Abstract

Background

   Comprehensive analysis of single-cell transcriptome and chromatin
   accessibility will contribute to interpret the heterogeneity of acute
   myeloid leukemia (AML). We hypothesize that integrating single-cell
   transcriptomic and chromatin accessibility landscapes underlying
   t(8;21) AML will provide valuable insights into its heterogeneous
   cellular properties and gene regulatory programs.

Methods

   Here, we conducted paired single-cell RNA-sequencing (scRNA-seq) and
   single-cell ATAC sequencing on bone marrow samples from newly diagnosed
   t(8;21) AML patients and healthy controls. Genetic signatures extracted
   from scRNA-seq were built and validated across three independent
   cohorts (German AMLCG1999, [36]GSE106291 and TCGA LAML).

Results

   We identified TCF12, a core component of AML1-ETO-containing
   transcription factor complex (AETFC), as the most active transcription
   factor in blast cells, driving a universally repressed chromatin state.
   Furthermore, we delineated two functionally distinct T cell subsets,
   revealing that EOMES-mediated transcriptional regulation promotes the
   expansion of a cytotoxic T cell population (T cells_2; high GNLY, NKG7
   and GZMB expression), with an increased clonality and a tendency for
   drug resistance. In addition, we discovered a novel leukemic CMP-like
   cluster characterized by high TPSAB1, HPGD and FCER1A expression.
   Leveraging machine learning-based integration of multi-omic profiles,
   we identified a robust 9-gene prognostic signature, which demonstrated
   significant predictive value for AML outcomes across three independent
   cohorts.

Conclusions

   This multi-omics study provides unprecedented insights into the
   transcriptional and epigenetic heterogeneity of t(8;21) AML, providing
   a potential actionable tool for clinical risk stratification.

Supplementary Information

   The online version contains supplementary material available at
   10.1186/s12967-025-06659-0.

   Keywords: scRNA-seq, scATAC-seq, TCR repertoire, T(8;21) AML

Introduction

   Acute Myeloid Leukemia (AML) is a heterogeneous and complex hematologic
   malignancy characterized by a diverse range of cellular subsets, which
   presents distinct responses to treatment. One of the most common
   chromosomal abnormalities in AML is the t(8;21) translocation, which
   produces the RUNX1::RUNX1T1 fusion protein (also known as AML1-ETO).
   This fusion protein plays a key role in the pathogenesis of AML, and
   epigenetic dysregulation is a distinctive hallmark of the disease
   [[37]1].

   Previous studies have revealed that the depletion of RUNX1::RUNX1T1 in
   leukemic cells would lead to a genome-wide alteration in transcription
   factor binding, suggesting its critical role of epigenetic
   reprogramming in t(8;21) AML [[38]2]. Furthermore, RUNX1::RUNX1T1
   resides in leukemic cells and functions through a stable
   AML1-ETO-containing transcription factor complex (AETFC), which
   contains several hematopoietic transcription (co)factors. These AETFC
   components co-localize genome wide, providing multiple DNA-binding
   domains for target genes and stabilizing the AETFC complex, thereby
   contributing to the pathogenesis of t(8;21) AML [[39]3, [40]4]. Despite
   AML with t(8;21) translocation being considered with a favorable
   prognosis, approximately 30–50% of patients experience relapse and drug
   resistance [[41]5]. The heterogeneity and complexity observed at both
   the cellular and epigenetic levels of t(8;21) AML present major
   challenges in curing this disease. We hypothesize that investigating
   gene expression profiling and chromatin accessibility underlying
   t(8;21) AML will provide valuable insights into its heterogeneous
   cellular properties and gene regulatory programs.

   Chromatin remodeling events, including alterations of chromatin
   accessibility and histone modification, have been reported to regulate
   the gene expression profiling in epigenetic mechanism patterns and are
   thus closely linked to the disease status of AML. Recent advances in
   high-throughput technologies, particularly single-cell analyses, have
   provided great opportunities to interrogate the cellular heterogeneity
   and global chromatin landscapes in AML. Although several studies have
   revealed the critical role of aberrant chromatin accessibility in the
   pathogenesis of hematological malignancies [[42]6, [43]7], the
   epigenetic regulatory mechanisms underlying t(8;21) AML at single-cell
   resolution remain elusive.

   To explore the heterogeneous cellular properties and epigenetic
   cis-regulatory programs of t(8;21) AML, in this study, we performed a
   comprehensive analysis using single-cell RNA sequencing (scRNA-seq),
   single-cell assay for transposase-accessible chromatin with sequencing
   (scATAC-seq), and single-cell T cell receptor sequencing (scTCR-seq) on
   bone marrow samples from newly diagnosed patients with t(8;21) AML.
   This multi-omic approach allows for an in-depth examination of gene
   expression patterns, global chromatin landscapes, and the exact T cell
   receptor clonotype profiles of T cell subtypes within the tumor
   microenvironment (TME) of t(8;21) AML. Through systematically comparing
   the AML samples with those of healthy individuals at the single-cell
   level, we characterized the substantial inter- and intra-patient
   heterogeneity, as well as the chromatin accessibility landscape within
   t(8;21) AML, and revealed a genetic signature that could serve as a
   novel prognostic indicator across multiple independent AML cohorts.

Methods

Human specimens

   The bone marrow samples of patients with primary t(8;21) AML were
   collected in compliance with the Declaration of Helsinki. The detailed
   clinical information of the samples and treatment regimens is described
   in Table [44]S1. All patients provided written informed consent, and
   this study was approved by the Ethics Committee of Ruijin Hospital
   Affiliated to Shanghai Jiao Tong University School of Medicine
   (2023-355) and Sun Yat-sen University Cancer Center
   (G2023-084-01,G2023-180-01).

Single-cell RNA-seq and V(D)J library preparation and sequencing

   Single-cell RNA sequencing and V(D)J libraries for t(8;21) AML samples
   were prepared using a 10x Single Cell Immune Profiling Solution Kit
   v2.0 (10x Genomics, 1000263), according to the manufacturer’s
   instructions. The libraries were subsequently sequenced on a NovaSeq
   6000 platform (Illumina).

Single-cell RNA-seq library processing

   Raw sequencing reads of scRNA-seq were processed using Cell Ranger (10x
   Genomics, default settings, version 5.0.0) and aligned to the human
   GRCh38 reference genome. Quality control and downstream analysis were
   performed with the Seurat (v3.0.2) R toolkit [[45]8]. All functions
   were run with default parameters, unless specified otherwise. We
   excluded cells with fewer than 200 or more than 6, 000 genes or with
   more than 10% mitochondrial RNA genes. The top 2, 000 highly variable
   genes were identified. Harmony was employed to adjust sample batch
   effect. Uniform manifold approximation and projection (UMAP) [[46]9]
   was performed to visualize the clusters. Cell identities were annotated
   based on the SingleR software [[47]10] and canonical markers, including
   CD34 for progenitor cells, CD14 for monocytes, CD3 for T cells, and
   CD79A for B cells.

Single-cell V(D)J library processing

   Raw sequencing reads of scTCR-seq were processed using the Cell Ranger
   VDJ pipeline (10x Genomics, version 5.0.0) to assemble the TCR
   sequences. Cells annotated as T cell cluster in the scRNA-seq data and
   with productive TCR α and β chains were further analyzed. V-J gene
   combinations in T cells were visualized via Sankey by using the ggplot2
   package in R software.

Functional enrichment and cell-cell communication analysis

   Gene Ontology and gene set enrichment analysis (GSEA) of differentially
   expressed genes were performed by using clusterProfiler R package (v
   3.14.0) [[48]11]. Cell-cell communication in primary t(8;21) AML was
   inferred using the CellChat toolkit [[49]12].

Single-cell ATAC-seq library preparation

   The nuclei were isolated with a Shbio Cell Nuclear Isolation Kit
   (Shbio, #52009–10). Libraries were constructed by using a Chromium™
   Single Cell ATAC GEM, Library & Gel Bead Kit v2.0 (10x Genomics,
   1000390) following the manufacturer’s guidelines. Libraries were
   sequenced with paired-end (2 × 100) by NovaSeq 6000 (Illumina).

Single-cell ATAC-seq data processing

   Cell Ranger-ATAC (version 2.0.0) was used for scATAC-seq data
   processing. Cells were removed based on the following criteria: a
   transcriptional start site (TSS) enrichment score lower than 4 and
   fewer than 3,000 total fragments in peaks, as determined using ArchR
   [[50]13]. In addition, doublets were filtered out using ArchR software.
   Harmony was used to remove batch effects cross-samples. Identities of
   cells from scATAC-seq were annotated from the corresponding scRNA-seq
   data. A total of 25,491 nuclei (from 3 leukemia samples) and 38,462
   nuclei (from 3 leukemia samples and 2 controls) were retained for
   cross-sample integration analysis. UMAP was used for visualization.

Integrated analysis of scRNA-seq and paired scATAC-seq

   To determine the cluster-specific marker genes, we applied a log2 fold
   change threshold greater than 1.25 and a false discovery rate (FDR) of
   less than 0.01 using the Wilcoxon test. For identifying
   cluster-specific peaks, we conducted pseudo-bulk replicates and applied
   MACS2 software to display the number of peaks in different regions.
   Motif deviation enrichment and transcription factor (TF) footprinting
   were also analyzed. Genomic track profiles of specific regions were
   visualized with CoveragePlot.

Construction of the leuk_CMP-like signature from joint integration of
scATAC-seq and scRNA-seq data

   The genetic markers expressed in the Leuk_CMP-like cluster were
   calculated through the joint analysis of scATAC-seq and scRNA-seq data.
   Overexpressed genes were defined as those with an average log2 (fold
   change) greater than 3.0 and a p value below 0.05. A total of 136 genes
   were identified (Table [51]S2). The German AMLCG1999 dataset was used
   as the training dataset and LASSO regression analysis was used to
   construct the final model.

Data availability

   The raw sequencing data reported in this paper have been deposited in
   the Genome Sequence Archive in National Genomics Data Center
   ([52]https://ngdc.cncb.ac.cn/gsa-human), under the accession number
   #HRA007073. Due to the legal restrictions, these data are under
   controlled access. Requests for access to these data for research
   purposes can be directed to the corresponding author Dr. Lu Jiang
   (jl11891@rjh.com.cn).

Statistical analysis

   The statistical analysis of this study was conducted by R software
   (version 4.2.0).

Results

Integrative single-cell analysis discriminated leukemic cells in primary
t(8;21) AML

   In order to characterize the immune repertoire as well as the chromatin
   accessibility underlying the bone marrow (BM) heterogeneity of newly
   diagnosed t(8;21) AML patients, we conducted paired scRNA-seq,
   scTCR-seq and scATAC-seq analyses with 10x genomics technology. The
   flow chart of the study was demonstrated in Fig. [53]1A. The clinical
   features of the analyzed samples were summarized in Table [54]S1. In
   brief, we collected a total of 28,000 cells from the bone marrow of AML
   patients harboring RUNX1-RUNX1T1 fusion and wild-type KIT, with an
   average of 54,634 mean reads per cell. In addition, we reanalyzed
   scRNA-seq data as well as scATAC-seq data of normal BM samples from
   healthy individuals [[55]14, [56]15] to construct a reference for the
   hematopoietic lineage as a normal control.

Fig. 1.

   Fig. 1
   [57]Open in a new tab

   Discrimination of leukemia cells in primary t(8;21) AML samples from
   single-cell RNA-seq and scATAC-seq. (A). Flow chart of the study
   design. We conducted paired single-cell RNA-sequencing, single-cell TCR
   repertoire and single-cell ATAC-seq of bone marrow mononuclear cells
   (BMMCs) from t(8;21) AML patients (n = 3) and healthy controls (n = 2).
   (B). Uniform manifold approximation and projection (UMAP) plot
   demonstrated single-cell RNA-seq profile of 32,457 cells from t(8;21)
   AML (n = 3) and healthy controls (n = 2). Dots indicates individual
   cells. (C). Bar plot showed the percentage of cells from each sample in
   each cluster. (D). Heatmap displayed inferred copy number variations
   (CNVs) from t(8;21) AML samples in comparison to healthy controls. (E).
   UMAP demonstrated the integration scATAC-seq analysis of t(8;21) AML
   (n = 3) with healthy controls (n = 2). (F). Summary of peaks
   distribution in the t(8;21) AML samples and healthy controls. (G).
   Footprint of leukemia-specific transcription factor RUNX1, TCF12 and
   LYL1 in t(8;21)AML and healthy controls

   First, to discriminate the leukemic cell identities of the primary
   t(8;21) AML samples, we adjusted for batch effects and performed an
   integrated analysis of both AML and normal cells. A total of eighteen
   cell clusters were identified (Fig. [58]1B and C and Fig. [59]S1).
   After combining of automatic annotation and manual verification
   referring to our previous work [[60]16], we determined the cell
   identities of the cells clusters. Cluster_4 cells, characterized by
   high expression of CD3D and LTB, were annotated as T cells. Cluster_8
   cells, exhibiting high expression of GNLY and NKG7, belonged to NK
   cells. Cluster_5 cells highly expressing CD79B and DNTT were annotated
   as B cells. In addition to elevated expression of CD79B and DNTT,
   Cluster_11 cells also presented increased expression of CD34 and thus
   were annotated as the progenitors of B cells. Cluster_3 cells and
   Cluster_13 cells, which highly expressed S100A8 and FCN1, were
   annotated as monocytes and macrophages, respectively. The remaining
   clusters, including Cluster_0, Cluster_2, Cluster_6–7, Cluster_9–10,
   Cluster_12, Cluster_14, Cluster_15-Cluster_17 were composed of leukemic
   cells. We also identified a small cluster of cells (Cluster_1) that
   highly expressed the cell-cycle related genes TYMS, UBE2C and KIAA0101,
   indicating a proliferative state. Interestingly, most of the cycling
   cells in Cluster_1 derived from leukemia samples, presented a highly
   proliferative state, whereas very few were presented in the healthy
   controls. Further exploration of chromosomal copy number variations
   (CNVs) of t(8;21) AML samples in comparison with healthy controls
   (Fig. [61]1D) also revealed recurrent alterations in the majority of
   leukemic cells in t(8;21)AML samples.

   To explore the alterations in chromatin accessibility, we performed
   scATAC-seq on the same matched samples. Our analysis obtained
   cis-regulatory data for 28,977 cells (Table [62]S1 and Fig. [63]1E),
   with an average of 20,187 unique fragments mapping to the nuclear
   genome. Approximately 60.2% of the Tn5 insertions were located within
   aggregate ATAC-seq peaks (Fig. [64]S2A and Fig. [65]S2B), providing a
   comprehensive view of chromatin accessibility in the t(8;21) AML
   samples.

   We compared the single-cell chromatin landscape of t(8;21) AML samples
   to that of healthy individuals [[66]15]. Consistent with the findings
   from the scRNA-seq, leukemia samples exhibited a substantial
   discrepancy in chromatin accessibility compared to normal controls
   (Fig. [67]1E). However, in terms of the distribution of peaks, there
   was no significant difference between leukemia samples and healthy
   controls (Fig. [68]1F).

   Next, we focused on the role of the AML1-ETO fusion protein in t(8;21)
   AML, which is known to act through the AETFC [[69]3, [70]4].
   Specifically, we examined the transcription factors associated with the
   AETFC, including E proteins TCF12 (HEB) and TCF3 (E2A), using the the
   scATAC-seq data. Compared to healthy controls, t(8;21) AML samples
   presented increased chromatin accessibility at loci of TCF12, LYL1, and
   RUNX1 (Fig. [71]1G), indicating the pathogenesis of AML1-ETO in t(8;21)
   AML.

Chromatin landscape of cell type-specific gene expression in primary t(8;21)
AML

   To decipher the characteristics of the leukemic cells in t(8;21) AML,
   we analyzed the scRNA-seq data from three samples from newly-dignosed
   patients (n = 3) and revealed the major cell clusters (Fig. [72]2A and
   B and Fig. [73]S3A). The cell cluster characterized by high expression
   of CD79A, JCHAIN and IGKC, was characterized as B cells, whereas the
   cell cluster expressing CD3D, CD5 and CD7 was identified as T cells
   (Fig. [74]2C and S3B). The cells that highly expressed S100A8, S100A9
   and FCN1 were identified as monocytes.

Fig. 2.

   [75]Fig. 2
   [76]Open in a new tab

   single-cell RNA-seq profiling of the leukemia clusters in primary
   t(8;21) AML. (A). UMAP demonstrated the annotation of each cluster in
   primary t(8;21) AML samples (n = 3). (B). Bar plot showed the
   percentage of each cluster from each sample. (C). Dot plot displayed
   the highly expressed genetic markers of each annotated cell cluster.
   Dot indicates the percentage of expression in each cluster and color
   represents the average expression. (D). Heatmaps displayed the top
   specific pathways that were enriched in each cluster

   We observed significantly higher expression levels of RUNX1T1 in
   leukemic cells than in cells from the tumor microenvironment (TME)
   (Fig. [77]S3C). Among the leukemic cells, seven distinct cell
   populations were observed. The cell cluster with high expression levels
   of ELANE, PRTN3 and AZU1 was identified as Leuk_abnormal myeloid cells
   with partial maturation (AM). Cell clusters of leukemic cells at cell
   cycle of G2M phase (highly expressing UBE2C and CDC20) and S phase
   (highly expressing TYMS and TK1) were annotated as leuk_G2M and leuk_S,
   respectively (Fig. [78]2C and Fig. [79]S3D). GSEA analyses revealed
   significant activation of the cell cycle pathways of these two cell
   clusters, which was consistent with their respective cell cycle stages
   (Fig. [80]2D). Additionally, cell cluster highly expressing EGR1,
   POU4F1, KIT and FLT3 was annotated as leuk_ribosome. Pathway analysis
   of this cluster revealed enrichment of RNA polymerase and ribosome
   biogenesis in eukaryotes (Fig. [81]2D). Cell cluster that presented
   high expression of CRIP1, EMP3 and LGALS1 was annotated as
   leuk_adhesion. A small cluster, characterized by high expression of
   CCL4 and CCL3, was designated leuk_chemokine. Last but not least, a
   subset of cells with elevated expression of TPSAB1, HPGD and FCER1A,
   showing a common myeloid progenitor (CMP) signature, was referred to as
   leuk_CMP-like.

   We then investigated the chromatin remodeling across distinct cell
   clusters and characterized the major cell populations identified by
   scRNA-seq from bone marrow of primary t(8;21) AML. Cell-specific marker
   genes derived from scATAC-seq were calculated. We further confirmed the
   cell identities (Fig. [82]3). B cells identified through scATAC-seq
   presented specific enrichment of MS4A1, BLC11A, and IRF8 peaks (Fig.
   [83]S5A), whereas T cells showed notable enrichment of CD3E. Monocytes
   presented specific enrichments of FCN1, CEBPA and CEBPB (Fig. [84]3B
   and C and Fig. [85]S5B). These findings further verified the identities
   of the cell clusters observed in the scRNA-seq data.

Fig. 3.

   [86]Fig. 3
   [87]Open in a new tab

   Single-cell chromatin accessibility profiling revealed the cell-cluster
   specific epigenetic features in t(8;21) AML. (A). (Left) UMAP showed
   the chromatin accessibility of 28,977 single cells from t(8;21) AML
   (n = 3). Dot indicates individual cells. (Right) Projection of leukemia
   cells in each sample. (B). Cell-specific accessible peaks with
   representative associated genes. Genome tracks showed the chromatin
   accessibility of MS4A1 (CD20) for B cells, FCN1 for monocytes and CD3E
   for T cells. (C). Heatmaps displayed the differentially expressed
   transcription factors in identified cell clusters in scATAC-seq

Heterogeneous clonality and chromatin accessibility of T cell subsets

   Integration of scRNA-seq and scATAC-seq revealed two distinct subsets
   of T cells, with T cells_1 characterized by the expression of CCR7 and
   LTB, and T cells_2 marked by GNLY, NKG7 and GZMB (Fig. [88]4A). In our
   previous longitudinal study of t(8;21) AML [[89]16, [90]17], we
   discovered a subset of T cells expressing GZMB, NKG7, and GNLY that
   expanded during the drug-resistant stage, of which the extracted gene
   signature was shown to be a potential biomarker of prognosis for AML
   [[91]17].

Fig. 4.

   [92]Fig. 4
   [93]Open in a new tab

   Single-cell TCR repertoire profiling and single-cell ATAC-seq analysis
   revealed heterogeneous clonality and chromatin accessibility of T cells
   subsets. (A). UMAP displayed the expression of T cells_1 marker genes
   (CCR7, LTB) and T cells_2 marker genes (GNLY, NGK7, GZMB) in the
   scRNA-seq landscape of t(8;21) AML (n = 3). (B). Clonotypes of the two
   clusters of T cells in each sample analyzed with scTCR. (C). Plots
   showing the most accessible motifs observed in T cells_1 (IRF8, IRF9,
   ETS1) and T cells_2 (EOMES, TBX10, TBR1). X axis showed the enriched
   transcription factors and Y axis showed the adjusted p value for the
   motif enrichment. (D). Deviation of EOMES in the clusters of t(8;21)
   AML samples analyzed with chromVAR. (E). UMAP showed the transcription
   accessibility of EOMES. (F). Footprint of subtype-specific
   transcription factor EOMES in T cells_1 and T cells_2. (G). Cell
   communication plot analyzed from scRNA-seq showed the interaction of T
   cells_1 and T cells_2 with the other cells

   To examine the clonality and rearrangements of these two subsets of T
   cells (T cells_1 and T cells_2), as well as their contributions to
   leukemia progression, we performed paired scTCR profiling on 1,099 T
   cells (Fig. [94]4). We observed significant heterogeneity in the TCR
   rearrangements and immune repertoire, presenting preferential usage of
   TCRα chains and TCRβ chains, as well as variable VJ gene combinations,
   across different AML samples (Fig. [95]S3E).

   Next, we assessed the clonality of the T cell populations with
   reference to previous studies [[96]18] and classified T cells clones
   according to the frequency into hyperexpanded, large, medium, small and
   single clones. There was no hyperexpanded or large clones within the
   two T cells subsets. Furthermore, we observed that T cells_2 exhibited
   a slightly greater proportions of small and medium clone types compared
   to T cells_1 (Fig. [97]4B and Fig. [98]S3F). These results indicated
   that T cells in t(8;21) AML demonstrated low clonal expansion, with
   most T cells expressing unique TCRs. Additionally, when calculating the
   overlap of TCR repertoires to assess their similarity, the Morisita
   index also indicated minimal overlap, supporting the notion of a
   diverse and individualized T cell repertoire across these primary AML
   samples (Fig. [99]S3G).

   We next investigated the chromatin regulatory programs underlying the
   distinct T cell subsets (Fig. [100]4C). For the T cells_1 cluster,
   scATAC-seq data showed that IRF8, IRF9, ETS1 and ELK4 were among the
   predominant transcription factors. ETS1 has been demonstrated to play a
   crucial role in CD8 T cell differentiation [[101]19]. In contrast, T
   cells_2 cluster presented a higher enrichment of transcription factor
   motifs associated with EOMES, TBR1 and TBX10. EOMES is responsible for
   the transcriptional regulation of the exhausted markers PD-1 and TIM-3
   [[102]20], which can lead to an inferior outcome in AML. The analyses
   of scATAC-seq data from primary t(8;21) AML further confirmed
   significantly increased accessibility of EOMES within T cells_2
   (Fig. [103]4D, E and F). In addition, T cells_2 presented a weaker
   cellular communication with other leukemic cells (Fig. [104]4G). In
   brief, we identified that T cells_2, which had an epigenetic feature of
   enrichment of EOMES motifs in opening peaks, presented heightened
   clonality and a propensity for expansion after drug resistance, which
   might contribute to a poorer prognosis for patients with AML.

Accessibility of transcription factors that inhibit the differentiation state
in primary t(8;21) AML

   We next explored the epigenetic features across the leukemic cell
   clusters. Our analysis identified a significant decrease in the overall
   accessibility of transcription factors within the leukemic cell
   cluster, indicating that the overall cell differentiation was in an
   inhibited state in the primitive cell subpopulations (Fig. [105]5A).
   scATAC-seq analysis showed that TCF12 and ASCL1 were among the top
   transcription factors within the primitive cell subpopulations
   (Fig. [106]5B). Plot tracks and motif matrix analyses demonstrated high
   accessibility of TCF12 and ASCL1 in leuk-ribosome cells (Fig. [107]5C
   and D). TCF12, also known as HEB, participates in the pathogenesis of
   t(8;21) AML as part of the AETFC complex, and plays an important role
   in the development of leukemia [[108]3]. Comparative genomic analysis
   revealed striking similarities in the binding patterns of AML1-ETO and
   HEB, with both factors exhibiting a preference for promoter-proximal
   regions and demonstrating comparable distributions relative to
   transcription start sites (TSS) (Fig. [109]S4A and B). Consistent with
   these findings, we identified the co-binding of AML1-ETO and HEB at key
   target loci (including EP300, HDAC7, RUNX3, and TP53) (Fig. [110]S4C),
   providing evidence for their cooperative role in transcriptional
   regulation. The increased accessibility of TCF12 and its synergistic
   interaction with AML1-ETO might collectively contribute to the altered
   regulatory landscape and consequent cellular state alterations observed
   in t(8;21) AML. In addition, we observed that another transcription
   factor, ASCL1, showed increased accessibility (Fig. [111]5E), which has
   not been previously reported. Our research indicated the potential
   value of ASCL1 in the pathogenesis in t(8;21) AML.

Fig. 5.

   [112]Fig. 5
   [113]Open in a new tab

   Single-cell chromatin profiling of ribosome leukemia clusters in
   primary t(8;21) AML. (a). Heatmap of peak marker from scATAC-seq showed
   the inhibited state in leuk_ribosome. (B). Plot displayed the top rank
   transcription factors for ribosome leukemia cells. Among the top
   including TCF12, SPIB and ASCL1. (C). Deviation of TCF12 and ASCL1 in
   the clusters of t(8;21) AML samples analyzed with chromVAR. (D). UMAP
   plot showed the transcription accessibility of TCF12 and ASCL1. (E).
   Footprint of subtype-specific transcription factor TCF12 and ASCL1 in
   the clusters of t(8;21) AML samples

   For the leuk_Cell Cycle cluster, we observed an enrichment of the
   transcription factors SP6, KLF4 and WT1 (Fig. [114]S5C-E, Fig.
   [115]3C), all of which have been reported to be involved in cell
   proliferation [[116]21, [117]22]. The increased transcriptional
   activity of WT1 facilitated leukemia blasts’ progression into the cell
   cycle, promoting the transition from G1 to S phase, and ultimately
   resulting in increased cell proliferation.

Accessibility of CMP-like leukemia cluster in primary t(8;21) AML

   We identified a distinct cluster of cells, characterized by high
   expression of TPSAB1, HPGD and FCER1A, designated the leuk_CMP-like
   cluster. KEGG pathway enrichment analysis revealed the activation of
   several pathways in this cell population, including the MAPK,
   chemokine, B cell receptor, and NOD-like receptor signaling pathways
   (Fig. [118]6). Joint analysis of scATAC-seq and scRNA-seq data revealed
   that the epigenetic features of the GATA family members GATA2, GATA1,
   GATA3 and GATA5 were significantly enriched within the leuk_CMP-like
   cluster (Fig. [119]6). These transcription factors are known to play
   important roles in lineage specification and transdifferentiation in
   hematologic disorders. We thus hypothesized that the genetic signatures
   presented within this cluster, whose expression levels are rigorously
   transcriptionally regulated by the GATA family, might serve as valuable
   prognostic indicators in AML. To this end, we calculated and extracted
   the genetic signature of the leuk_CMP-like cluster through joint
   analyses of scATAC-seq and scRNA-seq data. A distinct gene expression
   profile of this cluster was defined by genes exhibiting an average log2
   (fold change) more than 3.0 and a p value less than 0.05. A total of
   136 genes were extracted (Table [120]S2), exhibiting stemness- and
   progenitor-specific characteristics, marked by the high expression of
   relevant genes associated with AML, such as CD34, CD38 and RUNX1T1.

Fig. 6.

   [121]Fig. 6
   [122]Open in a new tab

   Single-cell chromatin profiling of Leuk_CMP-like cluster in primary
   t(8;21) AML. (A). UMAP plot showed the expression of FCER1A and IL5RA
   from scRNA-seq of primary t(8;21) AML (n = 3). (B). Kyoto Encyclopedia
   of Genes and Genomes (KEGG) analysis of the top enriched pathways in
   the Leuk_CMP-like cluster from scRNA-seq. (C). Genome browser tracks
   showed chromatin accessibility of Leuk_CMP-like cluster-specific genes
   FCER1A and IL5RA for the cell clusters in scATAC-seq. Each track showed
   merged signal. (D). Plot displayed the top rank transcription factors
   for Leuk_CMP-like cluster. Among the top including GATA2, GATA1 and
   GATA3. (E). Footprint of subtype-specific transcription factor GATA2,
   GATA1 and GATA3 in the clusters of t(8;21) AML samples

   Lastly, we developed a deconvolution-based prediction procedure to
   evaluate the potential prognostic value of the leuk_CMP-like signature
   by utilizing bulk RNA-seq data from the AML cohort. The GSE37642_GPL96
   (German AMLCG1999) dataset was used as the training dataset. By
   employing LASSO regression analysis, we constructed a leuk_CMP-like
   score derived from the estimated proportions of CMP-like cells in each
   patient. The final model incorporated nine key genes (Fig. [123]7A and
   B), specifically BACE2, EPX, FRY, IL18R1, IL5RA, PVALB, SCG2, TIMP3 and
   TPSAB1 (Fig. [124]7B), of which TIMP3 and PVSLB emerged as strong
   prognostic indicators of clinical outcome in patients with AML. We
   subsequently examined the relationship between the Leuk_CMP-like score
   and the clinical outcomes of AML patients. Remarkably, a high CMP-like
   score (high-risk subgroup) was significantly associated with inferior
   overall survival (P < 0.001, Fig. [125]7C) in the training cohort.
   Moreover, we observed similar findings in the validation cohorts of
   GSE37642_GPL570, [126]GSE106291 and TCGA LAML (Fig. [127]7D),
   reinforcing the prognostic significance of the leuk_CMP-like score.

Fig. 7.

   [128]Fig. 7
   [129]Open in a new tab

   Construction and validation of Leuk_CMP-like cluster related genetic
   signatures in AML. (A). Cross-validation and coefficients of LASSO
   regression to determine the prognostic model. (B). Forest plot showed
   the Cox regression model. Survival plots of Leuk_CMP-like cluster in
   training cohort (C) and validation cohorts (D). Log-rank tests were
   applied to compare the survival difference in the two risk groups

Discussion

   In this work, we characterized the cellular heterogeneity and global
   chromatin profiling in AML with t(8;21) translocation. Using
   scRNA-seq-based gene expression profiling, scATAC-seq-based chromatin
   accessibility profiling and scTCR-seq-based clonotype clustering, we
   identified distinct heterogeneous cell populations and explored the
   epigenetic cis-regulatory mechanism underlying t(8;21) AML. Our study
   provided the first paired single-cell transcriptomic and epigenomic
   atlas of t(8;21) AML, addressing critical knowledge gaps in
   understanding this specific AML subtype. To our knowledge, this study
   is the first integrative analysis utilizing multiple single-cell
   technologies to study the cellular and molecular underpinnings of
   t(8;21) AMLand to compare the results with those of healthy
   individuals. While previous pan-AML scRNA-seq and scATAC-seq studies
   have revealed AML cellular heterogeneity and regulatory mechanism,
   these investigations have either focused primarily on pediatric AML
   [[130]23], or examined chemotherapy-resistant post-treatment samples
   [[131]24]. Importantly, none have systematically explored the molecular
   regulatory mechanisms of the AML1-ETO fusion protein at single-cell
   resolution in treatment-naïve adult t(8;21) AML patients. Our
   multi-omics approach uniquely identifies t(8;21) AML-specific
   regulatory networks and cellular states that were previously obscured
   in bulk analyses [[132]25].

   One of the prominent discoveries of this study was the observation that
   the family members of the AETFC, namely TCF12 (HEB), TCF3 (E2A) and
   LYL1, presented increased chromatin accessibility in leukemic cells of
   t(8;21) AML. The AETFC complex plays a crucial role in the regulation
   of hematopoietic differentiation and the maintenance of multivalent
   interactions of diverse transcription factors, including the E proteins
   HEB and E2A, the AML1-binding partner CBF-β and the hematopoietic
   E-box-binding transcription factor LYL1, all of which contribute to
   leukemogenesis. Under the pathological state of t(8;21) AML, the
   AML1-ETO fusion protein aberrantly modulates these transcription
   factors, leading to increased chromatin accessibility at their target
   loci. This enhanced chromatin openness likely facilitates the binding
   of other transcription factors or co-factors, thereby promoting the
   expression of various target genes that contribute to leukemic
   transformation. Our findings provide valuable insights into the
   transcriptional dysregulation underlying the pathogenesis of AML1-ETO.
   To therapeutically disrupt the functional cooperation between TCF12/HEB
   and AML1-ETO, small-molecule inhibitors could be developed to target
   either their interaction domains or DNA-binding motifs. For example,
   synthetic peptides mimicking the E-protein interaction interface of
   AML1-ETO could serve as competitive inhibitors to effectively block
   this oncogenic partnership [[133]26, [134]27]. Furthermore, we also
   identified another transcription factor, ASCL1, presenting increased
   accessibility in leukemic cells in t(8;21) AML patients. ASCL1 encodes
   a member of the basic helix-loop-helix (BHLH) family of transcription
   factors, which acts as a master regulator of neurogenesis and has been
   shown to be overexpressed in neuroendocrine tumors [[135]28]. It has
   also been reported as a promising therapeutic target in small cell lung
   cancer [[136]29]. Although ASCL1 is a well-established regulator in
   neuroblastoma and other cancers, its role in hematopoietic malignancies
   is still under investigation. While ASCL1’s role in AML is emerging,
   its established function in neuroendocrine cancers suggests the
   tractability. Pharmacological degradation (e.g., proteolysis-targeting
   chimeras) or transcriptional suppression (e.g., CRISPR-dCas9) of ASCL1
   could be explored, especially given its aberrant accessibility in
   leukemic cells. Research is ongoing to better understand the molecular
   mechanisms by which ASCL1 promotes leukemogenesis in AML
   [[137]30–[138]33]. The above findings not only highlight the importance
   of transcriptional regulation in t(8;21) AML, but also provide
   potential therapeutic targets, as intervening in the activity of these
   transcription factors or their downstream pathways could offer new
   strategies for the treatment of t(8;21) AML.

   Another critical aspect of AML is the interaction between leukemic
   cells and the surrounding microenvironment, which plays pivotal roles
   in disease progression, chemoresistance, and relapse. In our previous
   work, we reported that a subset of T cells is associated with disease
   progression in AML [[139]17]. By incorporating scTCR-seq in this study,
   we further revealed a diverse T cell repertoire in patients with
   t(8;21) AML at initial onset, with T cell_2 clusters exhibiting an
   oligoclonal TCR repertoire and presenting key associations with drug
   resistance. In the T cell_2 cluster, the transcription factor EOMES was
   identified as a pivotal regulatory factor. EOMES is known to be
   involved in the differentiation of cytotoxic T lymphocytes and natural
   killer cells, playing a critical role in their function and survival
   [[140]34–[141]36]. Several studies have reported a potential link
   between EOMES and T cell exhaustion, particularly regarding exhaustion
   markers such as PD-1 and TIM-3 [[142]20, [143]37, [144]38]. PD-1 and
   TIM-3 are key checkpoint molecules that are highly expressed on
   exhausted T cells, which often accumulate in the tumor
   microenvironment, thus leading to T cell dysfunction. This observation
   has important translational implications. Immune checkpoint inhibitors
   (ICIs) targeting PD-1/PD-L1 or TIM-3 have shown efficacy in solid
   tumors and hematologic malignancies [[145]39, [146]40]. Given the
   association between EOMES and these checkpoint molecules, combinatorial
   strategies, such as EOMES-modulating agents alongside ICIs, could be
   explored to reverse T cell exhaustion and restore anti-leukemic
   immunity. Additionally, since oligoclonal T cell expansions may
   indicate antigen-driven responses, identifying the cognate antigens
   recognized by these T cells could uncover novel targets for
   immunotherapy, such as TCR-engineered T cells or vaccines.

   We also identified a subset of cells exhibiting CMP-like gene
   characteristics with high expression of marker genes specific to early
   progenitors, such as CD34, CD38 and GATA1, possibly representing
   leukemic CMP-like cells. Integrated analysis of the scRNA-seq and
   scATAC-seq data revealed that this cluster highly expressed TPSAB1,
   HPGD, and FCER1A, which was under the transcriptional regulation of
   GATA family members, including GATA2, GATA1, GATA3 and GATA5. A number
   of studies have reported that the GATA protein family comprises vital
   pluripotency-associated transcription factors determining the
   identities of pluripotent and differentiated cells [[147]41–[148]43].
   In addition, recent studies demonstrate that GATA2 dosage controls the
   generation and maintenance of the hematopoietic stem cell, with
   overexpression promoting stemness and is more resistant to the standard
   AML chemotherapy agent doxorubicin [[149]44]. Our results further
   revealed that the GATA family participated in the epigenetic
   reprogramming of leukemic cells with chromatin accessibility
   alterations, and is thus involved in stem cell maintenance and
   leukemogenesis promotion in t(8;21) AML. The co-expression of FCER1A
   (high-affinity IgE receptor) and HPGD (prostaglandin catabolizing
   enzyme) suggests potential microenvironmental interactions.
   HPGD-mediated regulation might exert tumorigenic effects through
   non-enzymatic mechanisms [[150]45]. These might create a pro-leukemic
   niche favoring immune evasion, consistent with our finding of
   dysfunctional cytotoxic T cells in patients with t(8;21) AML. These
   findings are consistent with the emerging concept of lineage infidelity
   in AML [[151]46], wherein epigenetic dysregulation drives cells into
   aberrant states. Future studies should functionally validate the
   leukemogenic potential of this CMP-like population through
   xenotransplantation models and explore whether its prevalence
   correlates with treatment outcomes.

   While our study provides valuable insights into the epigenetic
   landscape of AML, we acknowledge several limitations that warrant
   consideration. First, our analysis was restricted to three t(8;21) AML
   samples and two healthy controls due to the rarity of high-quality,
   multi-omics-matched specimens. While this sample size allowed us to
   identify robust cell-type-specific regulatory signatures, larger
   cohorts would improve statistical power to detect subtle heterogeneity
   and rare subpopulations. Future studies with expanded sample sizes are
   needed to validate and generalize our findings across the heterogeneous
   t(8;21) AML subtypes and clinical contexts.

   Addtionally, our scATAC-seq and scRNA-seq analyses were limited to
   baseline diagnostic samples, which was necessitated by the practical
   challenges associated with obtaining longitudinal bone marrow
   specimens. Future investigations incorporating paired diagnosis-relapse
   samples or serially collected post-treatment specimens would
   substantially enhance our understanding of how transcriptional and
   epigenetic reprogramming drives therapeutic resistance and relapse
   mechanisms, thereby offering opportunities to improve clinical outcomes
   through more targeted therapeutic strategies. Such investigations could
   capture the dynamic changes in chromatin accessibility during disease
   progression or in response to treatment, which is a critical aspect for
   comprehensively elucidating the epigenetic evolution of AML.

   Finally, while single-cell multi-omics technologies offer unprecedented
   resolution, they are subject to inherent technical limitations,
   including gene dropout effects in scRNA-seq, sparse coverage in
   scATAC-seq, and potential batch effects across samples. Computational
   challenges in integrating multimodal data may also influence
   interpretability. We mitigated these issues through rigorous quality
   control, batch correction, and cross-modal integration approaches, but
   residual confounding factors might persist. Advances in both
   experimental protocols and bioinformatic pipelines will further enhance
   the reliability of single-cell studies in t(8;21) AML.

Conclusions

   In summary, this multi-omics study, which integrates scRNA-seq,
   scATAC-seq and scTCR-seq, establishes a framework for a more
   comprehensive understanding of the gene expression alterations and
   transcriptional regulatory networks, as well as the role of immune cell
   dynamics in TME in t(8;21) AML. The high concordance between
   transcriptional- and epigenomic-based subtypes, together with the TCR
   repertoire profiling of the TME, revealed strong biological differences
   between distinct cell populations. Moreover, the inclusion of a genetic
   signature could be used as a novel and useful prognostic indicator for
   patients with t(8;21) AML.

Electronic supplementary material

   Below is the link to the electronic supplementary material.
   [152]Supplementary Material 1^ (5.9MB, zip)

Acknowledgements