Malignancies Translational Program, Tampa General Hospital, Tampa, Florida, United States, Leukemia Program, Tampa General Hospital, Tampa, Florida, United States
13 Hematologic Malignancies Translational Program, Tampa General Hospital, Tampa, Florida, United States, Tampa General Hospital Cancer Institute, Tampa, Florida, United States
14 Leukemia Program, Tampa General Hospital, Tampa, Florida, United States, Tampa General Hospital Cancer Institute, Tampa, Florida, United States, USF Health Morsani College of Medicine, Tampa, Florida, United States
15 Leukemia Program, Tampa General Hospital, Tampa, Florida, United States, Tampa General Hospital Cancer Institute, Tampa, Florida, United States
16 Hematologic Malignancies Translational Program, Tampa General Hospital, Tampa, Florida, United States, Leukemia Program, Tampa General Hospital, Tampa, Florida, United States, Tampa General Hospital Cancer Institute, Tampa, Florida, United States, USF Health Morsani College of Medicine, Tampa, Florida, United States
17 Genomic Testing Cooperative, Lake Forest, California, United States
18 Hematologic Malignancies Translational Program, Tampa General Hospital, Tampa, Florida, United States, Leukemia Program, Tampa General Hospital, Tampa, Florida, United States, Tampa General Hospital Cancer Institute, Tampa, Florida, United States, USF Health Morsani College of Medicine, Tampa, Florida, United States
ABSTRACT
Background: Our group previously showed that flow cytometry-defined AML differentiation arrest complements molecular risk stratification for predicting response to hypomethylating agent plus venetoclax (Andreadakis et al., ASH 2025). Although AML is both a disorder of aberrant hematopoietic differentiation and an immunogenic malignancy, how maturation states influence the immune microenvironment remains poorly defined. Clarifying which developmental program most strongly reshapes the leukemia immune microenvironment could identify biomarkers for selecting AML patients for immune checkpoint- or NK cell-directed therapies, an application not addressed by current genomic risk classifiers. We hypothesized that leukemia developmental programs shape immune transcriptional states. To test this, we designed a transcriptomic framework mirroring our six-state cell-of-origin (COO) model and validated developmental–immune relationships across independent AML cohorts. Methods: Baseline bulk RNA sequencing from 124 AML patients was used to develop a transcriptomic framework paralleling the flow cytometry-defined COO model (HSC, MPP, CMP, GMP, GP, MP). A 15-gene Random Forest (RF)-derived Primitive Score captured hematopoietic stemness and early myeloid biology, while complementary monocytic features defined the differentiation continuum. The resulting RF differentiation axis was oriented such that higher scores indicate more mature/monocytic states. The framework was externally validated in the 707-patient BeatAML cohort. Public AML single-cell RNA-sequencing data from the CZ CELLxGENE Census characterized donor-level immune programs. Analyses were performed in Python. Results: In the 124-patient training cohort, mean age was 63.8 years, 55% were male, and ELN22 risk was favorable (14%), intermediate (19.3%), and adverse (66.5%). COO distribution was HSC 2.9%, MPP 9.4%, CMP 29%, GMP 21.7%, GP 10.5%, and MP 26.4%. In 707 BeatAML samples, the mean-aligned transcriptomic COO score reproduced the prespecified differentiation hierarchy, consistent with independent primitive and monocytic gene signatures. Among 414 BeatAML patients with matched transcriptomic and mutational data, each 1-standard-deviation increase in COO score was associated with higher odds of NPM1 (OR, 2.29; 95% CI, 1.74–3.00; FDR<0.001) and FLT3 mutations (OR, 1.72; 95% CI, 1.24–2.38; FDR=0.002), and lower odds of TP53 (OR, 0.57; 95% CI, 0.39–0.83; FDR=0.004) and RUNX1 mutations (OR, 0.64; 95% CI, 0.47–0.88; FDR=0.006). To investigate immune axes within our designed differentiation hierarchy in the BeatAML cohort (n=707), an RF-prioritized monocytic program (LYZ, CTSS, CSF1R, FCER1G, SAT1, and CLEC7A) was projected onto three immune axes representing checkpoint regulation (CD274, PDCD1, CTLA4, HAVCR2) (ρ=0.366), cytotoxic effector activity (GZMB, PRF1, NKG7, IFNG) (ρ=0.302), and interferon (STAT1, IFNG, IRF1, CXCL10) (ρ=0.265). All three immune-module scores (all, P<10?¹²) were correlated with the monocytic signature. These findings identify monocytic polarization, rather than AML differentiation, as the principal transcriptional program associated with immune remodeling. In CELLxGENE donors with matched immune profiling (n=9), high 6-gene monocytic signatures correlated with reduced cytotoxic immune activity (ρ = -0.75, FDR = 0.03); after adjusting for NK-cell relative abundance, monocytic signature was independently associated with higher NK-cell checkpoint expression (adjusted standardized β=0.39, FDR<0.05), suggesting that monocytic-high microenvironments are associated with transcriptionally exhausted NK-cell states despite reduced overall cytotoxic activity. Together, these dual-compartment findings indicate that leukemia appears transcriptionally inflamed while its microenvironment remains functionally exhausted. Conclusions: We established a biologically interpretable transcriptomic COO framework that reconstructs the AML differentiation hierarchy. Monocytic polarization, rather than differentiation, emerged as the principal immune-associated transcriptional program, linking recurrent genomic alterations with immune remodeling and reduced cytotoxic activity across independent cohorts. This scalable, RNA-seq-based framework nominates monocytic polarization as a candidate marker of immune-suppressed AML microenvironments and provides a testable framework for evaluating immune checkpoint- or NK-directed therapies, complementing existing genomic risk classifiers.