Dear Editor,

The t(8;21)(q22;q22) translocation is a common chromosomal abnormality in acute myeloid leukemia (AML). Although t(8;21) AML has a favorable prognosis, ~40% of the patients will eventually relapse [1, 2]. Even with allogeneic hematopoietic stem cell transplantation (allo-HSCT), ~20% of patients still relapse, and further treatment is limited [3]. Notably, KIThigh mutation, particularly the KIT-D816 mutation, as well as FLT3-ITDhigh alterations, have been reported to confer poor prognosis in t(8;21) AML patients, although there remains controversy [4,5,6,7]. Other factors like aberrant immunophenotype, high white blood cell (WBC) counts, and high measurable residual disease (MRD) levels evaluated by multiparameter flow cytometry (MFC) also indicate worse outcomes [8, 9]. However, these do not fully illuminate the molecular or clinical variations observed in t(8;21) AML. Therefore, it is crucial to improve current risk stratification and identify those with a high risk of relapse as early as possible for tailored treatment strategies.

To elucidate the expression profile and molecular characteristics of t(8;21) AML, we incorporated transcriptome data and clinical information from 42 t(8;21) AML patients at diagnosis administered in our center between September 2020 and October 2022 (Supplemental Table 1). Using non-negative matrix factorization followed by consensus-clustering analysis (Supplementary Methods), three distinct transcriptional subgroups were identified based on 240 genes, named cluster 1 (C1) (38.1%, 16/42), cluster 2 (C2) (35.7%, 15/42), and cluster 3 (C3) (26.2%, 11/42), respectively (Fig. 1A, Supplemental Table 2). C1 was characterized by abnormal expression of leukemia stem and progenitor signatures (Fig. 1B). C2 and C3 were both enriched for interferon (IFN) signaling and immune response pathways but varied in the myeloid development stage. This was verified in gene set variation analysis (GSVA) for AML hierarchy signatures [10], where C1 was characterized by a blockage at the early stages of hematopoietic development, C2 displayed late-stage promono-like and monocyte-like cell properties, while C3 contained a spectrum of malignant cell types along the HSC-like to monocyte-like cells (Fig. 1C). We identified similar three clusters from 46 t(8;21) AML patients of the HOVON cohort [11], suggesting the accuracy of our clustering strategy (Supplemental Fig. 1). Additionally, we incorporated targeted sequencing data and identified that the adverse prognostic KIT-D816 mutation (p < 0.001) and FLT3-ITD (p = 0.031) alteration were significantly enriched in the C1 samples (Fig. 1D, supplemental Tables 3 and 4).

Fig. 1: Transcriptomics defined three distinct subgroups in t(8;21) AML.
figure 1

A Gene expression heatmap of three molecular categories in t(8;21) AML determined by consensus hierarchical clustering. Molecular subtypes and clustering confidence scores are displayed in the tracks. B Enrichment plots of each cluster. Columns showing the normalized enrichment scores (NES) and FDR q value of geneset enrichment by GSEA. C Heatmap for visualization of cell differentiation stages of each molecular subgroup by GSVA. D Genetic alterations in each cluster. Transcriptional clusters and gene mutations are shown for each sample. KIT others: KIT mutation except for KIT-D816 and KIT-N822.

Full size image

Notably, these subgroups exhibited obvious differences in prognosis. Patients in C1 showed a dismal prognosis, with 62.5% (10/16) patients eventually experiencing relapse or remaining refractory throughout the treatment course, while only 35.7% (5/14) patients relapsed or were refractory in C2, and all patients in C3 achieved complete remission (CR) and retained CR throughout the follow-up course (p = 0.006). Major molecular response (MMR) is defined as at least a 3-log reduction in RUNX1::RUNX1T1 transcript level, while complete molecular response (CMR) denotes the clear absence of detectable RUNX1::RUNX1T1. Here, our study revealed that only 1 (12.5%, 1/8) patient in C1 achieved MMR compared with that of 4 patients in C2 (33.3%, 4/12) and 4 patients in C3 (40%, 4/10) after the induction therapy. And CMR was only observed in patients of C2 (8.3%, 1/12) and C3 (50%, 5/10) during chemotherapy courses.

We further evaluated whether molecular subtyping had an independent impact on clinical outcomes. The median follow-up for our patients was 21.2 (95% CI = 19.3-24.2) months. Disease-free survival (DFS) and Event-free survival (EFS) were assessed according to the 2017 European LeukemiaNet (ELN-2017) criteria [12]. Considering the superior DFS of both C2 and C3 compared with C1 (p = 0.038 and p = 0.003, respectively), and the similar DFS between C2 and C3 (p = 0.054), we combined C2 and C3 patients into one group (hereafter called non-Cluster 1). DFS and EFS were censored at transplantation when specified as DFS-HSCT and EFS-HSCT. Univariate analysis indicated that elevated WBC counts, FLT3-ITD alterations, MFC-MRD positivity, and cluster C1 were associated with inferior DFS and EFS (Fig. 2A, B, supplemental Fig. 2A, B). KIT mutation was associated with poorer EFS. In the multivariable analysis adjusting for potentially cofounding variables, including WBC counts, FLT3-ITD alterations, MFC-MRD and transcriptional clusters, only MFC-MRD positivity (HR = 15.49, 95% CI = 3.86-62.07, p < 0.001 for DFS; HR = 24.21, 95% CI = 4.51-129.90, p < 0.001 for DFS-HSCT) and cluster C1 (HR = 10.48, 95% CI = 2.58-42.65, p = 0.001 for DFS; HR = 13.32, 95% CI = 2.53-70.04, p = 0.002 for DFS-HSCT) remained as independent adverse prognostic factors for DFS and DFS-HSCT after backward elimination (Fig. 2C). All these suggested the clinical significance of cluster subtyping in t(8;21) AML. The median DFS and DFS-HSCT for patients in C1 were both 8.26 (95% CI = 6.97-NA) month, which were not reached in non-C1 group (p < 0.001 and p < 0.001, respectively; Fig. 2D). The median EFS and EFS-HSCT for patients in C1 were 10.0 (95% CI = 6.97-NA) and 9.61 (95% CI = 6.97-NA) months, whereas they were also not reached in non-C1 group (p < 0.001 and p < 0.001, respectively; supplemental Fig. 2C).

Fig. 2: Transcriptional cluster was independently associated with DFS of t(8;21) AML.
figure 2

A Forest plot for univariate Cox regression analysis of factors impacting DFS-HSCT. DFS-HSCT: disease-free survival censored at transplantation. B Forest plot for univariate Cox regression analysis of factors impacting DFS. C Forest plots for multivariate Cox regression analysis of factors impacting DFS-HSCT and DFS. Variables that entered into the multivariate Cox model for DFS-HSCT and DFS initially included WBC counts, FLT3-ITD alterations, MFC-MRD, and transcriptional clusters. WBC: white blood cells. MFC-MRD: measurable residual disease levels evaluated by multiparameter flow cytometry. D DFS-HSCT of t(8;21) AML patients classified as Cluster 1 and non-Cluster 1 (upper) and DFS of t(8;21) AML patients classified as Cluster 1 and non-Cluster 1 (lower).

Full size image

In conclusion, our study proposed a novel classification strategy in t(8;21) AML based on multi-omics profiling that elucidated both the molecular and clinical variations observed in t(8;21) AML. We identified a high-risk subgroup (C1) that conferred poor prognosis and confirmed transcriptional cluster as an independent predictor of survival. Besides, previous studies have reported that KIT-D816 mutation predicted poor prognosis in t(8;21) AML. Two patients in C1 did not carry the KIT-D816 mutation, but they still experienced poor prognoses (28.6%, 2/7), with the DFS of 7.55 months and 13.87 months, respectively. Consequently, our transcriptomic profiling seemed to be more accurate in identifying high-risk groups compared with current risk classification based on genomic mutations.

Moreover, the distinct transcriptional signatures across these three clusters revealed the underlying differentiation hierarchy and cellular characteristics. The C1 group implied the most primitive stage and was thus associated with the poorest prognosis, which suggested the rational application of BCL-2 inhibitor. Interestingly, the IFN-related pathway was enriched in C2 and C3, which had relatively favorable outcomes, aligning with previous reports highlighting the advantage of IFN-α treatment in t(8;21) AML [13]. Therefore, our finding may help to identify which patients could benefit from IFN-α treatment based on our novel classification.

Taken together, these findings underscored the importance of transcriptional subtyping in t(8;21) AML for prognosis evaluation and therapeutic decision-making. Early risk identification at diagnosis and implementing targeted treatment strategies might improve patients’ outcomes. Further research is needed to validate the transcriptome-based risk stratification system in larger independent cohorts to confirm our pilot study and facilitate the design of corresponding clinical trials to guide precision-targeted chemotherapy and improve treatment outcomes.