2026 ASH®
Annual Meeting & Exposition December 12-15, 2026 ▪ New Orleans, Louisiana

Paper Details

Abstracts, authors, affiliations, and disclosures.

Saturday, December 12, 2026
04:00 p.m. - 05:30 p.m.
MCCNO - Great Hall BC
Technology and Procedures, Omics Technologies, Pathology, Bioinformatics, Computational Biology, Profiling, Treatment Considerations, Combination Therapy, Biological Therapies, Non-Biological Therapies, Chemotherapy, Artificial Intelligence (AI), Machine Learning, Biological Processes, Genomics, Multi-Systemic Interactions, Diseases, Myeloid Malignancies, Acute Myeloid Malignancies, Research, Fundamental Science, AML, Translational Research, Study Population, Human, Adult, Pediatric
PAPER

380 Single-cell multi-omics defines treatment-resistant cell states and a clinically informative classification in acute myeloid leukemia

Yang Song, Linjie Wu2, Yongjian Li3, Weiqian Dai4, Zhen Song5, Hui Wei, MD, PhD6, Sha Hao7, Jianxiang Wang Sr, MD, PhD8, Tao Cheng9
1 State Key Laboratory of Experimental Hematology, National Clinical Research Center for Blood Diseases, Haihe Laboratory of Cell Ecosystem, Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Tian, China
2 State Key Laboratory of Experimental Hematology, National Clinical Research Center for Blood Diseases, Haihe Laboratory of Cell Ecosystem, Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Tian, China
3 State Key Laboratory of Experimental Hematology, National Clinical Research Center for Blood Diseases, Haihe Laboratory of Cell Ecosystem, Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Tian, China
4 State Key Laboratory of Experimental Hematology, National Clinical Research Center for Blood Diseases, Haihe Laboratory of Cell Ecosystem, Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Tian, China
5 State Key Laboratory of Experimental Hematology, National Clinical Research Center for Blood Diseases, Haihe Laboratory of Cell Ecosystem, Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Tian, China
6 State Key Laboratory of Experimental Hematology, National Clinical Research Center for Blood Diseases, Haihe Laboratory of Cell Ecosystem, Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Tian, China
7 State Key Laboratory of Experimental Hematology, National Clinical Research Center for Blood Diseases, Haihe Laboratory of Cell Ecosystem, Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Tian, China
8 State Key Laboratory of Experimental Hematology, National Clinical Research Center for Blood Diseases, Haihe Laboratory of Cell Ecosystem, Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Tian, China
9 State Key Laboratory of Experimental Hematology, National Clinical Research Center for Blood Diseases, Haihe Laboratory of Cell Ecosystem, Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Tian, China

Background: Genetic classification has improved risk assessment in acute myeloid leukemia (AML), but it does not fully explain why patients with the same defining lesion have markedly different treatment responses. We reasoned that resistance may be determined by the functional organization of malignant cell states in addition to genotype.

Methods: We integrated single-cell transcriptomic, genomic, and DNA methylation data from 306 longitudinal bone marrow samples obtained from 198 patients with AML at diagnosis, non-response, complete remission, or relapse. Single-cell RNA sequencing was performed in all samples; whole-genome sequencing and whole-genome bisulfite sequencing were available for 286 and 128 samples, respectively. After quality control, 3,283,568 cells, including 1,923,917 malignant cells, were analyzed. Response-associated programs were defined within malignant stem/progenitor compartments and used to establish a seven-subtype Cellular State–based Classification of AML (CSCO-AML). Prognostic and therapeutic associations were evaluated in independent AML cohorts and ex vivo drug-response datasets.

Results: Primitive HSC-like, MPP-like, and LMPP-like malignant populations were enriched in patients who failed induction therapy; however, primitive-cell abundance alone did not account for resistance. Across these compartments, four conserved functional states converged on two opposing transcriptional programs. The impaired-differentiation program (Imp-Diff), enriched at diagnosis in non-responders and maintained in persistent refractory disease, coupled differentiation arrest to fatty-acid uptake and metabolism, TGF-β signaling, and ferroptosis defense, with increased CD36, SLC27A4, GPX4, DHODH, and AIFM2. The activated-differentiation program (Act-Diff) increased in patients who converted from non-response to remission and was associated with favorable survival in TCGA-LAML. DNA methylation analysis further showed that resistant states were reinforced by hypermethylation of differentiation-associated loci.

CSCO-AML separated diagnostic AML into seven biologically distinct subtypes. C1-C3 were enriched for differentiated, fusion-defined leukemias and had superior survival, whereas C4-C7 showed higher Imp-Diff activity, inferior outcomes, and distinct drug-response profiles. This separation was reproduced in TCGA-LAML, Beat AML, and OHSU AML and further resolved outcome heterogeneity within ELN intermediate- and adverse-risk disease.

The clinical resolution of CSCO-AML was evident in two settings. First, KMT2A-rearranged AML segregated predominantly into C4 and C5 despite sharing the same defining lesion. C4 was associated with more frequent treatment failure, enrichment of ROS-high and immune-evasive progenitor states, and increased Imp-Diff activity. C4 also harbored 13,015 differentially methylated regions, 95.2% of which were hypermethylated, including promoters of genes required for myeloid differentiation. In contrast, C5 retained transcriptional and epigenetic evidence of myeloid priming. Thus, CSCO-AML separated KMT2A-rearranged AML into biologically distinct entities with divergent treatment responses.

Second, CSCO-AML identified C7 as the subtype with the strongest venetoclax resistance, an association reproduced in pharmacogenomic data and in an independent cohort of primary AML samples tested ex vivo. Unlike the recognized BCL2-family resistance pattern in C1, C7 resistance was not accompanied by dominant BCL2 or MCL1 compensation. Instead, C7 was enriched for ASAH1 mutations (11/25, 44.0%; odds ratio, 3.44; P=0.0073) and showed sustained SLC27A4 upregulation, lipid-metabolic reprogramming, and increased Imp-Diff activity from the earliest leukemic progenitor stages. These findings define a distinct metabolic route associated with venetoclax resistance.

Finally, in the MOLM13 cell line, upregulation of CD36 can significantly increase its resistance to ferroptosis, as well as upregulation of SLC27A4. Interestingly, the SLC27A4 upregulation also markedly increases its resistance to venetoclax.

Conclusions: AML treatment resistance is organized by malignant cell state as well as genotype. CSCO-AML translates these states into a clinically informative framework that refines prognosis, resolves heterogeneity within KMT2A-rearranged AML, and identifies a metabolically defined venetoclax-resistant subtype. These findings provide a rationale for state-directed therapeutic strategies in AML.

Yang Song: No relevant conflicts of interest to declare.