09 Cellular Hierarchy Insights Reveal Leukemic Stem-Like Cells and Early Death Risk in Acute Promyelocytic Leukemia

Article https://doi.org/10.
1038/s41467-024-45737-7
Cellular hierarchy insights reveal leukemic

stem-like cells and early death risk in acute
promyelocytic leukemia
Received: 20 April 2023 Wen Jin 1,2,4, Yuting Dai 1,4, Li Chen 1, Honghu Zhu 3, Fangyi Dong1,
Hongming Zhu1, Guoyu Meng 1, Junmin Li1, Saijuan Chen 1, Zhu Chen1,2 ,
Accepted: 2 February 2024
Hai Fang 1 & Kankan Wang 1,2
Check for updates Acute promyelocytic leukemia (APL) represents a paradigm for targeted dif-
1234567890():,;
1234567890():,;
ferentiation therapy, with a minority of patients experiencing treatment failure

and even early death. We here report a comprehensive single-cell analysis of 16
APL patients, uncovering cellular compositions and their impact on all-trans
retinoic acid (ATRA) response in vivo and early death. We unveil a cellular
differentiation hierarchy within APL blasts, rooted in leukemic stem-like cells.
The oncogenic PML/RARα fusion protein exerts branch-specific regulation in
the APL trajectory, including stem-like cells. APL cohort analysis establishes an
association of leukemic stemness with elevated white blood cell counts and
FLT3-ITD mutations. Furthermore, we construct an APL-specific stemness
score, which proves effective in assessing early death risk. Finally, we show that
ATRA induces differentiation of primitive blasts and patients with early death
exhibit distinct stemness-associated transcriptional programs. Our work pro-
vides a thorough survey of APL cellular hierarchies, offering insights into cel-
lular dynamics during targeted therapy.
The landscape of cancer-targeted therapies has shifted from molecular the stability of PML/RARα to reverse transcriptional deregulation,
target identification to cellular heterogeneity characterization and affecting the proliferation and differentiation of APL cells. It is well-
targeting1,2. The direct targeting of molecular changes driving tumor- established that intratumoral heterogeneity at the cellular level is a
igenesis has proven to enhance therapeutic efficacy in cancer, as critical factor in leukemogenesis, disease progression, and therapy
exemplified by the success of molecular targeted therapy in acute response14. Advances in single-cell genomics technologies have
promyelocytic leukemia (APL)3. APL is characterized by its driver uncovered the cellular heterogeneity of chronic myeloid leukemia and
oncogenic fusion protein (PML/RARα), which plays a crucial role in lung cancer1,15, demonstrating the potential for exploring hetero-
initiating APL leukemogenesis, as supported by substantial evidence, geneous features to improve cancer therapy success rates. In the
including insights from murine models4,5. Notably, two drugs, all-trans context of APL, it has now become imperative to explore whether
retinoic acid (ATRA) and arsenic trioxide (ATO), have already achieved leukemic cells driven by the same oncogenic driver, PML/RARα,
remarkable therapeutic outcomes by directly targeting this oncogenic exhibit diverse cellular states; if so, to what extent the cellular com-
fusion protein6. We7–11 and others12,13 have long been focusing on position and transcriptional heterogeneity might impact the outcomes
molecular mechanisms to illustrate how ATRA and ATO directly target of targeted therapy in APL.
1
Shanghai Institute of Hematology, State Key Laboratory of Medical Genomics, National Research Center for Translational Medicine at Shanghai, Ruijin
Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai 200025, China. 2Sino-French Research Center for Life Sciences and Genomics, Ruijin
Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai 200025, China. 3Department of Hematology, Beijing Chao-Yang Hospital, Capital
Medical University, Beijing 100020, China. 4These authors contributed equally: Wen Jin, Yuting Dai. e-mail: [email protected]; [email protected];
[email protected]
Nature Communications | (2024)15:1423 1

Article https://doi.org/10.1038/s41467-024-45737-7
Long-term therapeutic efficacy in cancer seems to be attributable as controls. We adjusted for batch effects and performed an integrated
to targeting a rare cell population with stemness potentials16–18, such as analysis of both APL and normal cell populations to characterize the
leukemic stem cells (LSCs) in APL, although APL LSCs are far from malignant APL blasts and intratumoral heterogeneity (detailed in
clear, particularly in terms of cell-of-origin and their relationship with “Methods” section).
genetic events. Studies in mice have suggested that APL might arise Based on the APL blasts characterized in this study, we designed a
from myeloid-committed progenitors, including committed myeloid series of analyses and validations as follows: (i) we constructed the
progenitors (CMPs) and granulocyte-monocyte progenitors cellular architecture and differentiation trajectory of APL blasts, with a
(GMPs)19–21. Alternatively, APL LSCs might arise from more primitive specific focus on identifying APL stem-like cells; (ii) we validated the
progenitors that are earlier than CMPs, as PML/RARα has been expression of the PML/RARα fusion gene and FLT3-ITD using targeted
reported to be detectable in both CD34+CD38+ and CD34+CD38− scRNA-seq (scTarget) in two de novo APL patients; (iii) we determined
populations isolated from APL patients22. To reconcile these seemingly the association between cellular compositions and clinical presenta-
conflicting observations, it is imperative to gain a comprehensive tions, including the incidence of early death, by conducting decon-
understanding of APL cellular compositions, with a particular volution analysis on a large cohort of 323 APL patients (including 22
emphasis on the rare LSCs. patients with early death); (iv) we explored the impact of ATRA therapy
Furthermore, extensive investigations into the molecular on changes in cellular compositions for three patients, both at disease
mechanisms underlying effective targeted therapies in APL have pri- onset and on Day 2 after ATRA treatment; and (v) we performed
marily relied on in vitro analyses or in vivo murine models3,7,8. These deconvolution analysis on RNA-seq data from 10 newly diagnosed APL
studies have sought to elucidate the effectiveness of ATRA and/or ATO patients before and after ATRA treatment to establish the association
in reversing the aberrant transcriptional regulatory activity of PML/ of cellular compositions, especially APL stem-like cells, with ATRA
RARα. Additionally, functional analyses of cell behavior have show- responses in vivo.
cased the efficient induction of terminal granulocyte
differentiation3,7,8. However, the genuine in vivo responses of APL cells Single-cell characterization of APL blasts
to ATRA, particularly its impact on different cellular compartments, To unravel the cellular diversity within APL, we first performed analysis
remain ill-defined. This limitation restricts our ability to explain the by comparing BM samples collected from APL patients with those from
effectiveness of ATRA in treating APL patients with varying prognoses. healthy individuals. Using 23 normal BM samples, we established the
In this work, we use the single-cell RNA sequencing (scRNA-seq) baseline of cellular diversity, which revealed seven major cell popula-
technology to comprehensively dissect the cellular heterogeneity in tions, consistent with previously published findings23,24. These popu-
APL and its potential impact on ATRA therapy in vivo and early death, lations included hematopoietic stem/progenitor cells (HSPCs), GMPs,
as outlined in Fig. 1a. We generate a single-cell transcriptome resource monocytes (Mono), dendritic cells (DCs), B cells, T/Natural Killer (NK)
on the malignant APL blasts from 16 newly diagnosed APL patients. cells, and erythroid (Ery) cells (detailed in “Methods” section and
Subsequently, we conduct a series of data analyses to gain deeper Supplementary Fig. 1).
insights into APL cellular heterogeneity and its association with Next, we employed UMAP to project 136,497 cells from 16 newly
genomic and clinical characteristics of APL patients. The resources and diagnosed APL patients, along with normal hematopoietic cell popu-
findings presented in this study hold significant implications in four lations identified above, onto a two-dimensional space (Fig. 1b). While
aspects. Firstly, our resource enables the characterization of intratu- lymphoid and erythroid populations from APL patients formed clus-
moral heterogeneity with multiple branches, including a small sub- ters that corresponded to cell types also annotated by normal hema-
population of APL stem-like cells at the root of the differentiation topoietic cells, APL cells were predominantly grouped into a distinct
trajectories. Secondly, at the single-cell level, we show that the stem- cluster. The accuracy of APL blast identification was confirmed by the
ness characteristics of APL stem-like cells are determined by PML/ exclusive presence of PML/RARα expression, as observed in two APL
RARα target genes and can be further enhanced in the presence of patient samples using PML/RARα-targeted scRNA-seq (Fig. 1b, right
FLT3-ITD. Thirdly, deconvolution analysis conducted on a large cohort panel, and Supplementary Fig. 2). This cluster exhibited a high
of 323 APL patients reveals a significant association of higher APL stem- expression level of the gene MPO, which encodes a widely used diag-
like cell proportions with elevated white blood cell (WBC) counts and nostic marker for APL25, and also differed from the normal spectrum of
the presence of FLT3-ITD. We also successfully construct an APL- hematopoietic cell populations (such as HSPCs and myeloid cell
specific stemness score, which effectively assesses prognosis, espe- populations; Fig. 1b, c). The established markers of APL cells were
cially the risk of early death in APL patients. Lastly, our single-cell level highly expressed in this cluster, as highlighted by genes activated by
investigations into the in vivo effects of ATRA confirm that ATRA PML/RARα-associated super-enhancers8 (e.g., STAB1, CITED2, CCND2,
directly targets APL primitive blasts, leading to their differentiation and GFI1) (Fig. 1d). This finding further reinforced our previous find-
and maturation. ings, highlighting the role of PML/RARα in determining the identity of
APL blasts through super-enhancer regulation8. Moreover, this cluster
Results exhibited significantly elevated expression levels of GMP-specific
Study design and analysis overview genes, such as azurophilic granule genes (MPO, AZU1, ELANE, and
We performed single-cell transcriptome analysis on bone marrow (BM) CTSG) (Fig. 1d), supporting the notion that APL blasts may be blocked
samples collected from 16 newly diagnosed APL patients (Fig. 1a; also at the GMP stage14,23.
see “Methods” section for patient selection and Supplementary Data 1 We proceeded to compare the functional and regulatory char-
for detailed clinical characteristics and sample information). This acteristics of the APL blasts with those of the normal GMP cluster.
cohort included four patients who experienced early death, which is Firstly, we conducted gene ontology (GO) enrichment analysis on
defined as death occurring within 30 days from diagnosis. In this differentially expressed genes (DEGs) calculated using Seurat26. The
endeavor, we generated a total of 136,497 cells by combining 16 sepa- analysis revealed that significantly upregulated genes (adjusted P-
rate scRNA-seq datasets of APL BM samples at disease onset, forming value < 0.05) in APL blasts were of functional relevance to several key
the foundation for a comprehensive understanding of APL cellular processes, including HSC self-renewal/differentiation (RUNX1, MYC,
composition. In parallel, we reanalyzed 23 separate scRNA-seq data of and JAG1), histone modification (EP300) and DNA methylation
normal BM samples from healthy individuals (Gene Expression (DNMT3A and MBD1), cell cycle arrest and cell growth (CDK6, CCNA1,
Omnibus with accession IDs of GSE120221 and GSE130116) to construct and WT1), as well as the response to endoplasmic reticulum stress and
normal hematopoietic cell populations (totaling 102,792 cells) for use unfolded protein (XBP1, ATF6, and USP14) (Fig. 1e and Supplementary

16 scRNA-seq (10X) of 23 scRNA-seq (10X) of

de novo APL bone marrow cells healthy bone marrow cells
(including 4 early deaths) (public data)
Batch effects adjusted APL stem-like
Characteristic of malignant APL blasts Intra-tumoral heterogeneity of APL blasts
scTarget APL cohort In vivo treatment In vivo treatment
RA RA
AT AT
Day 0 Day 2 Day 0 Day 2

2 de novo APLs 323 de novo APLs 10 APLs, bulk RNA-seq
with FLT3-ITD (22 early deaths) 3 APLs, scRNA-seq
(3 early deaths)
lls
s
T/ lls ell
P C s ts
ce
B driti s
ce c c
yt lls
n te
d
HS bla
Er K ce
De ocy
oi
M s
M Ps
hr
b d
on
N
AP
Sample type Cell populations Cell populations
G
1 APL blasts 5 Dendritic cells Average
APL BM expression CLEC11A
Healthy BM 2 HSPCs 6 B cells
10 10 3 GMPs 7 T/NK cells 1.0 HGF
4 Monocytes 8 Erythroid cells 0.5 STAB1
0.0 IGFBP2
5 5 7 6 −0.5 CRNDE
Projection scTarget (KNN) MS4A3
UMAP_2
UMAP_2
PML/RARα Percent CYTL1

0 0 1 4
10 (>3 fus-reads) expressed CITED2
T/NK B 0 CCND2
APL 5 5 25
3 CCNA1
blasts 50
UMAP_2
-5 -5 2 FLT3
0 APL 75
blasts RNASE3
8
-5 PLPPR3
-10 -10 PTGDS
-10 Ery APOC2
APL blasts
FSCN1
-5 0 5 10 -5 0 5 10 -5 0 5 10 TNFRSF4
UMAP_1 UMAP_1 UMAP_1 GFI1
c MPO CD14 CD3E CD79A CA1
F11R
MEG3
OLIG1
IRX3
MMP2
Expression PLAU
high PRG2
WT1
UMAP_2
SPINK2
HSPCs FAM30A
low CD34
AVP
CRHBP
UMAP_1 PBX1
e g GMPs APL blasts Activity
HOXA9
MPO
response to endoplasmic reticulum stress
response to topologically incorrect protein AZU1
response to unfolded protein 2.1e-4 SMAD4 ELANE
proteasome−mediated ubiquitin−dependent protein catabolic # genes CTSG
IRE1−mediated unfolded protein response 9.2e-4 RNF6
40 LYST
ubiquitin−dependent ERAD pathway 3.1e-3 MAX CEBPD
ATF6−mediated unfolded protein response 80
Activated in APL blasts
120 3.2e-3 SP3 MNDA

histone modification 160 LYZ
histone acetylation 5.3e-3 ELF2
Myeloids
CES1
histone methylation -log10FDR 7.7e-3 RB1 MS4A4A
histone H3−K9 methylation FCN1
histone H3−K4 methylation 9.8e-3 HIF1A
10.0 VCAN
cell cycle arrest 1.1e-2 MBD1 S100A12
regulation of mitotic cell cycle phase transition
cell growth 5.0 3.0e-2 TOP2B CD14
3.1e-2 YY1 CLEC7A
myeloid cell differentiation FCER1A
erythrocyte differentiation 4.7e-2 PBX3
granulocyte differentiation LILRA4
5.6e-2 RXRA CLEC10A
f 3.5e-2 IRF1 CD1C
neutrophil activation involved in immune response CD79A
response to interferon−gamma 1.6e-2 FOS MS4A1
APP of peptide antigen via MHC class II IGHD
Lymphoids
1.4e-2 KLF10
itic cell antigen processing and presentation # genes
Repressed in APL blasts
1.3e-2 CD19
monocyte chemotaxis 20 IRF8
CD3E
reactive oxygen species metabolic process 40 1.3e-2 HOXA10 IL32
MyD88−dependent toll−like receptor signaling pathway 60 CD7
1.1e-2 SPI1
positive regulation of lymphocyte activation CCL5
-log10FDR 6.0e-3 ERG
T cell activation IL7R
lymphocyte proliferation 3.7e-3 STAT1 GNLY
10.0
B cell activation 1.9e-3 HES1 KLRB1
7.5
CD8A
Erythroids
positive regulation of cytokine production 5.0 1.5e-3 CEBPA

tumor necrosis factor production CA1
2.5 2.9e-4 BCL6 AHSP
interleukin−1 production
interferon−gamma production 2.2e-4 TFEC HBM
interleukin−8 production GYPA
P-value TF genes
Fig. 3a). In contrast, genes that were significantly downregulated confirmed the disruption of antigen presentation in APL cells27. Sec-
(adjusted P-value < 0.05) in APL blasts were enriched for immune ondly, we used VIPER28 to infer the transcription factor (TF) activity
response-related functions, including antigen processing and pre- differentially between the two clusters. Our analysis suggested that the
sentation (PSMB9, PSMB10, and CTSS), MHC class II protein complex hematopoietic TFs and cofactors (such as SPI1, ERG, FOS, and RXRA)
(CD74, HLA-DPA1, and HLA-DRA), regulation of cytokine production were repressed in APL blasts (Fig. 1g), supporting the differentiation
(CLEC7A, CCR2, and CCL3), and response to interferon-gamma (IRF8, blockade observed in APL blasts. Also repressed in APL blasts were the
IRF5, and IFI30) (Fig. 1f and Supplementary Fig. 3b). These findings mediators of interferon (IFN) signaling (e.g., STAT1, IRF8, and IRF1).

Fig. 1 | Characterization of APL blasts through an integrative analysis of scRNA- (f) in APL blasts compared with GMPs. g Inferred activated (red) and repressed
seq data from APL and normal bone marrow (BM) cells. a Overview of the (blue) TFs in APL blasts compared to normal GMPs. The central two-row graph
experimental strategy. b UMAP plots of APL and normal BM cells (n = 239,332 cells), illustrates the distribution of activated targets (depicted in red) and repressed
with color-coding indicating sample types (left panel), inferred cell populations targets (depicted in blue) of different TFs, with positions ranked according to the
(middle panel), and PML/RARα-positive cells detected by scTarget in two APL differential expression between APL blasts and normal GMPs (leftmost: most
samples (right panel). Cells detected with more than three PML/RARα fusion reads downregulated in APL blasts, rightmost: most upregulated in APL blasts). The
are illustrated. HSPCs hematopoietic stem/progenitor cells, GMPs granulocyte- regulatory model was based on the ARACNe-inferred interactome, provided in the
monocyte progenitors, NK Natural Killer, Ery erythroid. c UMAP plots with each cell build-in function of the VIPER R package. The P-value is shown on the left of the
(n = 239,332 cells) colored according to their normalized expression of MPO, CD14, column, and the inferred differential activity level is shown on the right. The P-
CD3E, CD79A, and CA1, respectively. d Normalized expression level and expression values were calculated using the msviper function in the VIPER R package. Two-
percentage of cell type-specific genes in eight cell populations in APL and normal sided P-values were calculated. APP antigen processing and presentation; MHC
BM cells. e, f Gene Ontology (GO) enrichment analysis showing significantly enri- major histocompatibility complex.
ched biological process terms for upregulated genes (e) and downregulated genes
Conversely, oncogenic TFs (e.g., RB1, HIF1A, and MAX), epigenetic (C17) and the CD14+ Promonocyte-like cell cluster (C18). This trajec-
regulators (e.g., SMAD4, MBD1, YY1, and SP3), and cell proliferation- tory supports the current view that, instead of being derived from
associated TFs (e.g., RNF6, ELF2, and TOP2B) were activated in APL GMPs37, MDPs might represent an earlier stage, possibly even earlier
blasts compared to normal GMPs (Fig. 1g). Consistent with these than the CMP stage. Furthermore, Monocle2 pseudotime ordering
findings, cell cycle analysis showed that the APL cluster significantly trajectory analysis also showed that the stem-like cell cluster was the
accumulated in the S or G2/M phases compared with normal myeloid starting point of APL blasts, giving rise to the S100hiGMP-like and GMP-
cells, further indicating the highly proliferative state of APL blasts like branches (Fig. 2e). This analysis also revealed that RUNX1, RUNX2,
(Supplementary Fig. 4). and interferon-related factors (i.e., STAT1 and FOS) might be involved
in the GMP-like differentiation trajectory, while CEBP family members
Characterizing the intratumoral heterogeneity of APL blasts (CEBPA, CEBPB, and CEBPE), MAFB, JUNB, and JUND were likely asso-
reveals a complex cell-state transition trajectory with leukemic ciated with the lineage decision towards the S100hiGMP-like APL
stem-like cells sitting at the top blasts (Fig. 2f).
Next, we determined the cellular architecture and differentiation tra- Furthermore, we investigated the role of PML/RARα in the
jectory within the characterized APL blasts. Through unsupervised identified APL trajectory. PML/RARα-targeted scRNA-seq confirmed
clustering and UMAP analysis, we identified 18 clusters, each char- that PML/RARα was uniformly expressed across all six branches of
acterized by distinct expression patterns of known marker genes23 APL blasts, with notable expression in the stem-like cell cluster
(Fig. 2a and Supplementary Data 2). Among these clusters, the 12 APL (Fig. 2g, h and Supplementary Fig. 7a). Furthermore, we integrated
clusters (C1-C12) accounted for 82.9% of the cells and, as expected, PML/RARα chromatin occupancy data31 obtained from CUT&Tag-seq
exhibited high expression of GMP-specific genes23, particularly those (Cleavage Under Targets and Tagmentation sequencing) in an APL
associated with azurophilic granules, such as ELANE, CTSG, and AZU1 patient-derived cell line, NB4 (Supplementary Data 4 and Supple-
(Fig. 2b, right panel). Notably, these GMP-like clusters displayed sig- mentary Fig. 7b). Notably, each branch possessed a considerable
nificant heterogeneity. For example, the three major clusters C1-C3 number of distinct PML/RARα targets (Fig. 2i and Supplementary
exclusively expressed GMP-specific genes, whereas clusters C6-C10 Data 5), suggesting the presence of branch-specific expression pat-
showed high expression of cell proliferation-related genes (TOP2A, terns for PML/RARα targets across the APL trajectory. GO analysis
MKI67, and PCNA)29, and clusters C11-C12 were marked by the highly provided insights into their functional significance, revealing that
expressed S100 family genes (S100A8/A9/A10). these targets were associated with distinct functional pathways
Of striking interest, clusters C14-C16 exhibited stemness-like (Supplementary Fig. 7c). For instance, the PML/RARα targets located
characteristics with high expression of marker genes specific to early within the APL stem-like cells were found to be predominantly
HSPCs and/or LSCs (such as CD20030, CD4431, CD9932, CD233, and involved in stem cell maintenance. Those within cycling GMP-like
FAM30A34) (Fig. 2b), master stemness-related TF genes (such as SOX435 cells were mainly linked to E2F targets involved in the G2-M check-
and MYC36), as well as APL characteristic genes (such as MPO) (Sup- point. Those within GMP-like cells exhibited marked enrichment in
plementary Fig. 5). Comparatively, C15 had the highest expression ribosomal functions. These results collectively illuminate the
levels of stemness- and progenitor-specific genes compared to C14 and impact of PML/RARα in shaping the intratumoral heterogeneity of
C16. In addition, C14 also expressed CD38, while C16 highly expressed APL cells.
marker genes associated with monocyte-DC progenitors (MDP)37, such
as CSF1R/CD115 and FLT3/CD135. Collectively, the cells in C15 appeared The characteristics of APL stem-like cells were determined by
to resemble the most primitive cells among APL blasts, possibly PML/RARα target genes and further enhanced by FLT3-ITD
representing leukemic stem-like cells. To delve into the characteristics of the APL stem-like cells defined in
Next, we performed RNA velocity analysis to verify the differ- our study, we identified APL-specific leukemic stemness genes by
entiation trajectory from the leukemic stem-like cell cluster (C15) to comparing the transcriptome data between the APL stem-like cell
the GMP-like cell clusters. The 18 APL clusters were reorganized into six cluster (C15) in Fig. 2a and the HSPC cluster in Fig. 1b (Supplementary
branches (Supplementary Data 3), with the stem-like cell cluster (C15) Data 6). We obtained the following findings.
sitting at the root of differentiation trajectories, supported by both the First, PML/RARα targets were significantly enriched in these APL-
velocity-based cell-state transition probabilities and the similarity of specific leukemic stemness genes (Fig. 3a and Supplementary Fig. 8).
expression patterns (Fig. 2c, d). Likely going through the Prog-like Notably, well-known LSC marker genes16,38–40, such as FCGR2A, CD9,
branch (C14), the stem-like cell cluster (C15) gave rise to the three main ITGA5, IL1RAP, and CD82, were exclusively expressed in APL stem-like
branches of APL blasts: the GMP-like branch (C1-C5 and C13), the cells but not in HSPCs. They were also direct targets of PML/RARα
cycling GMP-like branch (C6-C10), and the S100hiGMP-like branch (C11 (Fig. 3b and Supplementary Fig. 9). Moreover, well-known stemness/
and C12) (Fig. 2c, d and Supplementary Fig. 6). Another trajectory self-renewal-related PML/RARα targets8, such as HCK and GFI1, were
starting from C15 led to the MDP-like cell cluster (C16), which subse- highly expressed in APL stem-like cells. Genes closely related to APL
quently differentiated into cells in the CD1C+ PrecDC-like cell cluster leukemogenesis, such as FLT3 and JAG141,42, were not only targets of

a UMAP b Spearman’s correlation Gene mean expression in different cell types

C1 Cor.
17 16 C2
15 C3 1
4 18 C4
14 6 C5 0
7
C13
C8
UMAP_2
11 -1
C7
1 C6
0 9 8
12 C9 Exp.
C10 z-score
10 C18
13 2 C12 4
C11 2
-4 3 C14 0
4
C15 −2
C16
5 −4
C17
C1
C2
C3
C4
C5
C13
C8
C7
C6
C9
C10
C18
C12
C11
C14
C15
C16
C17
AZU1
CTSG
PRTN3
MZB1
MEG3
ARIH1
TOP2A
CCNB1
CENPA
TK1
PCNA
E2F1
CD14
CDA
FCN1
S100A9
TM4SF1
OLFML2A
CD34
FAM30A
TNFRSF4
IL3RA
CD99
CD44
CD9
LTB
MRC1
HLA−DPA1
HLA−DQA1
MYBL2
MCM2
S100A12
G0S2
CLC
HMMR
S100A8
S100A10
ANXA8
SOCS2
CD38
CD200
CD180
MEF2C
CD1C
CD36
MKI67
VIM
ELANE
CENPE
CENPF
MT1X
EPX
-5 0 5 10
UMAP_1
Cell type percentage

Cell Type
c d
C1 GMP-like 1
Stem-like
C17 (0.7%) C10 (0.3%) C2 GMP-like 2 MDP-like
C18 (0.9%) C2 (14.3%) C3 GMP-like 3
C5 (1.0%) C4 GMP-like 4 4 4 Prog-like
C16 (1.2%) C5 GMP-like 5
C4 (1.6%) C6 Cycling GMP-like 1 Cycling
C15 (1.8%) C7 Cycling GMP-like 2 S100hiGMP-like GMP-like
C3 (14.3%) C8 Cycling GMP-like 3
UMAP_2
C13 (2.8%)
UMAP_2
C9 (2.9%) C9 Cycling GMP-like 4 0 0
C8 (2.9%) C10 Cycling GMP-like 5
C7 (5.3%) C11 S100hiGMP-like 1
C12 S100hiGMP-like 2 GMP-like
C6 (7.2%) C11 (13.5%) C13 EOS-like
C12 (7.9%) C14 Prog-like -4 -4
C14 (9.6%) C1 (11.7%) C15 Stem-like
C16 MDP-like
C17 CD1C+ PrecDC-like
C18 CD14+ Promono-like
-5 0 5 10 -5 0 5 10
UMAP_1 UMAP_1
e Branches Pseudotime
g Projected scTarget Branches
h PML/RARα
MDP-like Stem-like 4
e
Expression level
lik
e
4 4 Prog-like
lik
P-
Stem-like Stem-like 3
P-
S100hi
G
M
UMAP_2
UMAP_2
hi
G
GMP-like Cycling- 2
00
hi
00
0 0 GMP-like
S1
S1
1
Branches Pseudotime
Sample GMP-like 0
GM
GM
Stem-like 6 -4 APL (10X) -4
ke
ke
M like
e
lik
lik
lik
APL06-scTarget
P-li
P-li
-li
-li
Prog-like
P-
P-
P-
P-
em
og
APL08-scTarget
D
S100hiGMP-like 2
ke
ke
Pr
St
G
-5 0 5 10 -5 0 5 10
hi
g-
0
00
GMP-like
lin
UMAP_1 UMAP_1
S1
yc
C
f S100hiGMP-like Stem-like GMP-like i scRNA-seq gene exp. of 16 APLs Mean exp.
TF genes PML/RARα targets
ZBTB16 PRDM2 REST SMARCA2/4
ERG NR3C1 MED12 DNMT3B
NFE2L2 ARID2 DPF2 MAPK11
MAX NFATC1
NFKB1 NFATC2 ALDH6A1 NFATC1/2
ETV6 RUNX1
APOC2 MRC2
Marker genes (PML/RARα targets)
KMT2A RUNX2
IRF1 CREB1 NPM1 RPL3
IKZF1 FOS XBP1 ID1
STAT1 SP3
STAT3 JUN BPI SRGN
SMAD4 FOXO3 RETN PLAU
STAT5B CTCF
Branch
NCOA4 LAIR1
REST DNMT1 CXCR4 GADD45A Stem-like
FOS SP100 Prog-like
RPL39 ARID1B S100hiGMP-like
RRBP1 ERN1
GMP-like
GATA2 MEF2C Cycling GMP-like
ELF1 SOX4 HNRNPL RREB1
MDP-like
NR4A1 IRX3 KDM1A KAT7
HHEX KLF9 PVT1 MBNL1 Gene type
WT1 ATF4 CDK13 RPL31/21 Marker genes
CEBPA MAFB
CEBPE KLF10 TOP2A E2F1 scRNA-seq exp.
Scaled row z-score
BCL6 ZNF467 exp. MYBL2 GINS2
SPI RXRA STMN1 PCNA
BATF NFE2 3 -2 0 2
GFI1 JUNB MAFB CD1C
CEBPB JUND Mean exp.
0 CD83 ISG15 row z-score
FLI1 LYL1
STAT6 MBD2 C5AR1 FGR
DDIT3 OLIG1 -3 IRF8 NLRP3 -3 0 3
PML/RARα but also exhibited elevated expression in APL stem-like Third, we performed VIPER analysis to elucidate potential TFs
cells (Fig. 3b). involved in the PML/RARα-induced transcriptional network in APL
Second, we performed pathway enrichment analysis to reveal the stem-like cells (Fig. 3d). By comparing with normal HSPCs, we found
involvement of the PML/RARα-dysregulated signaling pathways in APL that stemness-associated TFs, such as TGIF1 and HIF1A45,46 were acti-
stem-like cells. This analysis identified numerous LSC-associated vated, suggesting their potential roles in regulating the stemness of
pathways, including classical WNT, MAPK, VEGF, P53, and mTOR APL cells. Additionally, we observed the activation of regulators
signalings43,44 (Fig. 3c), all crucial for maintaining the LSC population. associated with histone modification, including MBD2/3 (methylated-
Furthermore, pathway crosstalk analysis based on these LSC- DNA binding proteins) and PRDM2 (H3K9 methyltransferase), in APL
associated PML/RARα target genes indicated that they coordinated stem-like cells. These findings suggested their potential involvement in
the regulation of APL stem-like cells (Supplementary Fig. 10 and Sup- epigenetic control of self-renewal and quiescence of APL stem-like
plementary Data 7). cells47. TFs linked to malignant transformation and stemness

Fig. 2 | Characterization of intratumoral heterogeneity in APL blasts reveals a by colors (upper panel), including stem-like cells (center), S100hiGMP-like cells
complex trajectory with multiple branches and a small subpopulation of APL (left), and GMP-like (right). Characteristic transcription factors (TFs) are listed on
stem-like cells. a UMAP plot of APL blasts (upper panel; n = 126,802 cells). Eighteen the right. g UMAP plots of the targeted scRNA-seq (scTarget) data from two APL
clusters are labeled in different colors and numbers (lower panel). GMP patients, with color coding for sample types (left panel) and branches (right panel).
granulocyte-monocyte progenitors, EOS eosinophils, Prog progenitors, MDP On the right panel, cells detected more than three PML/RARα fusion reads were
monocyte-DC progenitors, PrecDC pre-conventional dendritic cells, Promono illustrated. h The expression levels of PML/RARα in six branches of APL blasts.
promonocytes. b The left heatmap shows Spearman’s correlation between the 18 i Branch-specific expression patterns for PML/RARα targets across the APL trajec-
APL clusters, calculated using the average expression profiles of the clusters. The tory. The left heatmap visualizes the single-cell expression of PML/RARα-regulated
right heatmap illustrates the expression levels of cell type-specific genes in each branch-specific marker genes across branches, with rows representing genes and
cluster. c Visualization of RNA velocity-based cell-state transitions of APL blasts. columns for cells. To offer a clear and representative depiction of the branch-
d UMAP plot of APL blasts with six branches, i.e., stem-like, Prog-like, S100hiGMP- specific expression patterns for PML/RARα targets, we selected 1000 cells from
like, GMP-like, cycling GMP-like, and MDP-like branches. e Pseudotime-ordered each branch for interpretation. The right heatmap displays the mean gene
analysis of four major branches in APL blasts, including stem-like, Prog-like, expression (n = 1758 genes) across branches, accompanied by the annotations of
S100hiGMP-like, and GMP-like branches. f Heatmap showing the dynamic changes representative marker genes on the right side. Cor. correlation, Exp. expression.
in gene expression (n = 116 genes) along the pseudotime. Cell branches are labeled
properties, such as OLIG2 and ARID3B48,49, were also found to be active characteristics of APL patients (Supplementary Data 8). Firstly, a
in APL stem-like cells. These findings highlighted the critical roles of higher proportion of APL stem-like cells was significantly associated
PML/RARα in regulating APL stem-like cells at the single-cell level. with an elevated white blood cell (WBC) count (P < 0.0001) and a lower
Fourth, we proceeded to associate the cellular architecture of APL platelet count (P = 4.0e-3) (Fig. 5a). Notably, the APL stem-like cell type
blasts with the cooperating genetic alterations commonly found in showed the strongest correlation with the WBC count, followed by
APL, including FLT3 (FLT3-ITD and FLT3-TKD), WT1, and NRAS muta- GMP-like and cycling GMP-like cell types (Supplementary Figs. 11, 12
tions. As illustrated in Fig. 3e, FLT3-ITD was significantly associated and Supplementary Data 9). This finding also emphasized the inter-
with a more primitive disease phenotype than other investigated tumoral heterogeneity among APL patients. Further analysis revealed
mutations, suggesting that the presence of FLT3-ITD might play a that a higher proportion of stem-like cells was significantly associated
significant role in enhancing the leukemic stemness. We further used with a higher percentage of APL blasts in BM cells (R = 0.53, P < 0.0001;
two established leukemic stemness scorings34,50, LSC17 and LSC6, to Supplementary Fig. 13a) and a higher blast count in peripheral blood
compare the stemness characteristics between APL stem-like cells with (R = 0.35, P < 0.0001; Supplementary Fig. 13b). This was also notably
and without FLT3-ITD. The analysis revealed that the stem-like cells correlated with an increased WBC (P < 0.0001; Supplementary Fig. 14).
with FLT3-ITD received significantly higher scores than those without These results indicated that APL patients with a higher percentage of
FLT3-ITD (Fig. 3f), supporting the notion that FLT3-ITD might enhance APL stem-like cells in APL blasts might have an increased tendency for
the stemness characteristics. Moreover, we performed FLT3-ITD- blasts to circulate in peripheral blood. Secondly, we examined the
targeted scRNA-seq in two APL patients to verify its existence and, relationship between the percentage of APL stem-like cells and
more importantly, to confirm its higher expression in APL stem-like recurrent mutations in APL patients, including three common isoforms
cells (Fig. 3g, h). of PML/RARα (L-type, S-type, and V-type)54, FLT3 mutations (ITD and
TKD), and mutations involving WT1, NRAS, and ARID1A. Remarkably, a
The predictive power of the APL stemness score in early death higher percentage of APL stem-like cells was significantly associated
and therapy outcome in APL with the S-type PML/RARα transcript (S-type vs. L-type: P < 0.0001;
In this section, we explored how APL stem-like cells defined by scRNA- S-type vs. V-type: P = 7.3e-3; Fig. 5a and Supplementary Fig. 15) and
seq can be utilized to predict the clinical obstacles in APL, more pre- FLT3-ITD (P < 0.0001). In addition, FLT3-ITD was identified as the most
cisely, the occurrence of early death, risk stratification, and therapy significant co-occurrence event, supporting the importance of FLT3-
outcome. We first established deconvolution-based prediction proce- ITD in enhancing the stemness activity of APL stem-like cells (Fig. 3f).
dures, as graphically illustrated in Fig. 4a and detailed in the “Methods” Given the close association of APL stem-like cells with potential
section, and demonstrated their robustness and performance (Fig. 4b). unfavorable prognostic factors (including the high WBC count, S-type
At Step 1, we applied the CIBERSORTx algorithm51 to scRNA-seq data, PML/RARα, and FLT3-ITD), we sought to develop a stemness scoring
generating an APL-specific signature matrix that involved six APL blast system tailored for APL blast cells. Employing the LASSO algorithm, we
cell populations (i.e., stem-like, Prog-like, GMP-like, cycling GMP-like, established an APL-specific stemness score based on the estimated cell
S100hiGMP-like, and MDP-like) and three non-leukemic cell types (i.e., proportions of APL stem-like cells, which was then utilized to quantify
T/NK, B, and erythroid cells). At Step 2, we prepared the transcripts per the stemness of leukemic cells in each patient (detailed in “Methods”
kilobase of the exon model per million mapped reads (TPM) matrix section; Supplementary Fig. 16). We identified eleven genes (SKAP2,
from bulk RNA-seq data. At Step 3, we employed support vector IL1RAP, PLD1, HOPX, TRIM47, MAP2K1, TNFSF4, OLFML2A, P2RY14,
regression (SVR)52 to deconvolute both the signature and TPM matri- NPTX2, and RALA) to construct the APL stemness score, which showed
ces, resulting in a coefficient matrix. The percentage of each cell type a significant correlation with the proportion of APL stem-like cells
in the coefficient matrix was used to build a linear regression model for (Pearson’s correlation = 0.802; P < 0.0001). We then explored the
benchmarking. At Step 4, we constructed an 11-gene scoring model relationship between the APL stemness score and prognosis, including
through LASSO to evaluate the stemness of APL blast cells from bulk overall survival (OS), event-free survival (EFS), and disease-free survival
RNA-seq data, where a higher score indicates a higher stemness. (DFS). Remarkably, a high APL stemness score was significantly asso-
Additionally, we designed a leave-one-out test to demonstrate the ciated with a poorer OS (P = 5.7e-3) and EFS (P = 0.0342), but not DFS
robustness of the inferred APL stem-like cell proportions (median (P = 0.731; see Fig. 5b, with the cutoff optimized using the R ‘maxstat’
R = 0.933, 95% CI = 0.922–0.945, Fig. 4b). In other words, our decon- algorithm). Univariate Cox analysis also revealed that patients with a
volution approach could accurately predict APL stem-like cells from higher APL stemness score had a poor prognosis, as reflected by OS
bulk APL transcriptomes. (P = 3.4e-4; Fig. 5c) and EFS (P = 4.6e-4; Fig. 5d) using the optimized
Next, employing our established deconvolution prediction pro- cutoff. Multivariate Cox analysis confirmed the APL stemness score as
cedures, we examined a large cohort comprising 323 APL patients53 to an independent prognostic factor for OS and EFS (Supplemen-
explore the correlation between APL stem-like cells and the clinical tary Fig. 17).

a Top 500 PML/RARα targets b FCGR2A CD9
Expression Expression Expression
Expression Expression Expression

ranked by adjusted P-values (CUT&Tag) 6 CUT&Tag CUT&Tag
6
0.4 4 IgG 4 IgG
2 2
Gene Gene
0.3 0 0
Enrichment score (ES)
0.2 HCK GFI1

6 CUT&Tag CUT&Tag
0.1 NES = 1.562 4 4
P-value = 0.001 IgG IgG
FDR = 0.001 2 2
0.0 Gene Gene
7.5 0 0
5.0 FLT3 JAG1
2.5 6 CUT&Tag CUT&Tag
0.0 4 4
IgG IgG
−2.5 2 2
−5.0 Gene Gene
0 0
0 5,000 10,000 s Ps s Ps
like PC like PC
Genes ranked by fold change m- HS GM m- HS GM
between APL stem-like and HSPCs ste al ste al
L rm L rm
AP No AP No
c Upregulated PML/RARα targets Downregulated PML/RARα targets d TF activities

in APL stem-like cells in APL stem-like cells Cell types TF activities
row z-score
Pathways in cancer G2−M Checkpoint ARID3B/5B RUNX1/2 AR
MBD2/3 BATF NFATC1 4
Endocytosis
Ribosome BATF3 MEF2A/2D NCOA1/2/3 0
MAPK signaling pathway
Cell cycle NFIX TBP MAFB −4
Chemokine signaling pathway TFE3 GTF2B MAZ
p53 Pathway DNA replication HIF1A PRDM2 KAT7 Cell types
WNT Signaling OLIG2 FOXJ3 STAT1 HSPCs
TGIF1 SP100 NFE2L2 APL stem-like
Fc gamma R−mediated phagocytosis
MAX CIC ERG
PI3K/AKT/mTOR Signaling NFE2 CREB3 IRX3
Lysosome SOX8 TFAP2A TP73
T cell receptor signaling pathway ZBTB7C SMAD3/4 ATF4
Insulin resistance CREB3L3 FOXN3 KLF12
-log10FDR -log10FDR # genes IKZF4 PITX1 HOXA6
VEGF signaling pathway
15 10 HES5 HSF5 ZC3H8
Acute myeloid leukemia 10 12 20 BHLHE40 NANOG MSX1
ErbB signaling pathway 5 9 MYT1 BCL6 RFX5
30
6
Toll−like receptor signaling pathway
e f
FLT3-ITD NRAS mutation LSC17 score (stemness) LSC6 score (stemness)
100 not FLT3-ITD 100 NRAS wt P = 1.9e−25 P = 7e-11
0.50
FLT3-ITD NRAS mut 1.5
0.090 0.29
LSC6 score
0.25
LSC17 score
75 75
Percentage (%)
Percentage (%)
0.146
0.924 1.0
0.11 0.00
0.075
50 50
0.30 0.48 −0.25 0.5
0.039 0.143
25 25 0.22 0.30 −0.50
0.0
TD ) D
-IT e) TD ) D
-IT e)
3-I e T3 -lik 3-I ike T3 -lik
0 0 L T -lik FL stem LT m-l FL stem
n F tem n F ste
No PL s P L No PL L
(A (A (A (AP
FLT3-TKD WT1 mutation g Branches h FLT3-ITD expression (scTarget)

100 not FLT3-TKD 100 WT1 wt
FLT3-TKD WT1 mut MDP-like
Stem-like
0.391 0.332
75 75 4 Prog-like 2.0
Percentage (%)
Percentage (%)
0.952
0.236
Expression level
hi
S100
UMAP_2
0.591 1.5
50 0.162 50 0 GMP-like Cycling-
0.011 GMP-like
0.907 1.0
0.066 0.804 0.040
25 25 0.920 -4 GMP-like
0.5
0 0 0.0
like g-like -like like -like like like g-like -like like -like like -5 0 5 10
like g-like -like P-like -like -like
m- o P P- P P- m- o P P- P P- UMAP_1 m- MP MDP
Ste Pr hi GM GM g GM MD Ste Pr hi GM GM g GM MD Ste Pro hi GMP GM
0 n n gG
S1
0 c l i 0 0 c l i 00 clin
Cy S1 Cy S1 C y
Fig. 3 | The crucial role of PML/RARα and FLT3-ITD in regulating the properties means ± SE. The P-values were calculated using Student’s t-test and labeled in red
of APL stem-like cells. a Gene set enrichment analysis (GSEA) plot of top 500 PML/ when P-values < 0.05. Two-sided P-values were calculated. f Comparison of the
RARα targets. The gene set for GSEA analysis was defined based on the top LSC17 score (left panel) and the LSC6 score (right panel) of the stem-like cells in APL
500 PML/RARα targets according to adjusted P-values derived from CUT&Tag. patients with or without FLT3-ITD. n = 2344 stem-like cells were used for visuali-
Genes were ranked by the fold change between APL stem-like cells and HSPCs at the zation, excluding those with a score of 0 due to the absence of detected gene
mRNA level. NES normalized enrichment score. The P-value was calculated using expression. In the boxplot, a black line within the box marks the median. The
GSEA. A two-sided P-value was calculated. b Violin plots illustrating representative bottom and top of the box are located at the 25th and 75th percentiles, respectively.
genes highly expressed in APL stem-like cells compared with HSPCs. c GO enrich- The bars represent values more than 1.5 times the interquartile range from the
ment analysis showing the KEGG pathways enriched in upregulated (left panel) and border of each box. The P-values were calculated using the Wilcoxon rank-sum test.
downregulated (right panel) PML/RARα targets in APL stem-like cells compared Two-sided P-values were calculated. g Visualization of FLT3 expression through
with HSPCs. d Master regulator analysis to explore activated (red) and repressed projection onto the UMAP of APL blasts using the scTarget data from two patients.
(blue) transcription factors (TFs) in APL stem-like cells compared with HSPCs. Cells detected with more than three FLT3-ITD mutated reads were color-coded
e Comparison of the percentages of each branch with and without indicated according to the different branches. h The expression levels of FLT3-ITD in the six
mutations. n = 16 patients with FLT3-ITD/TKD mutation information and n = 12 branches of APL blasts, analyzed using data from FLT3-ITD-specific targeted scRNA-
patients with NRAS/WT1 mutation information. Error bars in bar plots represent the seq (scTarget).

a 9 Cell Types Stem-like Prog-like S100hiGMP-like GMP-like Cycling GMP-like MDP-like B T/NK Erythroid cells
b
STEP 1: Generation signature matrix from scRNA-seq STEP 3: Deconvolution using
Support Vector Regression (SVR)
0.975
Expression profile of
APL stem-like cell (%) in bulk RNA-seq and scRNA-seq

scRNA-seq on D0 (16 APLs) Signature matrix Coefficient matrix Proportion matrix
Pearson’s R
10,558 expressed genes
Median = 0.933
Pearson’s correlation coefficients between

95% CI = 0.922 - 0.945
12 patients
12 patients
3,388 genes
CIBERSORTx
Scaled Scaled
0.950
exp. exp. Coef
5 2
0.15
0 0
-5 -2 0
scRNA profilings of 9 cell types 9 cell types 9 cell types Percentage of
9 cell types (100%) -log10(P-value)
0.925
+ 6.0
STEP 2: Generation TPM matrix from bulk RNA-seq STEP 4: Calculation of APL stemness score 5.5
312 135 13 2 0 0 0
APL stemness 5.0
Bulk RNA-seq on D0 (12 APLs) TPM matrix LASSO score 0.900 4.5
0.10 0.14 0.18 0.22

Mean−Squared Error
4.0
3,388 genes
12 patients
Gene Scaled
expression exp. LASSO
(TPM matrix) 2 model
0
12 Patients with both -2 0.875
−8 −6 −4 −2 0 2
bulk RNA-seq and scRNA-seq 12 patients Log (λ)
Fig. 4 | Construction of an 11-gene APL stemness score. a Schematic depicting stemness signature genes, which were PML/RARα targets highly expressed in APL
the APL deconvolution approach and the generation of the APL stemness score stem-like cells; and (5) the APL stemness score was calculated by the mean
using the 16 APL scRNA-seq data as the reference. This improved deconvolution expression level of APL stemness signature genes. b Model performance of the
approach is based on the support vector regression (SVR) algorithm: (1) gen- deconvolution approach to predict the APL stem-like cell percentage from bulk
eration of signatures from scRNA-seq populations, including the six APL blasts RNA-seq of the 12 patients. The performance of the model is evaluated by Pear-
branches and T/NK cells, B cells, and erythroid cells; (2) calculating the TPM son’s correlation coefficients between the observed APL stem-like cell percentage
matrix of bulk RNA-seq of the 12 patients with matched scRNA-seq (detailed in from scRNA-seq and the predicted APL stem-like cell percentage from the
“Methods” section); (3) using SVR to calculate the coefficients of each scRNA-seq deconvolution approach. Leave-one-out (LOO) is used to evaluate the robustness
population from bulk RNA-seq and performing linear regression to benchmark of the model. The P-values were calculated using the Pearson’s correlation. Two-
the percentage and the coefficient of each population; (4) generation of APL sided P-values were calculated.
Notably, a higher APL stemness score was significantly associated death than those who achieved complete remission in bulk RNA-seq
with an increased risk of early death (P = 8.3e-3; Fig. 5b). This finding data of these 10 patients, consistent with our findings from the large
was consistent with our scRNA-seq data, revealing that compared to cohort (Supplementary Fig. 20b).
patients without early death, those patients with early death had a First, comparison analysis of scRNA-seq data revealed that diag-
relatively higher proportion of stem-like cells, as well as elevated nostic APL samples were enriched in more primitive cells than post-
expression of stemness-associated genes, such as FCGR2A, IL1RAP, treatment samples, which were relatively enriched in more mature
MAP2K1, and KLF9 (Supplementary Fig. 18). Moreover, further analysis progenitor cells (Fig. 6a). Especially on Day 2 after treatment, the stem-
showed that our APL stemness score was an independent risk factor like cells were almost undetectable in the post-treatment samples
(HR = 5.627; 95% CI, 1.981–15.980; P = 0.001) with a superior predictive (Fig. 6b). More precisely, ATRA treatment significantly increased the
value for early death (Supplementary Fig. 19). These results collectively percentages of the more mature GMP-like cell type and decreased the
underscore the utility of the APL stemness score in assessing APL risk, percentages of three primitive cell types (stem-like, Prog-like, and
including the risk of early death in APL. S100hiGMP-like), with stem-like cells almost undetectable after the
treatment, suggesting that ATRA had a notable impact on primitive
In vivo effect of ATRA on differentiation of primitive APL blasts APL cells, especially the stem-like cells (Fig. 6c, d). To confirm the
and its influence on early death risk ability of ATRA to induce the differentiation of APL primitive blasts
Administrating ATRA as early as possible has been proven essential in towards more mature progenitor cells, we also employed the estab-
reducing the early death rate in APL48,49. We then delved further into lished score to quantify the matureness of leukemic cells14, and
exploring the in vivo impact of ATRA treatment on APL cellular hier- observed a significant increase in this score in the 2-day post-treatment
archies, with a particular focus on the primitive blasts, as their stem- samples compared to samples at diagnosis (Fig. 6e), indicating that
ness might influence treatment response. We performed scRNA-seq on APL blasts became more mature upon ATRA treatment. Similar find-
BM samples collected from three patients (APL03, APL04, and APL05) ings were also obtained by analyzing bulk transcriptomes of 7 APL
after two days of ATRA therapy. At this time point, a notable increase in patients who achieved complete remission before and after ATRA
CD11b expression was observed (Supplementary Fig. 20a). Using the treatment. As illustrated in Fig. 6f, the proportion of APL stem-like cells
pre-defined six branches of APL blasts at diagnosis served as the and the stemness of leukemic cells were notably decreased following
reference, the cell types of APL cells on Day 2 after ATRA treatment ATRA treatment, especially on Day 2.
were determined by employing the KNN algorithm in a merged dataset Next, we looked at the expression changes of hematopoietic
that included cells from both Day 0 and Day 2. We then investigated differentiation-related CD markers and TFs to show that ATRA could
the in vivo effects of ATRA through changes in the abundance of cell induce a stepwise differentiation, starting from APL stem-like cells
groups and differential transcriptional regulation. We further applied towards more mature cells. Notably, we observed a significant down-
the deconvolution method on bulk transcriptomes from RNA-seq regulation of APL stemness CD markers (such as CD200, CD34,
performed on 10 APL patients (including three who suffered from early FCGR2A/CD32, CD9, and IL3RA/CD123) and TFs (such as HHEX, MYC,
death) before and after ATRA treatment to explore the potential JAG1, and ERG) in primitive cell types (Fig. 6g and Supplementary
influence of leukemic stemness on the differential in vivo response to Data 10). Conversely, markers and TFs associated with mature hema-
ATRA, which might be a contributing factor to early death. The topoietic cell lineages were upregulated following ATRA treatment.
stemness scores were indeed higher in patients who experienced early For example, CD38, CD84, CEBPA, and ELF4 were upregulated in the

a 323 de novo APL bulk RNA-seq cohort c Overall survival

100
Clinical features Percentage (%) 75 1.00 +++++
+++++++++++++++++++++++ +++++ +++++++++++++++++++
Percentage of
50 9 APL cell types +++++++ ++++ +++ ++ ++++++
+++++++++ +++++++++++
Survival probability
25 0.75
0
Gender (P=0.947)
Age (P=0.34) 0.50
Cons. strategy (P=0.367) APL stemness score low
Sanz group (P=5.6e-12) APL stemness score high
PML/RARα (P=3.9e-8) 0.25 P-value = 3.4e-4
WBC (P=2.0e-8) Hazard ratio (HR) = 4.003
Platelet (P=4.0e-3) 95% CI of HR = 1.763-9.089
0.00
FLT3-ITD (P=1.3e-10)
0 12 24 36 48 60
FLT3-TKD (P=0.0846)
Time (months)
WT1 (P=0.679)
Number at risk
NRAS (P=0.867)
ARID1A (P=0.334) Low 182 176 176 164 128 96
RREB1 (P=0.188) High 123 106 105 99 74 45
Mutations
ARID1B (P=0.791)
EBF3 (P=0.794)
KRAS (P=0.291)
USP9X (P=0.592)
EP300 (P=0.584)
SF3B1 (P=0.309)
SMARCB1 (P=0.412)
d
Event free survival
CALR (P=0.275)
1.00 ++++
NSD1 (P=0.404) +++++++++++++++++++++++ +++++ ++++++++++++
++++
UPF3B (P=0.864) +++++++ +++
b +++ ++ ++++++
++++ ++++ ++++++++++
Stemness score
Survival probability
0.8 0.75
0.6
Cutoff = 0.2689 APL stemness
0.4 score 0.50
0.2
APL stemness score low
APL stemness score high
OS status (P=5.7e-3) 0.25 P-value = 4.6e-4
EFS status (P=0.0342) Hazard ratio (HR) = 2.956
DFS status (P=0.731) 95% CI of HR = 1.564-5.585
Early death (P=8.3e-3) 0.00
0 12 24 36 48 60
Time (months)
Cell type Gender Consolidation strategy PML/RARα Platelet APL stemness score EFS/DFS status Number at risk
Stem-like Female (Cons. strategy) isoform
S-type ≥ 40×109/L High (≥0.2689) Event
Prog-like Male without ATO < 40×10 9
/L Low (<0.2689) No event Low 182 176 176 164 128 96
hi
S100 GMP-like with ATO L-type
GMP-like V-type High 123 106 105 99 74 45
Cycling GMP-like Age Sanz group WBC group Mutation OS status Early death
MDP-like 18-45 High-risk
B cells 45-60 Intermediate-risk > 10×109/L Mutation Dead Yes
T/NK cells 60-80 Low-risk ≤ 10×10 9
/L Wild-type Alive No
Erythroid cells
Fig. 5 | Informativeness of the APL stemness score in predicting prognosis and Wilcoxon rank-sum test to reveal the correlation between the APL stemness score
early death in APL. a Profiling of the predicted APL cell type percentage, clinical and the prognosis of the 323 APL patients. * PML/RARα isoforms mainly
features, and gene mutations in the 323 APL patients. Columns represent indivi- encompass three typical types: long (L), short (S), and variant (V) types, respec-
dual APL patients, and the P-values were calculated using the Wilcoxon rank-sum tively defined by the breakpoint of PML on intron 6, intron 3, and exon 6. OS
test (statistical tests with two groups) or analysis of variance (ANOVA, with more overall survival, EFS event-free survival, DFS disease-free survival.
than two groups) to illustrate the correlation between the APL stem-like cell c, d Kaplan–Meier estimates of overall survival (OS) (c) and event-free survival
percentage and clinical features/genetic alternations. Two-sided P-values were (EFS) (d) of APL patients (n = 305 patients with prognostic information) in the
calculated. b Examination of the APL stemness score and its relationship to the entire cohort stratified by the APL stemness score. The P-values were calculated
prognosis of the 323 APL patients. The P-values were calculated using the using the log-rank test. Two-sided P-values were calculated.
Prog-like cells; ITGAM/CD11b, C5AR1, TFEC, and MAFB were upregu- death, potentially explaining the significant association we observed
lated in the S100hiGMP-like cells; and CD24, CEACAM6, CEBPE, and between the stemness activity of leukemic cells and early death, as
CEBPB upregulated in the GMP-like cells (Fig. 6h and Supplementary revealed by our deconvolution analysis of transcriptomes from 323
Data 10). APL patients (Fig. 5b).
Of particular interest, we identified distinct transcriptional
responses induced by ATRA in APL patients with and without early Discussion
death. As illustrated in Fig. 6i, the consistent downregulation of several Single-cell RNA sequencing (scRNA-seq) provides a powerful means to
stemness-associated CD markers (CD200 and CD9) and TFs (HHEX and precisely identify complex cell type compositions, especially rare cell
NFATC2) was not observed in APL patients with early death (Supple- types. In this study, we employed scRNA-seq to comprehensively
mentary Data 11). Similarly, the consistent upregulation of decipher APL heterogeneity and draw an overall picture of the APL
differentiation-related CD markers (FUT4 and IL1R1) and TFs (FOS and hierarchy composition at the single-cell level. Our findings illuminated
EGR1) was also not observed in patients with early death (Fig. 6j and the pivotal role of PML/RARα in orchestrating the intratumoral het-
Supplementary Data 11). GO analysis revealed similar results: in addi- erogeneity of APL blasts, with a special emphasis on its influence on the
tion to the induction of differentiation, the repression of stemness- stem-like cells identified in our study. Furthermore, we discovered that
associated pathways, such as the MAPK cascade, was also not observed FLT3-ITD further enhanced the stemness attributes of these cells. By
in patients with early death after ATRA treatment (Supplementary integrating scRNA-seq data with a large cohort of bulk RNA-seq data
Fig. 20c, d). Additionally, a significant decrease of the APL stemness and in vivo ATRA treatment data, we uncovered the complex and
scores after ATRA treatment was observed in patients who achieved multifaceted contributions of APL stem-like cells to early death in APL.
complete remission, but not in patients who experienced early death This included a notable increase in cell proportion, elevated expres-
(Supplementary Fig. 20b). Our findings suggest that ATRA treatment sion of stemness-associated genes, and the persistence of the stem-
had a lesser impact on the stemness program in APL patients with early ness program post-ATRA treatment in patients with early death. More

a b Day 0 (de novo APL) Day 2 (after ATRA treatment) c

ATRA
Day 0 (de novo APL) Stem-like treatment
10 10 B cells 10 B cells Day 0
Day 2 (after ATRA treatment)
Erythroid cells Erythroid cells Day 2
Stem-like Stem-like
5 5 MDP-like 5 MDP-like Prog-like
UMAP_2
UMAP_2
T/NK cells T/NK cells S100hiGMP-like Prog-like
Prog-like Prog-like
GMP-like
S100hiGMP-like hi
S100 GMP-like Cycling GMP-like MDP-like
0 0 0 MDP-like
B cells
GMP-like GMP-like T/NK cells
Erythroid cells S100hi Cycling
−5 −5 Cycling GMP-like −5 Cycling GMP-like GMP-like GMP-like GMP-like
−10 −5 0 5 −10 −5 0 5 −10 −5 0 5
UMAP_1 UMAP_1 UMAP_1
d e f Stem-like cell (%) APL stemness score
Primitive to mature score bulk RNA-seq bulk RNA-seq
Stem-like Prog-like S100hiGMP-like GMP-like
(ANOVA P=4.7e-4) (ANOVA P=2.4e-4)
% of stem-like cells in APL blasts

P = 0.0559 P = 0.0453 P = 0.006 P = 0.117 scRNA-seq (P < 1e-100)
Percentage of APL blasts (%)
10.0 ATRA 0.011 0.0012

1.5 0.3 ATRA
20 75 0.56 0.035
5 0.4 0.0070
Stemness score
0.2 0.0095
7.5 70 4
Score
1.0 15 0.1
ATRA 65 3 ATRA 0.3 ATRA
5.0 0.0
10 2
0.5 ATRA 60 ATRA ATRA
−0.1 +ATO +ATO
1 0.2
2.5 5
ATRA 55 −0.2
0.0 0
Day 0 Day 2 Day 0 Day 2 Day 0 Day 2 Day 0 Day 2 Day 0 Day 2 7) 7) 6) 7) 7) 6)
(n= 2 (n= 5 (n= n= (n= (n =
y0 y y y 0( y 2 ay 5
Da Da Da Da Da D
g Stemness CD markers Stemness TFs
h Differentiation markers Hematopoietic TFs
−0.010 CD200 −0.183 RUNX2 0.507 CD38 0.216 CEBPA
−0.006 CLEC10A −0.044 NFATC2 0.094 CD84 0.026 ELF4
−0.007 CD34 −0.149 WT1 0.005 CEACAM1 0.073 IRF1
−0.227 IFITM1 −0.379 HHEX 0.361 LAIR1 0.054 ARID3A
−0.008 TNFSF14 −0.105 KLF9 0.123 ANPEP/CD13 0.436 SPI1
−0.019 FCGR2A/CD32 −0.082 JAG1 0.084 ICAM1 0.029 TFEC
−0.274 CD96 −0.175 KLF10 0.052 ITGAM/CD11b 0.012 MAFB
−0.370 FLT3 −0.044 DLL1 0.012 C5AR1 0.227 ZBTB7A
−0.111 IL3RA/CD123 −0.503 BCL11A 0.106 ITGAX/CD11c 0.006 RARG
−0.184 TNFRSF18 −0.158 MYC 0.085 CD300LF 0.482 CEBPE
−0.601 CD9 −0.056 ERG 0.006 CEACAM6 0.609 CEBPB
−0.477 TNFRSF4 −0.615 SOX4 0.001 LRP1 0.181 BHLHE40
−1.086 CD99 −0.153 RUNX1 0.013 FCER2 0.336 EGR1
−1.578 CD44 −0.132 IRX3 0.027 CD24 0.220 FOS
FC NA FC NA FC NA FC NA
log 2 e e e e log 2 like like like like log 2 e e e e log 2 e e e e

m -lik g-lik P-lik P-lik m- rog- i MP- MP- m -lik g-lik P-lik P-lik m -lik g-lik P-lik P-lik
o o o i M
Ste Pr 0hi GM GM Ste P h G G Ste Pr 0hi GM GM Ste P r h G GM
0 0 0 0 0 0
S1 S1 S1 S1
log2FC (Day 2 vs. Day 0) ATRA Day 0 Mean expression

scRNA-seq -2 0 2 treatment Day 2
(z-score) -2 0 2
i Stemness CD markers Stemness TFs j Differentiation markers Hematopoietic TFs
CD200 ERG ITGAE CEBPA

IFITM1 IRX3 FUT4 CEBPE
TNFRSF18 IRX5 CEACAM1 FOS
TNFSF14 KLF9 CXCR5 TRIB1
TNFSF10 HHEX C5AR1 KLF5
CD9 MAP2K1 IL1R1 ZNF454
IL3RA NFATC2 IL10RB EGR1
3 4 0 1 5 8 9 1 7 2 3 4 0 1 5 8 9 1 7 2 3 4 0 1 5 8 9 1 7 2 3 4 0 1 5 8 9 1 7 2
L0 L0 L2 L2 L0 L1 L1 L1 L1 L2 L0 L0 L2 L2 L0 L1 L1 L1 L1 L2 L0 L0 L2 L2 L0 L1 L1 L1 L1 L2 L0 L0 L2 L2 L0 L1 L1 L1 L1 L2
AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP AP
log2FC (Day 2 vs. Day 0) Clinical outcomes Alive Early death

bulk RNA-seq -2 0 2
Fig. 6 | In vivo analysis revealing direct targeting of APL primitive blasts in (e and f), a black line marks the median, and the bottom and top of the box are
(including stem-like cells) by ATRA and induction of their differentiation and located at the 25th and 75th percentiles, respectively. The bars represent values more
maturation. a UMAP plot showing the integration of APL BM cells (n = 39,415 cells) than 1.5 times the interquartile range from the border of each box. The P-values
collected on Day 0 (red) and Day 2 after ATRA treatment (blue). b UMAP plots were calculated using the Wilcoxon rank-sum test. Two-sided P-values were cal-
showing all defined cell populations of APL BM cells (n = 39,415 cells) collected on culated. g Heatmap showing the normalized expression of stemness CD markers
Day 0 and Day 2 after ATRA treatment. c Comparison of the percentages of six and TFs in APL stem-like cells, Prog-like, S100hiGMP-like, and GMP-like clusters on
branches before (red) and after two days of ATRA treatment (blue). d Comparison Day 0 (red) and Day 2 after ATRA therapy (blue). h Heatmap showing the nor-
of the percentages of the stem-like, Prog-like, S100hiGMP-like, and GMP-like cells on malized expression of differentiation markers and hematopoietic TFs in APL stem-
Day 0 (red, n = 3 samples) and Day 2 after ATRA treatment (blue, n = 3 samples). The like cells, Prog-like, S100hiGMP-like, and GMP-like clusters on Day 0 (red) and Day 2
P-values were calculated using Student’s t-test. A one-sided P-value was calculated. (blue) after ATRA therapy. i Heatmap illustrating the log2(fold changes) (log2FC) of
In the boxplot, a black line within the box marks the median. The bottom and top of gene expression levels for stemness-associated CD markers and TFs between APL
the box are located at the 25th and 75th percentiles, respectively. e Comparison of samples on Day 0 and Day 2 (Day 2 vs. Day 0) in different APL patients. j Heatmap
the Primitive-to-Mature scores of APL blasts on Day 0 (red, n = 13,966 cells) and Day illustrating the log2FC of gene expression levels for differentiation markers and
2 after ATRA treatment (blue, n = 10,852 cells). f Comparison of APL stem-like cell hematopoietic TFs between APL samples on Day 0 and Day 2 (Day 2 vs. Day 0) in
percentages (left panel) and stemness scores (right panel) on Day 0 (in red, different APL patients. Gray represents patients alive after induction therapy, and
n = 7 samples), Day 2 (in blue, n = 7 samples treated with ATRA alone), and Day 5 (in black for patients with early death.
orange, n = 5 samples treated with ATRA + ATO). Notably, within the boxplot shown

importantly, we developed an APL-specific stemness score that proved as CD99 and CD9, have been reported to play a role in leukocyte
to be a robust independent risk factor with superior predictive power migration69,70, potentially contributing to extramedullary infiltration.
for poor prognosis and early death, surpassing traditional risk factors, Further investigation is required to comprehensively understand these
such as WBC, FLT3-ITD, and the Sanz score. underlying mechanisms. We also acknowledge that early death is a
Our results provide valuable single-cell insights into the identi- complex event influenced by numerous factors beyond the stemness
fication of leukemic stem cells in APL. Our scRNA-seq analysis score, such as the tumor microenvironment, cooperating mutations,
revealed a small portion of stem-like leukemic cells (C15) directly epigenetic modifications, and various clinical variables such as the
from APL patients, expressing stemness markers (CD200, CD99, and patient performance status (ECOG-PS) and the treatment protocol.
CD9) and PML/RARα, and sitting at the top of the differentiation Our preliminary comparative analysis, using cohort data, suggested
trajectories in the APL hierarchy. APL LSCs have not been well that the IL8-related pathway and abnormal metabolic processes might
defined, partly due to a lower density of the CD34 expression on APL also play a role in influencing the survival outcome of patients with
cells compared to other forms of AML55. Interestingly, previous stu- higher stemness scores (Supplementary Fig. 21). Our results high-
dies have reported variable CD34 expression in APL, with frequencies lighted the leukemic stemness as one of the critical factors for early
ranging from 20% to 31%56–58 and occasionally as high as 43% when death in APL, which shed deeper insights into the complicated
considering a low cutoff level59. In our study, we indeed observed mechanisms underlying early death in APL and provide promising
CD34 expression in a subset of APL stem-like cells. Detailed inspec- targets for mitigating the risk of early death.
tion showed that these CD34-positive cells within APL stem-like cells Our study has several limitations that warrant consideration. First,
predominantly originated from three patients who were CD34- given the complex effects of ATRA on primitive stem cells71, further
positive based on their immunotyping results. This observation investigations to validate and explore the impact of ATRA on APL stem-
aligns with the known variability of CD34 expression in APL patients. like cells we identified are required. Second, expanding the sample
Notably, LSCs can be found in both CD34-positive and CD34-negative size, particularly for patients experiencing early death, is essential to
populations60. In NPM1-mutated AML (commonly CD34-negative), robustly confirm our findings regarding the prognostic significance of
LSCs were found in both CD34+ and CD34− cells, suggesting that leukemic stemness in early death.
their presence is not limited to CD34+ progenitor cells60,61. Further- In conclusion, our work provides valuable single-cell insights into
more, CD200 was identified as a novel LSC marker and highly APL leukemogenesis by comprehensively elucidating intratumoral
expressed in both CD34+ and CD34– LSCs, including those with heterogeneity, the cellular composition hierarchy in APL, and their
mutant NPM130, further validating our identification of leukemic differential responses to ATRA therapy in vivo, which may contribute
stem-like cells in APL for its characteristic expression of CD200. to early death in APL patients. We have characterized a small sub-
Furthermore, APL stem-like cells co-expressed lymphoid-related population of APL stem-like cells at the single-cell level, closely cor-
genes, such as T-lineage-affiliated glycoprotein CD2, also supporting related with FLT3-ITD and poor prognosis, laying the foundation for
its more primitive status33. further exploration of cellular therapeutic targeting strategies. Future
Based on the discovery and characterization of APL stem-like studies will be geared towards elucidating the biological mechanisms
cells, we introduced the APL stemness score, which has demonstrated of APL stem-like cells to extend the success of APL-targeted therapy to
the ability to predict the prognosis and risk of APL, notably with other cancers.
regards to early death in APL patients. The concept of leukemic
stemness as a clinical predictor has gained growing attention in the Methods
field of AML. Various measures have been developed to evaluate leu- Patient samples
kemic stemness in non-APL AML, including metrics like the Bone marrow (BM) samples were collected according to the Declara-
LSC17 score50 and the frequency of leukemic progenitor cells. How- tion of Helsinki at the initial diagnosis of APL. Patients were randomly
ever, applying these measures to APL has been challenging due to the recruited in the clinical setting and we also confirmed that the patients
limited understanding of APL stem cells. Our study marks the endeavor in each analysis were representative of the broader APL patient
to characterize APL stem-like cells and establish a measure based on demographic. The selection/recruitment was also contingent upon the
APL leukemic stemness. In comparison to other commonly used pre- availability of high-quality samples, which is a prerequisite for the
dictors, such as high WBC counts and FLT3-ITD status, the APL stem- reliable scRNA-seq analysis that our study relies on. Written informed
ness score exhibited superior performance in predicting poor consent was obtained from the patient allowing for the publication of
prognosis and early death. This superiority can be partially attributed clinical information, and ethical approval was obtained from the Ethics
to the central role of leukemic stemness in APL, as it significantly Committees of Ruijin Hospital, Shanghai Jiao Tong University School
contributes to the development of the disease. Incorporating our of Medicine (2021/154).
stemness score with the examination of WBC levels and aggressive
mutations offers a valuable addition to the current APL risk assess- Single-cell sequencing alignment and data preprocessing
ment. While the detection of WBC levels helps clinicians evaluate risk Single-cell RNA sequencing data (10X Genomics) from 23 healthy BM
promptly, our stemness score may serve as a reliable measure for more samples were obtained from the Gene Expression Omnibus (GEO)
intensive surveillance, which is both practical and of clinical relevance. public database with accession IDs of GSE12022172 and GSE13011673.
Early death remains a formidable obstacle to achieving favorable Raw sequencing data of healthy BM cells were downloaded and con-
outcomes in APL patients62, and considerable research efforts have verted into FASTQ format using SRA-toolkit (version 2.11.0). Our
been made to investigate its underlying mechanisms53,63–65. Our data scRNA-seq data for the 16 APL BM samples were generated through the
highlight the multifaceted impacts of stem-like cells on early death in separate scRNA-seq experiments and were integrated to explore cel-
APL, encompassing a relatively higher proportion of stem-like cells lular heterogeneity in APL cells. To obtain gene expression matrices
(“quantity” aspect), elevated expression of stemness-associated genes, from both healthy and APL BM cells, we used cellranger (10X Geno-
and persistence of the stemness program (“quality” aspect) post-ATRA mics, default settings, version 6.0.2) to align the scRNA-seq data to the
program in patients who experienced early death compared to those human GRCh38 reference (2020-A version). Both the cellranger soft-
who achieved complete remission. Several studies have linked stem- ware and the human reference were downloaded from the 10X
ness genes, such as CD2 and CD9, with an increased risk of thrombosis Genomics website (https://www.10xgenomics.com). The gene
and coagulopathy66–68, the common causes of early death in APL. expression matrices were then imported into the Seurat R package26.
Moreover, many stemness genes encoding adhesion molecules, such For quality control, cells expressing fewer than 300 genes, having

unique counts exceeding 30,000 or falling below 500, expressing Constructing single-cell trajectories of APL blast populations
mitochondrial RNA exceeding 10%, or identified as doublets using The APL blasts were subjected to downstream analysis using Seurat.
DropletUtils (https://bioconductor.org/packages/DropletUtils/) were To ensure robust results and account for potential batch effects
excluded. As a result, a total of 102,792 cells from healthy BM cells and across samples, we applied the Harmony algorithm for batch corre-
a total of 136,497 from APL BM cells were retained for downstream lation. Subsequently, UMAP was employed to visually demonstrate
data analysis. the effectiveness of our batch correction (Supplementary Fig. 23).
Additionally, the results in Supplementary Fig. 23b, c demonstrated
Annotation of cell populations of healthy BM cells that factors such as age and gender did not affect our identified
We used Seurat to perform highly variable gene identification, clusters of APL blasts. To estimate the potential dynamic process of
dimensionality reduction, graph-based clustering, and differentially cell differentiation in APL blasts, we performed RNA velocity analysis.
expressed genes (DEGs) calculation. For identifying cell populations Velocyto was used to run the RNA velocity analysis and generate the
of healthy BM cells, the top 3000 highly variable genes were selected spliced, unspliced, and ambiguous fractions of each cell. A
using the “FindVariableFeatures” function and enrolled for principal likelihood-based dynamical model was utilized to learn the tran-
component analysis (PCA). The R package Harmony was used for scriptional states and infer intratumoral differentiation of APL blasts
sample-wide batch effect adjustment. Uniform Manifold Approx- using scVelo75. To gain additional insights into genes during differ-
imation and Projection (UMAP) and t-distributed Stochastic Neigh- entiation, the pseudotime transitional trajectory of the four APL blast
bourhood Embedding (tSNE) were performed for dimensionality cell populations was utilized using the R package Monocle2 (version
reduction using cell embeddings generated by Harmony. UMAP was 2.24.0)76. To eliminate the effects of cell cycle blasts and better
chosen for visualization to better interpret the hematopoietic line- interpret intratumoral heterogeneity, cells that were APL stem-like,
age. We first performed cell annotation on normal BM cells. Unsu- Prog-like, S100hiGMP-like and GMP-like were enrolled to construct
pervised clustering was performed on the 102,792 cells from 23 the trajectory. To reduce computation time, the cell number of each
normal BM samples using the default parameters of the “FindClus- cell type was down-sampled to 2,000. The DDRTree method was
ters” function. After unsupervised clustering, 26 clusters were clas- used to perform dimensionality reduction based on the top 30 DEGs
sified. Marker genes for each cluster were calculated using the in each cell type. Based on the trajectory result, APL stem-like cells
“FindAllMarkers” function and defined under the following criteria: were defined as the initiating point of the trajectory. Differentially
log2(fold changes) (log2FC) > 0.58 (FC > 1.5), min.pct >0.1, and expressed TFs in S100hiGMP-like and GMP-like branches were calcu-
adjusted P < 0.05. Cells were annotated using both machine-learning- lated using the BEAM subprogram in Monocle2. Q-value < 0.05 was
based software SingleR and high expression of canonical hemato- used to filter TFs. The visualization was performed via the plot_gen-
poietic markers (i.e., CD34 for hematopoietic stem/progenitor cells es_branched_heatmap function.
(HSPCs), CD14 for monocytes, CD1C for dendritic cells, CD3 for
T cells, CD79A for B cells, and CA1 for erythroid cells) in each cluster. Targeted single-cell RNA sequencing data analysis
Finally, a total of seven major cell populations (HSPCs, GMPs, To validate the expression level of PML/RARα and FLT3-ITD tran-
monocytes, DCs, B cells, T/NK cells, and erythroid cells) from normal scripts, two samples (APL06 and APL08) were performed for targeted
BM cells were identified. single-cell RNA sequencing (scTarget; Singleron Biotechnologies) and
bulk RNA-seq. For gene expression analysis and cell type annotation of
Characterizing malignant APL blasts through integration with scTarget, gene expression matrices of cells were generated using
healthy BM cells and batch effect correction using Harmony CeleScope (https://github.com/singleron-RD/CeleScope) with default
To characterize malignant APL blasts, we performed the integration of parameters. We used the K-nearest neighbor (KNN)23 algorithm to
the scRNA-seq data from both APL and normal BM cells. We first predict cell types, and the calculation steps were described as follows:
identified the top 3,000 highly variable genes across APL and healthy (i) build a single-cell-based expression reference using the pre-
BM cells, and then adjusted batch effects using Harmony. UMAP was annotated 16 APL blasts (10X Genomics); (ii) integrate the expression
performed to visualize the cell distribution of healthy and APL BM cells profiles of the 16 APL blasts and the two scTarget samples. Batch
before (Supplementary Fig. 22) and after batch effect adjustment effects adjustment was performed across samples using Harmony; (iii)
(Fig. 1b). The classification of APL blasts was determined based on the for each cell from scTarget, the top 50 nearest cells were calculated
following criterion: cells were grouped into a distinct cluster that was using the KNN algorithm by the R package BiocNeighbors; (iv) the
clearly separated from cell populations of healthy BM cells, as annotations were based on the highest frequency of the annotated
observed in the UMAP plot. This classification was further validated nearest cell types; (v) cells from scTarget were projected onto the
using the following methods: (i) targeted scRNA-seq analysis, (ii) UMAP of APL blasts for visualization. Next, for the identification of
examination of the expression of canonical APL markers (such as PML/RARα and FLT3-ITD transcripts, the calculation steps were
clinical immunophenotype markers like MPO) and other genes like described as follows: (i) build a reference that contained wild-type
azurophilic granule genes (AZU1, ELANE, and CTSG); and (iii) annota- gene regions of PML, RARA, and FLT3, and the different isoforms of
tion as progenitor cells using SingleR. PML/RARα fusion transcripts; (ii) FLT3-ITD were called from bulk RNA-
seq using RNAmut; (iii) preprocessing steps of scTarget data to dis-
Differentially expressed genes analysis and transcription factors cover fusions/mutations were performed using CeleScope; (iv) a 12-bp
activities estimation of scRNA-seq of nucleotides across the PML/RARα fusion point was used to scan the
Differentially expressed genes (DEGs) between different cell popu- reads that mapped onto PML and RARA genes and identify the
lations were calculated using the “FindMarkers” function in Seurat expression level of PML/RARα transcripts of each cell; (v) detection of
with logfc.threshold set to 0.01 and min.pct set to 0.01. Significance FLT3-ITD according to the mutated nucleotides obtained from bulk
level of DEGs was set to log2FC > 0. 26 (FC > 1.2) and adjusted RNA-seq; (vi) merge the expression level of PML/RARα and FLT3-ITD
P < 0.05. Two methods, VIPER28 and DoRothEA74, were used to esti- transcripts with gene expression data. There were, however, certain
mate the protein activities of transcription factors (TFs). The msviper limitations in scTarget. For example, the presence of dropouts, a well-
function from VIPER package was used for master regulator infer- known challenge in single-cell analysis77, implies that only a fraction of
ence analysis and visualization. P < 0.05 was set as the significance the transcriptome in each cell may have been captured. Therefore,
level to demonstrate the difference of TF activities between different there is room for improving the sensitivity and accuracy of scTarget in
cell types. detecting targeted gene expression at the single-cell level.

Bulk RNA sequencing, alignment, and data analysis Functional enrichment analysis and pathway crosstalk analysis
Raw FASTQ files of bulk RNA sequencing (RNA-seq) data were aligned Gene ontology (GO) enrichment and gene set enrichment analysis
to the human reference genome GRCh38 (release 38). The human (GSEA) were performed using the clusterProfiler package. Gene sets
reference genome and its annotation file were downloaded from the used in GSEA were downloaded from MSigDB. We used single-sample
GENCODE database (https://www.gencodegenes.org/). Salmon (ver- GSEA (ssGSEA) to evaluate the expression level of pathways in each
sion 1.5.1) was used to generate the count and transcripts per kilobase sample through the GSVA package. The public gene sets used in
of the exon model per million mapped reads (TPM) matrix. For cal- GSEA were downloaded from MSigDB, including Hallmark gene sets
culating differentially expressed genes, the limma package was used (H) and KEGG gene sets (C2). To perform GSEA analysis with PML/
with significance level settings of adjusted P < 0.05 and log2FC > 0.58 RARα targets generated from CUT&Tag (GSE195776)78, we initiated
or < −0.58. The R package pheatmap (https://CRAN.R-project.org/ NES evaluation by randomly selecting 500 PML/RARα targets with
package=pheatmap) was used to perform and visualize hierarchical different random seeds spanning from 1 to 100 (Supplementary
clustering in the heatmap plot. Fig. 24). Genes included in the GSEA analysis were ranked based on
their log2FC between APL stem-like cells and normal HSPCs. The
Deconvolution analysis and construction of the LASSO ultimate gene set of PML/RARα targets was defined according to the
score model top 500 PML/RARα targets, determined by adjusted P-values
Deconvolution analysis was performed to estimate APL cell popula- derived from CUT&Tag. We employed OpenXGR79 to identify path-
tions from bulk RNA-seq data. The reference was constructed from way crosstalk through integrative analysis of differential expression
scRNA-seq of the 16 de novo APLs. 10,000 cells were randomly selec- data (PML/RARα targets expressed in APL stem-like cells compared
ted from the total 136,497 cells of APL BM cells. The normalized with HSPCs) and KEGG-derived pathway interaction data. Enrichr80
expression profile was uploaded to CIBERSORTx (https://cibersortx. was also conducted for functional enrichment analysis of differen-
stanford.edu/) and the generation of the signature matrix of the six tially expressed genes. The differentially expressed pathways depic-
APL branches (stem-like, Prog-like, S100hiGMP-like, GMP-like, cycling ted were the most relevant to hematological malignancies or
GMP-like, and MDP-like) and three non-APL populations (B cells, T/NK stemness.
cells, and erythroid cells) was performed. Then the signature matrix
containing 3388 genes was generated and enrolled for deconvolution Statistics and reproducibility
analysis. Support Vector Regression (SVR) was performed to estimate Statistical analysis was performed using R version 4.2.1. The detailed
the coefficients of each cell type from the bulk TPM matrix using R descriptions of the statistical tests used were provided in the
package e1071 (https://CRAN.R-project.org/package=e1071). The legends of the corresponding figures. APL patient samples were
coefficients could be used to reflect the abundance of cell types in bulk randomly recruited in the clinical setting without sample size cal-
RNA-seq. In this study, we had data from 12 patients who had both culated. All analyses are reproducible using codes available through
scRNA-seq and matched bulk RNA-seq data available. To estimate the the project-dedicated website ‘TACH’ (http://www.genetictargets.
cell proportions, we employed the linear regression model to link the com/tach).
coefficients calculated from the bulk RNA-seq and the observed cell
proportions from scRNA-seq. For evaluation, leave-one-out (LOO) Reporting summary
cross-validation was performed using the 12 patients as the “ground Further information on research design is available in the Nature
truth” to assess the accuracy and reliability of the model. The model Portfolio Reporting Summary linked to this article.
was then applied to the 323 de novo APL cohort (GSE172057)53 to
estimate cell proportions. For better application on bulk RNA-seq, we Data availability
constructed a least absolute shrinkage and selection operator (LASSO) The raw sequencing data reported in this paper have been deposited in
based score model to reveal the stemness score of APL. This model the Genome Sequence Archive in National Genomics Data Center,
considered two factors: (i) the proportion of APL stem-like cells by China National Center for Bioinformation/Beijing Institute of Geno-
using genes that were exclusively expressed in APL stem-like cells mics, Chinese Academy of Sciences (https://ngdc.cncb.ac.cn/gsa-
relative to other APL blast populations (quantity-driven); and (ii) the human). These data are accessible under the accession number ‘GSA-
characteristics of APL stem-like cells by analyzing differentially Human: HRA003777’. These data are under controlled access by
expressed genes (DEGs) between APL stem-like cells and normal human privacy regulations and are only available for research pur-
HSPCs (quality-driven). From this analysis, 898 significant DEGs were poses. Access to the data can be granted following approval from the
selected for use in the LASSO regression model. The optimal value of Data Access Committee of the GSA-human database, as detailed at
the penalty parameter λ was determined through 10-fold cross-vali- https://ngdc.cncb.ac.cn/gsa-human/document/GSA-Human_Request_
dation, using the R package glmnet (https://CRAN.R-project.org/ Guide_for_Users_us.pdf. Data are accessible to researchers who meet
package=glmnet). The top 10% estimated proportions of APL stem- the criteria for access as defined by the GSA-human database guide-
like were used to separate the samples. Finally, eleven genes were lines. Access requests are usually processed within approximately four
selected for constructing the APL stemness score model. weeks and data will be available for three months once access is
granted. All sequencing data, including scRNA-seq and bulk RNA-seq
Analysis of APL BM cells after ATRA treatment data, are also available in NODE under the accession number
For scRNA-seq analysis, cells from Day 0 and Day 2 (after ATRA treat- OEP003829. The public datasets utilized in this study are available in
ment) were integrated using Seurat. Batch effects were adjusted using the GEO database. These include scRNA-seq data from 23 healthy BM
Harmony. In the realm of cell type annotation, the cell types identified samples (accession code GSE120221 and GSE130116), the APL cohort
in Day 0 samples were consistent with annotations from the original data (GSE172057), and the PML/RARα target data generated from
analysis of 16 de novo APL scRNA-seq datasets. Subsequently, for Day CUT&Tag (GSE195776). Accession links are as follows: GSE12022172,
2 samples, we annotated cell types based on the KNN algorithm, GSE13011673, GSE17205753, and GSE19577678. Source data are provided
considering their top 50 nearest cells in Day 0. For bulk RNA-seq as a Source Data file within the paper, also accessible through the
analysis, we leveraged deconvolution analysis to estimate the pro- project-dedicated website ‘TACH’ (http://www.genetictargets.com/
portions of the 9 APL cell types, providing insights into the cellular tach). The remaining data are available within the Article, Supple-
composition within the samples. mentary Information, or Source Data file.

Code availability 19. Guibal, F. C. et al. Identification of a myeloid committed progenitor

The single-cell RNA data were processed using cellranger (https:// as the cancer-initiating cell in acute promyelocytic leukemia. Blood
www.10xgenomics.com, version 6.0.2) and analyzed using the R 114, 5415–5425 (2009).
package Seurat (https://satijalab.org/seurat, version 4.3.0). The codes 20. Wojiski, S. et al. PML-RARalpha initiates leukemia by conferring
used in the paper are available on GitHub (https://nrctm-bioinfo. properties of self-renewal to committed promyelocytic progeni-
github.io/APL_stemness) and Zenodo (https://doi.org/10.5281/zenodo. tors. Leukemia 23, 1462–1471 (2009).
10437695)81. 21. Reinisch, A. et al. A humanized bone marrow ossicle xeno-
transplantation model enables improved engraftment of healthy
References and leukemic human hematopoietic cells. Nat. Med. 22,
1. Giustacchini, A. et al. Single-cell transcriptomics uncovers distinct 812–821 (2016).
molecular signatures of stem cells in chronic myeloid leukemia. 22. Edwards, R. H. et al. Evidence for early hematopoietic progenitor
Nat. Med. 23, 692–702 (2017). cell involvement in acute promyelocytic leukemia. Am. J. Clin.
2. McGranahan, N. & Swanton, C. Biological and therapeutic impact of Pathol. 112, 819–827 (1999).
intratumor heterogeneity in cancer evolution. Cancer Cell 27, 23. van Galen, P. et al. Single-cell RNA-seq reveals AML hierarchies
15–26 (2015). relevant to disease progression and immunity. Cell 176, 1265–1281.
3. de The, H., Pandolfi, P. P. & Chen, Z. Acute promyelocytic leukemia: e1224 (2019).
a paradigm for oncoprotein-targeted cure. Cancer Cell 32, 24. Wu, J. et al. A single-cell survey of cellular hierarchy in acute
552–560 (2017). myeloid leukemia. J. Hematol. Oncol. 13, 128 (2020).
4. Brown, D. et al. A PMLRARalpha transgene initiates murine acute 25. Tran, V. T. et al. The diagnostic power of CD117, CD13, CD56, CD64,
promyelocytic leukemia. Proc. Natl Acad. Sci. USA 94, and MPO in rapid screening acute promyelocytic leukemia. BMC
2551–2556 (1997). Res. Notes 13, 394 (2020).
5. He, L. Z. et al. Acute leukemia with promyelocytic features in PML/ 26. Butler, A., Hoffman, P., Smibert, P., Papalexi, E. & Satija, R. Inte-
RARalpha transgenic mice. Proc. Natl Acad. Sci. USA 94, grating single-cell transcriptomic data across different conditions,
5302–5307 (1997). technologies, and species. Nat. Biotechnol. 36, 411–420 (2018).
6. Chen, L. et al. Arsenic trioxide replacing or reducing chemother- 27. Yang, X. W. et al. Coordinated regulation of the immunoproteasome
apy in consolidation therapy for acute promyelocytic leukemia subunits by PML/RARalpha and PU.1 in acute promyelocytic leu-
(APL2012 trial). Proc. Natl Acad. Sci. USA 118, e2020382118 kemia. Oncogene 33, 2700–2708 (2014).
(2021). 28. Alvarez, M. J. et al. Functional characterization of somatic mutations
7. Wang, K. et al. PML/RARalpha targets promoter regions containing in cancer using network-based inference of protein activity. Nat.
PU.1 consensus and RARE half sites in acute promyelocytic leuke- Genet. 48, 838–847 (2016).
mia. Cancer Cell 17, 186–197 (2010). 29. Sagar et al. Deciphering the regulatory landscape of fetal and adult
8. Tan, Y. et al. A PML/RARalpha direct target atlas redefines tran- gammadelta T-cell development at single-cell resolution. EMBO J.
scriptional deregulation in acute promyelocytic leukemia. Blood 39, e104159 (2020).
137, 1503–1516 (2021). 30. Ho, J. M. et al. CD200 expression marks leukemia stem cells in
9. Song, H. et al. Recurrent noncoding somatic and germline WT1 human AML. Blood Adv. 4, 5402–5413 (2020).
variants converge to disrupt MYB binding in acute promyelocytic 31. Jin, L., Hope, K. J., Zhai, Q., Smadja-Joffe, F. & Dick, J. E. Targeting of
leukemia. Blood 140, 1132–1144 (2022). CD44 eradicates human acute myeloid leukemic stem cells. Nat.
10. Zhang, X. W. et al. Arsenic trioxide controls the fate of the PML- Med. 12, 1167–1174 (2006).
RARalpha oncoprotein by directly binding PML. Science 328, 32. Chung, S. S. et al. CD99 is a therapeutic target on disease stem cells
240–243 (2010). in myeloid malignancies. Sci. Transl. Med. 9, eaaj2025 (2017).
11. de The, H. & Chen, Z. Acute promyelocytic leukaemia: novel 33. Grimwade, D. et al. The T-lineage-affiliated CD2 gene lies within an
insights into the mechanisms of cure. Nat. Rev. Cancer 10, open chromatin environment in acute promyelocytic leukemia
775–783 (2010). cells. Cancer Res. 62, 4730–4735 (2002).
12. Vitaliano-Prunier, A. et al. Clearance of PML/RARA-bound pro- 34. Elsayed, A. H. et al. A six-gene leukemic stem cell score identifies
moters suffice to initiate APL differentiation. Blood 124, high risk pediatric acute myeloid leukemia. Leukemia 34,
3772–3780 (2014). 735–745 (2020).
13. Dos Santos, G. A., Kats, L. & Pandolfi, P. P. Synergy against PML- 35. Liu, J. et al. SOX4 maintains the stemness of cancer cells via tran-
RARa: targeting transcription, proteolysis, differentiation, and self- scriptionally enhancing HDAC1 revealed by comparative pro-
renewal in acute promyelocytic leukemia. J. Exp. Med. 210, teomics study. Cell Biosci. 11, 23 (2021).
2793–2802 (2013). 36. Yoshida, G. J. Emerging roles of Myc in stem cell biology and novel
14. Zeng, A. G. X. et al. A cellular hierarchy framework for under- tumor therapies. J. Exp. Clin. Cancer Res. 37, 173 (2018).
standing heterogeneity and predicting drug response in acute 37. Yanez, A. et al. Granulocyte-monocyte progenitors and monocyte-
myeloid leukemia. Nat. Med. 28, 1212–1223 (2022). dendritic cell progenitors independently produce functionally dis-
15. Stewart, C. A. et al. Single-cell analyses reveal increased intratu- tinct monocytes. Immunity 47, 890–902.e894 (2017).
moral heterogeneity after the onset of therapy resistance in small- 38. Liu, Y. et al. CD9, a potential leukemia stem cell marker, regulates
cell lung cancer. Nat. Cancer 1, 423–436 (2020). drug resistance and leukemia development in acute myeloid leu-
16. Saito, Y. et al. Identification of therapeutic targets for quiescent, kemia. Stem Cell Res. Ther. 12, 86 (2021).
chemotherapy-resistant human leukemia stem cells. Sci. Transl. 39. Bonardi, F. et al. A proteomics and transcriptomics approach to
Med. 2, 17ra19 (2010). identify leukemic stem cell (LSC) markers. Mol. Cell Proteom. 12,
17. Saygin, C., Matei, D., Majeti, R., Reizes, O. & Lathia, J. D. Targeting 626–637 (2013).
cancer stemness in the clinic: from hype to hope. Cell Stem Cell 24, 40. Marjon, K. D. et al. Tetraspanin CD82 regulates bone marrow
25–40 (2019). homing of acute myeloid leukemia by modulating the molecular
18. Naldini, M. M. et al. Longitudinal single-cell profiling of che- organization of N-cadherin. Oncogene 35, 4132–4140 (2016).
motherapy response in acute myeloid leukemia. Nat. Commun. 14, 41. Cheng, J., Qu, L., Wang, J., Cheng, L. & Wang, Y. High expression of
1285 (2023). FLT3 is a risk factor in leukemia. Mol. Med. Rep. 17, 2885–2892 (2018).

42. Grieselhuber, N. R. et al. Notch signaling in acute promyelocytic 63. Dong, F. et al. Predictive values of plasma TNFalpha and IL-8 for
leukemia. Leukemia 27, 1548–1557 (2013). intracranial hemorrhage in patients with acute promyelocytic leu-
43. Majeti, R. et al. Dysregulated gene expression networks in human kemia. Front. Med. 16, 909–918 (2022).
acute myelogenous leukemia stem cells. Proc. Natl Acad. Sci. USA 64. Lin, J. et al. TRIB3 stabilizes high TWIST1 expression to promote
106, 3396–3401 (2009). rapid APL progression and ATRA resistance. Clin. Cancer Res. 25,
44. Rodrigues, A. et al. Cell signaling pathways as molecular targets to 6228–6242 (2019).
eliminate AML stem cells. Crit. Rev. Oncol. Hematol. 160, 103277 65. Kwaan, H. C., Weiss, I. & Tallman, M. S. The role of abnormal
(2021). hemostasis and fibrinolysis in morbidity and mortality of acute pro-
45. Prunier, C. et al. Disruption of the PHRF1 tumor suppressor network myelocytic leukemia. Semin. Thromb. Hemost. 45, 612–621 (2019).
by PML-RARalpha drives acute promyelocytic leukemia pathogen- 66. Breccia, M. et al. Occurrence of thrombotic events in acute pro-
esis. Cell Rep. 10, 883–890 (2015). myelocytic leukemia correlates with consistent immunophenotypic
46. Coltella, N. et al. HIF factors cooperate with PML-RARalpha to and molecular features. Leukemia 21, 79–83 (2007).
promote acute promyelocytic leukemia progression and relapse. 67. Osterroos, A. et al. A risk score based on real-world data to predict
EMBO Mol. Med. 6, 640–650 (2014). early death in acute promyelocytic leukemia. Haematologica 107,
47. Cheedipudi, S. et al. A fine balance: epigenetic control of cellular 1528–1537 (2022).
quiescence by the tumor suppressor PRDM2/RIZ at a bivalent 68. Ondrussek, R. et al. Prognostic value and multifaceted roles of
domain in the cyclin a gene. Nucleic Acids Res. 43, 6236–6256 tetraspanin CD9 in cancer. Front. Oncol. 13, 1140738 (2023).
(2015). 69. Vaikari, V. P. et al. Clinical and preclinical characterization of CD99
48. Liao, T. T. et al. let-7 modulates chromatin configuration and target isoforms in acute myeloid leukemia. Haematologica 105,
gene repression through regulation of the ARID3B complex. Cell 999–1012 (2020).
Rep. 14, 520–533 (2016). 70. Touzet, L. et al. CD9 in acute myeloid leukemia: prognostic role and
49. Ligon, K. L. et al. Olig2-regulated lineage-restricted pathway con- usefulness to target leukemic stem cells. Cancer Med. 8,
trols replication competence in neural stem cells and malignant 1279–1288 (2019).
glioma. Neuron 53, 503–517 (2007). 71. Purton, L. E. Roles of retinoids and retinoic acid receptors in the
50. Ng, S. W. et al. A 17-gene stemness score for rapid determination of regulation of hematopoietic stem cell self-renewal and differ-
risk in acute leukaemia. Nature 540, 433–437 (2016). entiation. PPAR Res. 2007, 87934 (2007).
51. Newman, A. M. et al. Determining cell type abundance and 72. Oetjen, K. A. et al. Human bone marrow assessment by single-cell
expression from bulk tissues with digital cytometry. Nat. Biotechnol. RNA sequencing, mass cytometry, and flow cytometry. JCI Insight
37, 773–782 (2019). 3, e124928 (2018).
52. Newman, A. M. et al. Robust enumeration of cell subsets from tissue 73. Witkowski, M. T. et al. Extensive remodeling of the immune micro-
expression profiles. Nat. Methods 12, 453–457 (2015). environment in B cell acute lymphoblastic leukemia. Cancer Cell
53. Lin, X. et al. Integration of genomic and transcriptomic markers 37, 867–882.e812 (2020).
improves the prognosis prediction of acute promyelocytic leuke- 74. Holland, C. H. et al. Robustness and applicability of transcription
mia. Clin. Cancer Res. 27, 3683–3694 (2021). factor and pathway analysis tools on single-cell RNA-seq data.
54. Slack, J. L. et al. Molecular analysis and clinical outcome of adult Genome Biol. 21, 36 (2020).
APL patients with the type V PML-RARalpha isoform: results from 75. Bergen, V., Lange, M., Peidli, S., Wolf, F. A. & Theis, F. J. Generalizing
intergroup protocol 0129. Blood 95, 398–403 (2000). RNA velocity to transient cell states through dynamical modeling.
55. Paietta, E. et al. A surrogate marker profile for PML/RAR alpha Nat. Biotechnol. 38, 1408–1414 (2020).
expressing acute promyelocytic leukemia and the association of 76. Qiu, X. et al. Single-cell mRNA quantification and differential ana-
immunophenotypic markers with morphologic and molecular lysis with Census. Nat. Methods 14, 309–315 (2017).
subtypes. Cytom. B Clin. Cytom. 59, 1–9 (2004). 77. Qiu, P. Embracing the dropouts in single-cell RNA-seq analysis. Nat.
56. Lee, J. J. et al. CD34 expression is associated with poor clinical Commun. 11, 1169 (2020).
outcome in patients with acute promyelocytic leukemia. Am. J. 78. Wu, S. et al. Interplay between hypertriglyceridemia and acute
Hematol. 73, 149–153 (2003). promyelocytic leukemia mediated by the cooperation of peroxi-
57. Foley, R. et al. CD34-positive acute promyelocytic leukemia is some proliferator-activated receptor-alpha with the PML/RAR alpha
associated with leukocytosis, microgranular/hypogranular mor- fusion protein on super-enhancers. Haematologica 107,
phology, expression of CD2 and bcr3 isoform. Am. J. Hematol. 67, 2589–2600 (2022).
34–41 (2001). 79. Bao, C. et al. OpenXGR: a web-server update for genomic summary
58. Albano, F. et al. The biological characteristics of CD34+ CD2+ adult data interpretation. Nucleic Acids Res. 51, W387–W396 (2023).
acute promyelocytic leukemia and the CD34 CD2 hypergranular 80. Kuleshov, M. V. et al. Enrichr: a comprehensive gene set enrichment
(M3) and microgranular (M3v) phenotypes. Haematologica 91, analysis web server 2016 update. Nucleic Acids Res. 44,
311–316 (2006). W90–W97 (2016).
59. Breccia, M. et al. Negative prognostic value of CD34 antigen also if 81. Dai, Y. et al. Cellular hierarchy insights reveal leukemic stem-like
expressed on a small population of acute promyelocytic leukemia cells and early death risk in acute promyelocytic leukemia. https://
cells. Ann. Hematol. 93, 1819–1823 (2014). doi.org/10.5281/zenodo.10437695 (2023).
60. Quek, L. et al. Genetically distinct leukemic stem cells in human
CD34- acute myeloid leukemia are arrested at a hemopoietic Acknowledgements
precursor-like stage. J. Exp. Med. 213, 1513–1535 (2016). This work was supported by the National Natural Science Foundation of
61. Taussig, D. C. et al. Leukemia-initiating cells from some acute China (81890994, 82350710226, 82370178, and 32170663) and the
myeloid leukemia patients with mutated nucleophosmin reside in National Key R&D Program of China (2023YFA1800401).
the CD34(-) fraction. Blood 115, 1976–1984 (2010).
62. Sanz, M. A. et al. Management of acute promyelocytic leukemia: Author contributions
updated recommendations from an expert panel of the European K.W. designed the experiments; W.J. performed the experiments; Y.D.
LeukemiaNet. Blood 133, 1630–1643 (2019). and H.F. analyzed the data; W.J., H.H.Z., and F.D. collected clinical

samples; L.C., H.H.Z., H.M.Z., and J.L. followed up with the patients; K.W. Reprints and permissions information is available at
and H.F. supervised the study and analysis; K.W., H.F., W.J., and Y.D. http://www.nature.com/reprints
wrote the manuscript; Z.C., S.C., and G.M. gave conceptual advice; and
all authors discussed the results and implications and reviewed the Publisher’s note Springer Nature remains neutral with regard to jur-
manuscript. isdictional claims in published maps and institutional affiliations.
Competing interests Open Access This article is licensed under a Creative Commons
The authors declare no competing interests. Attribution 4.0 International License, which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as
Additional information long as you give appropriate credit to the original author(s) and the
Supplementary information The online version contains source, provide a link to the Creative Commons licence, and indicate if
supplementary material available at changes were made. The images or other third party material in this
https://doi.org/10.1038/s41467-024-45737-7. article are included in the article’s Creative Commons licence, unless
indicated otherwise in a credit line to the material. If material is not
Correspondence and requests for materials should be addressed to Zhu included in the article’s Creative Commons licence and your intended
Chen, Hai Fang or Kankan Wang. use is not permitted by statutory regulation or exceeds the permitted
use, you will need to obtain permission directly from the copyright
Peer review information Nature Communications thanks Bernhard holder. To view a copy of this licence, visit http://creativecommons.org/
Gentner, Derek Janssens, and the other, anonymous, reviewer(s) for licenses/by/4.0/.
their contribution to the peer review of this work. A peer review file is
available. © The Author(s) 2024

09 Cellular Hierarchy Insights Reveal Leukemic Stem-Like Cells and Early Death Risk in Acute Promyelocytic Leukemia

Uploaded by

Copyright:

Available Formats

09 Cellular Hierarchy Insights Reveal Leukemic Stem-Like Cells and Early Death Risk in Acute Promyelocytic Leukemia

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

09 Cellular Hierarchy Insights Reveal Leukemic Stem-Like Cells and Early Death Risk in Acute Promyelocytic Leukemia

Uploaded by

Copyright:

Available Formats

Article https://doi.org/10.

Cellular hierarchy insights reveal leukemic

ferentiation therapy, with a minority of patients experiencing treatment failure

Nature Communications | (2024)15:1423 1

Nature Communications | (2024)15:1423 2

16 scRNA-seq (10X) of 23 scRNA-seq (10X) of

Batch effects adjusted APL stem-like

Characteristic of malignant APL blasts Intra-tumoral heterogeneity of APL blasts

scTarget APL cohort In vivo treatment In vivo treatment

Day 0 Day 2 Day 0 Day 2

PML/RARα Percent CYTL1

120 3.2e-3 SP3 MNDA

positive regulation of cytokine production 5.0 1.5e-3 CEBPA

Nature Communications | (2024)15:1423 3

Nature Communications | (2024)15:1423 4

a UMAP b Spearman’s correlation Gene mean expression in different cell types

Cell type percentage

Stem-like 6 -4 APL (10X) -4

Nature Communications | (2024)15:1423 5

Nature Communications | (2024)15:1423 6

a Top 500 PML/RARα targets b FCGR2A CD9

Expression Expression Expression

Expression Expression Expression

0.2 HCK GFI1

c Upregulated PML/RARα targets Downregulated PML/RARα targets d TF activities

FLT3-TKD WT1 mutation g Branches h FLT3-ITD expression (scTarget)

Nature Communications | (2024)15:1423 7

APL stem-like cell (%) in bulk RNA-seq and scRNA-seq

Pearson’s correlation coefficients between

0.10 0.14 0.18 0.22

Nature Communications | (2024)15:1423 8

a 323 de novo APL bulk RNA-seq cohort c Overall survival

Nature Communications | (2024)15:1423 9

a b Day 0 (de novo APL) Day 2 (after ATRA treatment) c

% of stem-like cells in APL blasts

10.0 ATRA 0.011 0.0012

log 2 e e e e log 2 like like like like log 2 e e e e log 2 e e e e

log2FC (Day 2 vs. Day 0) ATRA Day 0 Mean expression

i Stemness CD markers Stemness TFs j Differentiation markers Hematopoietic TFs

CD200 ERG ITGAE CEBPA

log2FC (Day 2 vs. Day 0) Clinical outcomes Alive Early death

Nature Communications | (2024)15:1423 10

Nature Communications | (2024)15:1423 11

Nature Communications | (2024)15:1423 12

Nature Communications | (2024)15:1423 13

Code availability 19. Guibal, F. C. et al. Identiﬁcation of a myeloid committed progenitor

Nature Communications | (2024)15:1423 14

Nature Communications | (2024)15:1423 15

Nature Communications | (2024)15:1423 16

You might also like