Graphical abstract

   graphic file with name fx1.jpg
   [64]Open in a new tab

Highlights

     * •
       RAS mutations induce inflammatory network activation and immune
       evasion in CMML
     * •
       Expansion of NF-κB-dependent HSPCs drives blast progression (BP)
       after HMA therapy
     * •
       RAS mutant CMML HSPCs rely on MCL1 for survival during BP
     * •
       NF-κB pathway activation in RAS mutant HSPCs drives venetoclax
       resistance
     __________________________________________________________________

   In brief, Montalban-Bravo et al. show that RAS pathway mutations induce
   multi-step reprogramming of distinct cellular populations in CMML.
   Expansion of myeloid progenitors at blast progression following
   hypomethylating agent therapy is driven by NF-κB pathway activation and
   MCL1 dependency. These data support the potential of NF-κB-targeted
   therapies in CMML.

Introduction

   Chronic myelomonocytic leukemia (CMML), a clonal disorder of mutant
   hematopoietic stem cells (HSCs),[65]^1 is characterized by
   myelodysplastic and myeloproliferative bone marrow (BM)
   features,[66]^2^,[67]^3 and a high risk of progression to acute myeloid
   leukemia (AML).[68]^4^,[69]^5^,[70]^6 Hypomethylating agent (HMA)
   therapy, the current standard of care for most patients with
   CMML,[71]^7 can overcome CMML cells’ aberrant proliferation and achieve
   improved outcomes in some patients. However, most patients only have
   transient responses to HMA therapy, owing to these agents’ inability to
   effectively deplete HSCs and decrease tumor burden. CMML patients whose
   disease undergoes transformation to AML upon HMA therapy failure have
   dismal clinical outcomes.[72]^8^,[73]^9

   Despite advances in the genetic characterization of CMML, the
   development of alternative frontline treatments or more effective
   second-line therapies to improve the outcomes of CMML patients with
   high-risk biological features has been delayed because of an incomplete
   understanding of the ways in which different hematopoietic populations
   that persist throughout HMA therapy contribute to disease maintenance
   and progression.

   Mutations in RAS pathway signaling genes (BRAF, CBL, KRAS, NF1, NRAS,
   and PTPN11) confer adverse biological features that increase the risk
   of disease progression and poor overall survival, particularly when
   they are concurrently present with loss-of-function mutations in the
   ASXL transcriptional regulator 1 gene, ASXL1.[74]^10

   Herein, we used single-cell technology-based approaches to elucidate
   the biological and molecular landscape of RAS pathway-mutated CMML to
   guide the selection of future therapeutic interventions and achieve
   durable responses in CMML patients in whom blast progression (BP)
   occurs after failure to HMA therapy.

Results

Mutations in RAS pathway signaling genes predict a high risk of CMML BP after
HMA therapy failure

   We first evaluated whether specific mutations predict a high risk of
   CMML BP in a cohort of 108 CMML patients who received HMA therapy
   ([75]Table S1). After a median follow-up of 19 months (95% confidence
   interval [CI], 15.8–23.9 months), 57 patients experienced HMA therapy
   failure; 36 patients had BP at the time of therapy failure. Mutations
   in RAS pathway genes were significantly associated with BP (odds
   ratio = 3.35; 95% CI, 1.46–7.70; p = 0.004) ([76]Figure 1A) and shorter
   time to BP (hazard ratio = 2.21; 95% CI, 1.13–4.33; p = 0.021)
   ([77]Figure 1B). Similarly, logistic regression analysis showed that
   RAS pathway mutations were associated with a higher risk of BP (p =
   0.01158) ([78]Table S2). To assess whether BP was associated with
   mutations that were not detected at diagnosis or with the clonal
   expansion of pre-existing mutations, we sequenced BM cells isolated
   from samples collected at the time of BP after HMA failure from 22 of
   the 36 patients and compared the cells’ genomic landscape with that of
   BM cells isolated at diagnosis ([79]Figure 1C). Among 22 patients with
   BP, 14 (64%) had RAS pathway mutations at diagnosis, and 20 (91%) had
   RAS pathway mutations at BP. Nine patients (41%) acquired newly
   detectable RAS pathway mutations at BP (4 patients had no detectable
   RAS pathway mutations at diagnosis, and 5 patients acquired other RAS
   pathway mutations). Of the 14 patients with RAS pathway-mutated CMML at
   diagnosis, 10 had BP without RAS pathway mutation-induced clonal
   evolution ([80]Figure 1D).

Figure 1.

   [81]Figure 1
   [82]Open in a new tab

   Mutations in RAS pathway signaling genes predict a high risk of CMML BP
   after HMA therapy failure

   (A) Bar chart showing the frequencies of detectable mutations and
   cytogenetic abnormalities among 108 CMML patients who received HMA
   therapy and whose disease progressed (green) or did not progress
   (blue). Asterisks indicate significantly different frequency changes
   (∗p < 0.05).

   (B) Kaplan-Meier survival curves showing the cumulative incidence of BP
   after HMA therapy in previously untreated CMML patients with or without
   RAS pathway mutations. N, number; HR, hazard ratio; CI, confidence
   interval.

   (C) Bar chart showing the overall frequencies of detectable mutations
   among 22 CMML patients whose disease progressed and in whom targeted
   sequencing was performed at the time of BP. Mutations at diagnosis and
   BP are indicated by pink and green, respectively. Paired samples were
   analyzed.

   (D) Detected mutations and their variant allele frequencies (VAFs) in
   matched samples obtained at diagnosis and at the time of BP in the 22
   CMML patients shown in (C). Columns represent the mutations and VAFs
   from sequential samples of individual CMML patients at diagnosis and
   BP. Patient identifiers are shown at the top of each column. Asterisks
   indicate the presence of multiple mutations in a particular gene. The
   numbers of RAS mutations are shown in red gradient; the VAFs of each
   mutation are shown in blue gradient.

   These results were validated using single-cell DNA sequencing coupled
   with cell-surface immunophenotyping analysis of mononuclear cells
   (MNCs) isolated from sequential BM samples obtained at the time of
   diagnosis or BP from two representative RAS pathway-mutated CMML
   patients whose disease never responded to therapy ([83]Figures S1A and
   S1B) or underwent clonal evolution after an initial response
   ([84]Figures S2A and S2B). Taken together, these data demonstrate that
   patients with RAS pathway-mutated CMML have a high risk of BP at the
   time of HMA therapy failure. This observation has important clinical
   implications in light of our recent study showing that RAS pathway
   mutations also drive resistance to and/or BP following venetoclax-based
   second-line therapy.[85]^11 These data underscore the urgent need to
   dissect the biological mechanisms of RAS pathway mutation-induced
   therapy resistance, as such an understanding could lead to the
   development of future therapeutic approaches to prevent or overcome
   disease progression.

RAS pathway mutations activate cell-intrinsic and -extrinsic inflammatory
networks

   To dissect the molecular mechanisms underlying the progression of RAS
   pathway-mutated CMML, we first evaluated the molecular determinants of
   disease initiation. We performed single-cell RNA sequencing (scRNA-seq)
   analysis of lineage-negative (Lin^–) CD34^+ hematopoietic stem and
   progenitor cells (HSPCs) isolated from five untreated RAS pathway
   mutant CMML patients and two age-matched healthy donors (HDs)
   ([86]Table S3). This analysis identified eight cellular clusters driven
   by the differentiation profile of the cells ([87]Figure 2A), which we
   defined based on the differential expression of validated
   lineage-specific transcriptional factors (TFs) and cellular
   markers[88]^12^,[89]^13 ([90]Figure S3A; [91]Table S4). Compared with
   HSPCs from HDs, Lin^–CD34^+ HSPCs from RAS pathway mutant CMML patients
   had a predominant myeloid differentiation route with higher frequencies
   of early myeloid hematopoietic progenitor cells (eMyHPCs) (clusters 1
   and 3, characterized by the high expression of CD34, BTF3, and CEBPA
   but low expression of CD38) and more differentiated MyHPCs (dMyHPCs)
   (clusters 0, 4, and 6, marked by the expression of CEBPD and/or CEBPA,
   as well as that of MPO) at the expense of more primitive HSCs (cluster
   2, marked by the high expression of MLLT3, MEG3, and CLEC9A), and
   erythroid/megakaryocyte (Ery/Mk) HPCs (clusters 5 and 7, marked by the
   expression of KLF1, GATA1, and GATA2) ([92]Figure 2B). Differential
   expression analysis revealed that genes upregulated in RAS pathway
   mutant CMML HSCs compared with HD HSCs were mainly involved in
   oxidative phosphorylation, interferon (IFN) response, and apoptosis
   ([93]Figures 2C and [94]S3B). Similar results were observed in eMyHPCs
   and dMyHPCs ([95]Figure 2C).

Figure 2.

   [96]Figure 2
   [97]Open in a new tab

   RAS pathway-mutated CMML cells activate cell-intrinsic and -extrinsic
   inflammatory networks

   (A) UMAP of scRNA-seq data for pooled single Lin^–CD34^+ cells isolated
   from BM samples of two HDs (n = 895) and five RAS pathway mutant CMML
   patients (n = 3,161). Each dot represents one cell. Different colors
   represent the cluster cell-type identity (left) or sample origin
   (right). HSC, hematopoietic stem cells; eMyHPC, early myeloid
   progenitor cells; dMyHPC, differentiated myeloid progenitors;
   Ery/MkHPC, erythroid/megakaryocyte hematopoietic progenitor cells.
   Dashed lines indicate single clusters in each cell-type population.

   (B) Distribution of HD (top) and RAS pathway mutant CMML (bottom)
   Lin^–CD34^+ cell types among the clusters shown in (A).

   (C) Pathway enrichment analysis of the genes that were significantly
   upregulated in HSCs (left), eMyHPCs (middle), and dMyHPCs (right) from
   the five RAS pathway mutant CMML samples shown in (A) compared with
   those from HD samples (adjusted p ≤ 0.05). The top 10 hallmark gene
   sets are shown.

   (D) UMAP of scRNA-seq data for pooled single MNCs isolated from BM
   samples of three HDs (n = 12,836), three RAS pathway wild-type
   (RAS[wt]) (n = 16,038), and five RAS pathway mutant (RAS[mut]) (n =
   12,234) CMML patients. Each dot represents one cell. Different colors
   represent the cluster cell-type identity (left) or sample origin
   (right). HSC, hematopoietic stem cells; MKP, megakaryocyte precursors;
   Ery, erythroid precursors; MyHPC, myeloid hematopoietic progenitor
   cells; Mono, monocytes; CD16 Mono, non-classical CD16^+ monocytes; cDC,
   classical dendritic cells; pDC, plasmacytoid dendritic cells; B cell, B
   lymphocytes; PC, plasma cells; nCD4T, naive CD4^+ T cells; mCD4/CD8,
   memory CD4^+ and CD8^+ T cells; eCD8T, effector CD8^+ T cells; NKC,
   natural killer cells. Dashed lines indicate single clusters in each
   cell-type population.

   (E) Distribution of HD (top), and RAS pathway wild-type (RAS[wt])
   (middle) or mutant (RAS[mut]) CMML (bottom) BM MNC populations among
   the clusters shown in (D).

   (F) Pathway enrichment analysis of the genes that were significantly
   upregulated in RAS pathway wild-type (top left) or mutant (top right)
   CMML monocyte clusters compared with those in HD and RAS pathway mutant
   monocyte clusters compared with those in RAS pathway wild-type monocyte
   clusters (bottom left) (adjusted p ≤ 0.05). The top 10 hallmark gene
   sets are shown.

   To evaluate the contribution of downstream myelo/monocytic (My/Mo)
   populations to disease maintenance, we performed scRNA-seq analysis of
   BM MNCs isolated from three HDs, and five untreated RAS pathway mutant
   CMML samples. To dissect the specific role of RAS pathway mutations in
   disease initiation, we also included BM MNCs from three untreated RAS
   pathway wild-type CMML samples. This analysis identified 18 cellular
   clusters inclusive of all major BM cell types that we defined based on
   the expression of lineage-specific TFs and cellular markers and using
   the single-cell transcriptome to protein prediction with deep neural
   network pipeline[98]^14^,[99]^15 ([100]Figures 2D and [101]S3C;
   [102]Table S5). Consistent with the predominant myeloid differentiation
   bias of CMML HSPCs, differential analysis of BM cell lineage
   composition revealed that the monocyte population (clusters 0 and 4)
   increased in BM CMML samples compared with that in BM HD samples,
   regardless of the presence of RAS pathway mutations ([103]Figure 2E).

   However, although CMML monocytes from RAS pathway wild-type CMML
   underwent transcriptional reprogramming compared with those from HDs,
   RAS pathway mutant monocytes had significantly enhanced upregulation of
   IFN and NF-κB signaling-mediated inflammatory responses compared with
   RAS pathway wild-type monocytes ([104]Figures 2F and [105]S4A).

   Inflammatory networks modulate the immune microenvironment and
   contribute to immune escape.[106]^16 To assess whether CMML monocytes
   directly suppress the immune response and whether RAS pathway mutations
   modulate such interactions, we dissected the intercellular crosstalk
   and communication networks between CMML cells and all other BM cells.
   We inferred cell-to-cell communications from the combined expression of
   multi-subunit ligand-receptor complexes using CellPhoneDB, a repository
   of ligands and receptors and their interactions.[107]^17 After
   generating a homeostatic interactome of BM MNCs from HDs, we analyzed
   the cellular communication networks that were upregulated in RAS
   pathway wild-type and mutant CMML BM samples ([108]Figure S4B;
   [109]Table S6). Compared with HD MNCs, RAS pathway mutant CMML MNCs had
   significantly more ligand-receptor interactions involving monocytes,
   classical dendritic cells (cDCs), plasmacytoid DCs (pDCs), MyHPCs,
   effector CD8^+ T (eCD8T) cells, and natural killer (NK) cells
   ([110]Figure S4B). Monocytes, cDCs, and immune populations from
   patients with RAS pathway mutant CMML gained significantly more
   ligand-to-receptor interactions compared with those without RAS pathway
   mutations ([111]Figure S4B). Specifically, expression levels of
   chemokine genes (CCL3 and CCL3L1) and cytokine genes (IL1B, TNFSF10,
   MIF, and HGF) involved in inflammatory signaling and NF-κB-mediated
   cell survival were significantly increased in CMML monocytes, cDCs,
   pDCs, and MyHPCs, and enriched in patients with RAS pathway mutations
   ([112]Figure S4C). Monocytes and cDCs from patients with RAS pathway
   mutations expressed higher levels of the receptors of these ligands
   (CCR1, CCR5, CD74, and TNFRSF10B), which suggests that an
   aberrant feedback loop among different cell types preferentially
   contributes to CMML maintenance in RAS pathway mutant CMML. Together,
   these data are consistent with previous findings showing that, in other
   cancers, NF-κB signaling activation is essential for RAS pathway
   mutation-induced tumorigenesis.[113]^18^,[114]^19^,[115]^20^,[116]^21

   CMML monocytes, pDCs, and cDCs also gained cell-to-cell interactions
   with NK and eCD8T cells. Interactions involving the HLA-E-KLRC1/2,
   CDH1-KLRG1, LGALS9-HAVCR2, and TGFB1-TGBR1/3 ligand-receptor pairs
   (known to inhibit the immune cell
   functions[117]^22^,[118]^23^,[119]^24^,[120]^25^,[121]^26^,[122]^27^,[1
   23]^28^,[124]^29) were the most common ([125]Figure S4C;
   [126]Table S6). To evaluate whether CMML BM monocytes and immune cells
   spatially co-localized, we performed multiplex immunofluorescence
   analysis of BM biopsy sections obtained from CMML patients (n = 4) at
   the time of diagnosis. This analysis revealed that BM monocytes
   (CD14^+CD68^− cells) resided within a median of 19.73 μm (95% CI,
   12.75–32.25 μm) from CD8^+ T cells and 22.62 μm (95% CI,
   15.73–32.17 μm) from NK cells (CD3^−CD56^+ cells; [127]Figures S5A and
   S5B), which suggests that these cell populations interact with each
   other. Accordingly, both CMML NK cells (cluster 1) and eCD8T cells
   (cluster 3) had increased expression levels of immune checkpoint genes
   associated with these cells’ exhaustion (e.g., KLRG1, KLRC1, TIGIT,
   LAG3, CD244, B3GAT1, and CD160) compared with those from HDs
   ([128]Figure S5C).[129]^30^,[130]^31^,[131]^32 To further characterize
   the functional state of CD8^+ T and NK cells in CMML, we evaluated the
   expression of activation markers on these cells after antigen exposure.
   After co-culture with K562 AML cells, the frequencies of IFN-γ^+ CD8^+
   T cells and activated CD16^+ NK cells were significantly lower in RAS
   pathway mutant CMML but not in RAS pathway wild-type CMML, compared
   with those in HDs ([132]Figure S5D). In addition, IFN-γ^+ CD8^+ T cells
   and NK cells from patients with RAS pathway mutant CMML, but not those
   from RAS pathway wild-type CMML, had significantly lower IFN-γ and
   perforin expression levels, respectively ([133]Figure S5D).

   Taken together, these data suggest that CMML HSPCs and downstream My/Mo
   cells undergo significant transcriptional rewiring and that RAS pathway
   mutations enhance the activation of cell-intrinsic and -extrinsic
   inflammatory networks in CMML monocyte populations to maintain cell
   proliferation and suppress the immune microenvironment, thus enabling
   immune escape and clonal expansion.

RAS pathway-mutated HSCs upregulate NF-κB transcriptional programs and drive
CMML BP after HMA therapy failure

   To evaluate the cellular and molecular dynamics of CMML progression, we
   performed scRNA-seq analysis of Lin^–CD34^+ HSPCs isolated from BM
   samples sequentially obtained from five RAS pathway mutant CMML
   patients at diagnosis and BP ([134]Figures 3A and [135]S6A;
   [136]Table S7). HSPCs isolated from BM samples obtained at BP
   maintained aberrant differentiation toward the My/Mo lineage
   ([137]Figure 3B) and had upregulated genes belonging to the NF-κB
   signaling pathway ([138]Figure 3C). Importantly, MCL1, an
   anti-apoptotic member of the BCL2 family and a downstream effector of
   the NF-κB pathway, was significantly upregulated in HSCs (cluster 6)
   and eMyHPCs (clusters 0 and 1) at BP compared with those at diagnosis
   ([139]Figures S6B and S6C; [140]Table S8).

Figure 3.

   [141]Figure 3
   [142]Open in a new tab

   RAS pathway-mutated HSCs undergo epigenetic reprogramming and drive
   CMML BP after HMA therapy failure

   (A) UMAP of scRNA-seq data for pooled single Lin^–CD34^+ cells isolated
   from BM samples of five RAS pathway mutant CMML patients at diagnosis
   (n = 1,840) and at BP after HMA therapy failure (n = 1,711). Each dot
   represents one cell. Different colors represent the cluster cell-type
   identity (left) or sample origin (right). HSC, hematopoietic stem
   cells; eMyHPC, early myeloid hematopoietic progenitor cells; dMyHPC,
   differentiated myeloid hematopoietic progenitor cells; Ery/MkHPC,
   erythroid/megakaryocyte hematopoietic progenitor cells. Dashed lines
   indicate single clusters in each cell-type population.

   (B) Distribution of Lin^–CD34^+ cell types at diagnosis (top) and BP
   (bottom) among the clusters shown in (A).

   (C) Pathway enrichment analysis of the genes that were significantly
   upregulated in HSCs (left) and dMyHPCs (right) at the time of BP after
   HMA therapy failure compared with those at diagnosis (adjusted p ≤
   0.05). The top 10 hallmark gene sets are shown.

   (D) UMAP of scATAC-seq data for pooled Lin^–CD34^+ cells isolated from
   BM samples obtained from three RAS pathway mutant CMML patients at
   diagnosis (n = 5,066) and at BP after HMA therapy failure (n = 8,603).
   Each dot represents one cell. Different colors represent the cluster
   identity (left) or sample of origin (right). HSC, hematopoietic stem
   cells; MyHPC, myeloid progenitor cells; Ery/MkHPC,
   erythroid/megakaryocyte hematopoietic progenitor cells.

   (E) Pathway enrichment analysis of genes whose promoters were enriched
   in open chromatin regions in HSCs (clusters 0 and 4, shown in D) at the
   time of BP as compared with those at diagnosis (p ≤ 10^−4). The top 10
   hallmark gene sets are shown.

   (F) Pathway enrichment analysis of genes whose distal elements were
   enriched in open chromatin regions in HSCs (clusters 0 and 4, shown in
   D) at the time of BP as compared with those at diagnosis (adjusted p ≤
   0.05). The top 10 hallmark gene sets are shown.

   To evaluate whether HSPCs’ transcriptional changes at BP resulted from
   epigenetic reprogramming in the more primitive HSCs, we performed
   single-cell assays for transposase-accessible chromatin with
   high-throughput sequencing (scATAC-seq) to profile the chromatin
   accessibility landscape in Lin^–CD34^+ HSPCs isolated from BM samples
   sequentially obtained from three RAS pathway mutant CMML patients at
   diagnosis or BP. Our analysis identified five clusters with distinct TF
   binding motif enrichment in the open chromatin regions ([143]Figures 3D
   and [144]S6D; [145]Table S9). MyHPCs (clusters 1 and 2) were
   characterized by open chromatin regions in the binding motifs of the
   myeloid TFs SPI1B and CEBPA and TFs belonging to the FOS and JUN
   families. HSCs (clusters 0 and 4) had the highest activities of TFs
   involved in stemness maintenance, such as HLF and TFs belonging to the
   nuclear retinoid receptor and EGR families. Ery/MkHPCs (cluster 3) were
   characterized by open chromatin regions in binding motifs for GATA TFs.

   Consistent with our transcriptomic data, HSCs at BP had increased open
   chromatin peaks at the promoters of genes involved in NF-κB pathway
   activation and inflammatory response pathways ([146]Figure 3E),
   including MCL1 ([147]Figure S6E; [148]Table S10). HSCs also showed
   increased open chromatin peaks of genes involved in NF-κB pathway
   activation and inflammatory response pathways at the genes’ distal
   elements, which define cell identity and differentiation trajectories
   more precisely than promoter regions do[149]^33 ([150]Figures 3F and
   [151]S6F).

   Taken together, these data suggest that RAS pathway-mutated CMML HSCs
   exacerbate the activation of inflammatory and NF-κB pathway
   transcriptional programs and promote transcriptional upregulation of
   NF-κB signaling-mediated anti-apoptotic pathways to maintain survival
   at BP after HMA failure.

RAS pathway-mutated CMML cells rely on MCL1 overexpression to maintain their
survival at BP

   To elucidate whether MyHPCs’ transcriptional and epigenetic
   reprogramming drives BP, we performed scRNA-seq analysis of MNCs
   isolated from sequential RAS pathway-mutated CMML BM samples obtained
   from six CMML patients at diagnosis and BP ([152]Figures 4A and
   [153]S7A; [154]Table S11). MNCs at BP had a significantly higher
   frequency of CD34^+ MyHPCs (22.5% vs. 2.8%, respectively; clusters 7,
   8, and 14) compared with those at diagnosis ([155]Figures 4B and
   [156]S7B). These results were confirmed by flow cytometry analysis of
   Lin^–CD34^+ cells in 70% of the patients ([157]Figure S7C), which
   suggests that BP after HMA therapy is mostly driven by the expansion of
   the HSPC compartment. Consistent with our results in the Lin^–CD34^+
   compartment, differential expression analysis confirmed that MyHPCs at
   BP had upregulated genes involved in TNF-α-mediated NF-κB activation
   ([158]Figure S7D), including MCL1 ([159]Figures S7E and S7F;
   [160]Table S12). The upregulation of these genes was maintained in
   downstream My/Mo progenitors (My/MoPs; clusters 5 and 17) and was
   significantly increased in the monocytic populations (clusters 2, 3,
   11, 15, and 20) ([161]Figures 4C and [162]S7G; [163]Table S12).
   Consistent with these transcriptomic results, CD34^+ BM cells from
   patients with RAS pathway mutant CMML at BP (n = 2) had higher MCL1
   protein expression than did cells from patients at diagnosis (n = 3;
   [164]Figure S7H).

Figure 4.

   [165]Figure 4
   [166]Open in a new tab

   RAS pathway-mutated CMML cells rely on MCL1 overexpression to maintain
   their survival at BP

   (A) UMAP of scRNA-seq data for pooled single MNCs isolated from BM
   samples of six RAS pathway mutant CMML patients at diagnosis (n =
   16,372) and at BP after HMA therapy failure (n = 19,541). Each dot
   represents one cell. Different colors represent the cluster cell-type
   identity (left) or sample of origin (right). MyHPC, myeloid
   hematopoietic progenitor cells; My/MoP, myelo/monocytic progenitors;
   Mono, monocytes; cDC, classical dendritic cells; pDC, plasmacytoid
   dendritic cells; MKP, megakaryocyte precursors; Ery-E, early erythroid
   precursors; Ery-L, late erythroid precursors; B cell, B lymphocytes;
   PC, plasma cells; nCD4T, naive CD4^+ T cells; mCD4T, memory CD4^+
   T cells; eCD8T, effector CD8^+ T cells; NKC, natural killer cells.
   Dashed lines indicate single clusters in each cell-type population.

   (B) Distribution of MNC populations at diagnosis (top) and progression
   (bottom) among the clusters shown in (A).

   (C) Pathway enrichment analysis of the genes that were significantly
   upregulated in the monocytic populations (clusters 2, 3, 11, 15, and
   20) shown in (A) at the time of BP after HMA therapy failure compared
   with those at diagnosis (adjusted p ≤ 0.05). The top 10 hallmark gene
   sets are shown.

   (D) Numbers of live Lin^–CD34^+CD38^– HSCs and Lin^–CD34^+CD38^+ MyHPCs
   from CMML patients with BP after treatment with vehicle or AMG-176 (n =
   4, 20 nM) for 48 h. Lines represent means ± SD. Statistical
   significance was calculated using a two-tailed Student’s t test
   (∗∗∗p < 0.001, ∗∗∗∗p < 0.0001).

   (E) UMAP of scRNA-seq data for pooled single MNCs isolated from BM
   samples obtained from a representative CMML patient at the time of BP
   after HMA therapy failure (n = 6,209) and subsequent failure to
   venetoclax-based therapy (n = 6,795). Each dot represents one cell.
   Different colors represent the cluster cell-type identity (left) or the
   sample of origin (right). HSC, hematopoietic stem cells; MyHPC, myeloid
   hematopoietic progenitor cells; My/MoP, myelo/monocytic progenitors;
   Mono, monocytes; Ery/MkHPC, erythroid/megakaryocytic hematopoietic
   progenitor cells; Ery-E, early erythroid precursors; Ery-L, late
   erythroid precursors; Pre-E, pre-erythrocytes; mCD8T, memory CD8^+
   T cells; eCD8T, effector CD8^+ T cells; NKC, natural killer cells.

   (F) Pathway enrichment analysis of the genes that were significantly
   upregulated in MyHPCs at the time of venetoclax failure compared with
   those at the time of BP after HMA therapy failure (adjusted p ≤ 0.05).
   The top 10 hallmark gene sets are shown.

   (G) Distribution of myeloid cell types among the myeloid compartments
   at BP after HMA therapy failure (left) and venetoclax-based therapy
   failure (right).

   MNCs at BP exacerbated the cellular communication networks between
   cDCs, MyHPCs, My/MoPs, pDCs, monocytes, and eCD8T cells, compared with
   MNCs at baseline. Exacerbation of the cellular communication networks
   mainly occurred through immune-suppressive interactions between the
   LGALS9-HAVCR2 ligand-receptor pair, as well as increased CCL3/CCR1 and
   HGF/CD44 interactions between monocytes, MyMoPs and MyHPCs
   ([167]Figures S8A and S8B; [168]Table S13).

   To determine whether MCL1 upregulation was a hallmark of BP in RAS
   pathway mutant CMML or a general mechanism of treatment resistance and
   progression in CMML, we performed scRNA-seq analysis of BM MNCs
   isolated from RAS pathway wild-type CMML samples at diagnosis (n = 3)
   and BP after HMA failure (n = 3) ([169]Figures S9A and S9B;
   [170]Table S14). RAS pathway wild-type BM MNCs at progression had a
   higher frequency of MyHPCs compared with those at diagnosis (8.1% vs.
   3.9%, respectively; cluster 7; [171]Figure S9C). MyHPCs at BP
   upregulated genes involved in TNF-α-mediated NF-κB activation but not
   MCL1 ([172]Figures S9D and S9E; [173]Table S15). Similar data were also
   observed in downstream My/MoPs and monocytes ([174]Figure S9D;
   [175]Table S15). Consistent with these findings, CD34^+ BM cells from
   patients with RAS pathway wild-type CMML at diagnosis (n = 3) and BP
   (n = 3) had similar MCL1 protein expression levels ([176]Figure S7H).

   Together, these data suggest that only RAS pathway-mutated CMML MyHPCs
   and monocytes rely on MCL1-driven anti-apoptotic pathways to maintain
   survival and expand after therapy failure. To test this hypothesis, we
   treated Lin^–CD34^+ HSPCs isolated from the BM of patients with
   RAS pathway-mutated CMML with the MCL1 inhibitor AMG-176[177]^34 (at
   a dose that did not deplete Lin^–CD34^+CD38^– or Lin^–CD34^+CD38^+
   HSPCs isolated from the BM of HDs in co-culture system with mesenchymal
   stromal cells; [178]Figure S10A). AMG-176 significantly decreased the
   numbers of Lin^–CD34^+CD38^– and Lin^–CD34^+CD38^+ HSPCs isolated from
   BM samples obtained from patients with RAS pathway-mutated CMML at BP
   ([179]Figure 4D). AMG-176 did not significantly affect the survival of
   HSPCs isolated from BM samples obtained from patients at diagnosis
   ([180]Figure S10B), which confirms that CMML HSPCs maintain an intact
   apoptotic program at disease initiation. Consistent with our scRNA-seq
   analysis showing that MCL1 was not upregulated in RAS pathway wild-type
   HSPCs, the treatment with AMG-176 did not deplete RAS pathway wild-type
   HSPCs at the time of BP after HMA failure ([181]Figure S10C).

   Importantly, BCL2 was not overexpressed in either RAS pathway mutant
   CMML HSPCs or downstream My/Mo populations at BP after HMA therapy
   failure ([182]Figure S10D). BCL2 expression was significantly
   downregulated at progression in MyHPCs and My/MoPs from CMML patients
   without detectable RAS pathway mutations at the time of BP
   ([183]Figure S10E). These findings are consistent with our previous
   clinical observation that CMML patients in whom HMA therapy has failed
   do not benefit from second-line therapy with venetoclax.[184]^11
   Indeed, scRNA-seq analysis of MNCs isolated from sequential BM samples
   obtained from one representative RAS pathway mutant CMML patient whose
   disease progressed after HMA therapy failure and did not respond to
   venetoclax therapy ([185]Figures 4E and [186]S10F; [187]Table S16)
   revealed that MyHPCs further exacerbate the expression of genes
   involved in TNF-α-mediated NF-κB pathway activation ([188]Figure 4F),
   including MCL1 ([189]Figure S10G). Venetoclax failure was associated
   with a significant expansion of downstream My/MoPs in the myeloid
   compartment ([190]Figure 4G) and these cells’ high expression of MCL1
   ([191]Figure S10G).

   Taken together, our findings suggest that venetoclax therapy cannot
   overcome HMA failure-induced transcriptional reprogramming in My/MoPs
   and provide a rationale for targeting effectors of the NF-κB signaling
   pathway, such as MCL1, in patients with RAS pathway mutant CMML to
   improve the dismal outcome of CMML patients whose disease is resistant
   to available therapies.

Discussion

   Whereas the dissection of the molecular landscape of CMML initiation
   and progression has significantly advanced our understanding of the
   pathogenesis of CMML,[192]^1^,[193]^35^,[194]^36^,[195]^37^,[196]^38
   the development of more effective therapeutic approaches to improve
   patient survival has been delayed by our limited understanding of the
   ways in which genetic alterations affect distinct transcriptional
   states of My/Mo differentiation.

   Mutations in RAS pathway genes, which are present in 30% of CMML
   patients,[197]^38 are enriched during disease progression in up to 90%
   of the cases and predict a higher risk of and a shorter time to relapse
   after HMA and venetoclax therapy.[198]^39 Currently, there are no other
   therapies that improve the survival duration of patients with RAS
   pathway-mutated CMML.

   Using single-cell multi-omics technologies, we sought to dissect the
   biological mechanisms behind RAS pathway mutation-induced CMML
   evolution with the overall goal of identifying cellular vulnerabilities
   that could be therapeutically targeted to halt disease progression. We
   found that, at disease initiation, RAS pathway mutant CMML HSPCs
   significantly upregulated genes involved in the cell-intrinsic IFN
   signaling pathway such as IRF1, IRF7, IRF9, IFI44, IFI44L, IFIH1,
   IFIT3, or STAT2 that drive these cells’ differentiation toward the
   My/Mo lineage while maintaining an intact apoptotic program. Consistent
   with this observation and prior studies showing that KRAS or NRAS
   mutations directly activate intrinsic IFN-stimulated genes,[199]^40 IFN
   signaling activation in HSPCs was not associated with IFN receptor
   (IFNAR1, IFNAR2, IFNGR1, or IFNG2) or ligand (IFNG or IFNA)
   overexpression. In addition, this inflammatory reprogramming was
   exacerbated in downstream RAS pathway mutant monocyte populations,
   which expressed high levels of cytokines and cell surface receptors
   involved in NF-κB pathway activation and immune evasion. These results
   suggest that disease initiation and maintenance, as a result of RAS
   pathway mutations, rely on the activation of both cell-intrinsic and
   -extrinsic inflammatory networks in distinct cell populations and
   provide a rationale for using inhibitors of NF-κB-associated
   inflammatory signaling cascades as a frontline treatment for patients
   with RAS pathway-mutated CMML. These findings, which are consistent
   with previous studies showing the role of inflammatory cell populations
   in myeloid malignancies,[200]^41^,[201]^42 have significant
   implications since several inflammation-targeting therapies that are
   currently in clinical development have shown great potential to treat
   patients with myeloid malignancies.[202]^43^,[203]^44^,[204]^45

   Consistent with the long-standing observation that inhibition of
   apoptosis contributes to therapy resistance and cancer progression, we
   found that RAS pathway mutant CMML HSPCs isolated from BM samples at
   the time of BP depended on MCL1, an anti-apoptotic downstream effector
   of the NF-κB pathway, to maintain their survival and undergo clonal
   expansion. Consistent with this observation, we had demonstrated
   previously that TNF-α-mediated NF-κB pathway activation represents a
   cell-intrinsic adaptive mechanism to overcome cell death in response to
   therapeutic pressure.[205]^46 In addition, our findings align with
   prior data demonstrating that RAS mutations can directly induce NF-κB
   hyperactivation.[206]^21^,[207]^47 Notably, targeting MCL1 activity
   with the small molecule AMG-176 only significantly depleted HSPCs from
   RAS mutant CMML but not those from RAS wild-type CMML, a finding that
   supports the selective use of MCL1 inhibitors to treat patients with
   RAS pathway mutant CMML in whom BP occurs at the time of HMA therapy
   failure. These results are consistent with previous findings showing
   that CMML monocytes rely on MCL1, but not BCL2, for survival,[208]^48
   and that NRAS-mutant monocytic subclones that emerge at AML relapse
   depend on MCL1, not BCL2, for energy production.[209]^49 Consistent
   with this observation, our scRNA-seq analysis of BM MNCs from one
   representative patient with venetoclax-resistant disease confirmed that
   BCL2 inhibition cannot overcome the activation of NF-κB
   pathway-mediated inflammatory and survival mechanisms in HSPCs and
   downstream My/Mo populations.

   In conclusion, this study highlights the importance of dissecting how
   specific genetic drivers affect the cell-of-origin in cancer to gain
   mechanistic insights into therapy failure and, thereby, develop
   selective therapeutic approaches to halt disease progression. Given
   that the RAS pathway mutation-induced reprogramming of CMML cells is a
   multi-step process that affects multiple biological signaling pathways
   (e.g., inflammation, apoptosis, and immune escape) in distinct BM cell
   types, our findings also suggest that only combination therapies that
   simultaneously target these pathways could effectively overcome disease
   progression and prolong the survival of patients whose disease is
   resistant to current therapeutic approaches.

Limitations of the study

   In this study, we used 3′ RNA-seq by 10X Genomics, which evaluates RNA
   transcript expression levels for individual genes at the single-cell
   level but does not capture the entire RNA sequence, hence not allowing
   inference of the complete cDNA sequence and somatic mutation detection.
   Therefore, we were not able to correlate transcriptome to RAS pathway
   mutation status with single-cell resolution in all sequenced cases.
   Although we attempted to mitigate this intrinsic limitation to our
   sequencing technique by selecting samples with high RAS pathway mutant
   variant allele frequencies (VAFs), future studies will require the use
   of alternative single-cell sequencing technologies able to
   simultaneously capture genotype and transcriptome at the single-cell
   level to invariably characterize the specific features of RAS pathway
   mutant vs. wild-type cells. In an attempt to mitigate the impact of
   such a limitation in our identification of MCL1 upregulation as a
   preferential RAS mutant cell survival mechanism, we confirmed MCL1
   upregulation at the protein level by evaluating RAS pathway mutant
   samples with high VAFs. In addition, although we were able to confirm
   the selective sensitivity of RAS pathway mutant Lin^−CD34^+ cells to
   MCL1 inhibition at the time of progression after HMA therapy failure,
   there are inherent limitations to the extent to which these studies can
   capture the in vivo effects of MCL1 inhibition and how this could
   affect distinct cell types and functionalities. Finally, although we
   validated our transcriptomic findings related to cell-cell
   communication networks between key BM populations with multiplex
   immunofluorescence and immunophenotypic immune cell characterization,
   future studies will require deeper investigation and validation of
   these interactions.

STAR★Methods

Key resources table

   REAGENT or RESOURCE SOURCE IDENTIFIER
   Antibodies
     __________________________________________________________________

   Anti-human CD2 BD Biosciences Cat#: 555326; RRID: [210]AB_395733
   Anti-human CD3 BD Biosciences Cat#: 349201; RRID: [211]AB_400405
   Anti-human CD4 ThermoFisher Cat#: MHCD0401; RRID:[212]AB_10392546
   Anti-human CD7 BioLegend Cat#: 343104; RRID: [213]AB_1659216
   Anti-human CD11b ThermoFisher Cat#: 11-0118-42; RRID: [214]AB_1582242
   anti-human CD14 BD Biosciences Cat#: 347493; RRID: [215]AB_400311
   Anti-human CD19 BD Biosciences Cat#: 340409; RRID: [216]AB_400024
   Anti-human CD20 BD Biosciences Cat#: 555622; RRID: [217]AB_395988
   Anti-human CD33 ThermoFisher Cat#: 11-0337-42; RRID: [218]AB_1603221
   Anti-human CD56 BD Biosciences Cat#: 562794; RRID: [219]AB_2737799
   Anti-human CD235 BD Biosciences Cat#: 559943; RRID: [220]AB_397386
   Anti-human CD34 BD Biosciences Cat#: 562577; RRID: [221]AB_2687922
   Anti-human CD38 BioLegend Cat#: 303534; RRID: [222]AB_2561605
   Anti-human CD3 BioLegend Cat#: 317339; RRID: [223]AB_2563407
   Anti-human CD4 BioLegend Cat#: 300502; RRID: [224]AB_314070
   Anti-human CD8 BioLegend Cat#: 344710; RRID: [225]AB_2044010
   Anti-human CD56 BioLegend Cat#: 362544; RRID: [226]AB_2565922
   Anti-human CD16 BioLegend Cat#: 302028; RRID: [227]AB_893262
   Anti-Human IFN-γ BD Biosciences Cat#: 554702; RRID: [228]AB_398580
   Anti-human Perforin BioLegend Cat#: 308130; RRID: [229]AB_2687190
   Zombie UV BioLegend Cat#: 423107
   Protein Transport Inhibitor Cocktail eBioscience Cat#: 00-4980-93
   Fc Receptor Binding Inhibitor Polyclonal Antibody ThermoFisher Cat#:
   14-9161-73
   CD3e Cell Signaling Technology Cat#: 85061BF; RRID: [230]AB_2721019
   CD4 Abcam Cat#: ab133616; RRID: [231]AB_2750883
   CD8 Thermo Scientific Catalog#: MA5-13473; RRID: [232]AB_11000353
   CD14 Abcam Cat#: ab183322; RRID: [233]AB_2909463
   CD68, Clone PG-M1 Dako Cat#: M0876; RRID: [234]AB_2074844
   CD56 Dako Cat#: M7304; RRID: [235]AB_2750583
   DAPI Akoya Biosciences Cat#: FP1490; RRID: N/A
   MCL1 Cell Signaling Technology Cat#: 4572; RRID:[236]AB_2281980
   Vinculin Sigma-Aldrich Cat#: V9131; RRID:[237]AB_477629
     __________________________________________________________________

   Biological samples
     __________________________________________________________________

   Bone marrow aspirates from healthy donors AllCells (Alameda, CA) and MD
   Anderson’s Department of Stem Cell Transplantation N/A
   Bone marrow aspirates from CMML patients MD Anderson Bank N/A
   Human bone marrow-derived mesenchymal stem cells Cells provided by Dr.
   M. Andreeff N/A
     __________________________________________________________________

   Chemicals, peptides, and recombinant proteins
     __________________________________________________________________

   AMG-176 Med Chem Express Cat#: HY-101565
     __________________________________________________________________

   Critical commercial assays
     __________________________________________________________________

   Cytofix/Cytoperm kit BD Biosciences Cat#: 554714
     __________________________________________________________________

   Deposited data
     __________________________________________________________________

   scRNA-seq data This paper GEO: [238]GSE218390
   scATAC-seq data This paper GEO: [239]GSE218390
   scDNA-seq data This paper GEO: [240]GSE218390
     __________________________________________________________________

   Experimental models: Cell lines
     __________________________________________________________________

   K562 cells ATCC ATCC CCL243
     __________________________________________________________________

   Software and algorithms
     __________________________________________________________________

   SPSS 23.0 SPSS, Inc [241]https://www.ibm.com/products/spss-statistics
   R v4.1.2 R Core Team [242]https://www.r-project.org
   Seurat v4 R package ‘ Hao et al.[243]^50
   [244]https://github.com/satijalab/seurat
   CellphoneDB Vento-Tormo et al.[245]^51
   [246]https://www.cellphonedb.org/
   GraphPad Prism version 10.0 GraphPad software
   [247]https://www.graphpad.com/
   Spotfire TIBCO N/A
   FlowJo BD Biosciences [248]https://www.flowjo.com/
   GSEA Metascape [249]https://metascape.org/
   Spotfire TIBCO [250]https://www.spotfire.com/
     __________________________________________________________________

   Other
     __________________________________________________________________

   Ficoll-Paque PLUS ThermoFisher Cat#:45-001-752
   CD34 Microbead Kit Miltenyi Biotec Cat#:130-046-702
   NK Cell Isolation Kit Miltenyi Biotec Cat#: 130-092-657
   Opal 9 Kit Akoya Biosciences Cat#: NEL797001KT
   Qubit Thermo Fisher Scientific N/A
   [251]Open in a new tab

Resource availability

Lead contact

   Further information and requests for resources and reagents should be
   directed to and will be fulfilled by the lead contact, Simona Colla
   (scolla@mdanderson.org).

Materials availability

   This study did not generate new unique reagents.

Data and code availability

     * •
       scRNA-seq, scATAC-seq, and scDNA-seq data are accessible at GEO
       under accession number [252]GSE218390
       . No custom computer codes were generated in this study.
     * •
       The [253]lead contact can provide any additional information
       required to reanalyze the data reported in this work paper upon
       request.

Experimental model and study participant details

Primary human samples

   BM aspirates were obtained from patients with CMML who were seen in the
   Department of Leukemia at the University of Texas MD Anderson Cancer
   Center. Samples were obtained with the approval of the Institutional
   Review Board and in accordance with the Declaration of Helsinki. CMML
   diagnoses were assigned according to the World Health Organization
   criteria.[254]^3

   RAS pathway mutations were identified by targeted amplicon-based
   next-generation sequencing (NSG).[255]^52 Genomic DNA was extracted
   from whole BM aspirate samples and was subject to targeted PCR-based
   sequencing using an NGS platform evaluating a total of 81 genes, as
   previously described.[256]^52 This analysis was performed within the
   MDACC CLIA-certified Molecular Diagnostics Laboratory after informed
   consent (additional details in [257]supplemental information). For
   NGS-based analysis, the limit of detection for variant calling was 2%.
   Previously described somatic mutations registered at the Catalog of
   Somatic Mutations in Cancer (COSMIC:
   [258]http://cancer.sanger.ac.uk/cosmic) were considered potential
   driver mutations.

   All available samples carrying RAS pathway mutations were included in
   the study. Baseline BM aspirates were collected from patients before
   any treatment. Sequential BM samples were collected after HMA or
   venetoclax therapy failure. The clinical characteristics of the
   patients with RAS pathway mutated CMML are shown in [259]Tables S1 and
   [260]S3. BM samples from HDs were obtained from AllCells (Alameda, CA)
   and the Department of Stem Cell Transplantation at MD Anderson Cancer
   Center. Written informed consent was obtained from all donors.

   MNCs were collected from each BM sample immediately after BM aspiration
   using the standard gradient separation approach with Ficoll-Paque PLUS
   (catalog number #45-001-752, Thermo Fisher Scientific). MNCs were
   cryopreserved and stored in liquid nitrogen until they were used. For
   cell sorting applications, MNCs were enriched in CD34^+ cells using
   magnetic-activated cell sorting (MACS) with the CD34 Microbead Kit
   (catalog number #130-046-702, Miltenyi Biotec, Germany) and further
   purified by fluorescence-activated cell sorting (FACS) as described
   below.

Method Details

Clinical data analysis

   A clinical dataset of 108 CMML patients treated with HMA therapy at the
   Department of Leukemia at the University of Texas MD Anderson Cancer
   Center was evaluated to identify predictors of therapy outcomes. HMA
   therapy failure was defined as a lack of response (based on IWG 2006
   criteria) after at least 4 cycles of therapy or as relapse or
   progression after any number of cycles of therapy. Blast progression
   (BP) was defined as 1) the presence of >5% blasts in the BM at the time
   of primary HMA failure in patients with <5% blasts at baseline or an
   increase of at least 50% blasts in patients with 5–9% blasts at
   baseline; 2) BM blasts >20% or myeloid sarcoma regardless of primary or
   secondary failure; 3) BM blasts >5% at the time of secondary HMA
   failure (relapse or progression). Associations between gene mutations
   and BP were assessed using data from 108 patients whose samples were
   sequenced using the 81-gene panel; these analyses were performed at MD
   Anderson Cancer Center. Clinical datasets were analyzed using the SPSS
   23.0 (SPSS, Inc., Chicago, IL, USA) and R (version 3.5.1) statistical
   software programs. Logistical regression analysis was performed using
   clinical, cytogenetic, and molecular characteristics in correlation
   with responses to HMA therapy. The dataset was randomly divided into a
   training set (30 patients with BP) and a testing set (5 patients with
   BP). A combination rule derived from selected features was trained
   using logistic regression in the training set and a fixed model in the
   testing set. Receiver operating characteristic (ROC) curves were
   generated using the “pROC” package in R (version 3.6.0). The 95% CIs
   for the areas under the ROC curves were estimated using the DeLong
   method.[261]^53 The chi-square or Fisher exact test was used to analyze
   differences between categorical variables. Survival curves were
   generated using the Kaplan-Meier method and compared using log rank
   tests. Responses to HMA- or venetoclax-based therapies were evaluated
   based on the International Working group 2003[262]^54 and 2006[263]^55
   criteria for patients with secondary AML or CMML, respectively.

Flow cytometry analysis and fluorescence-activated cell sorting (FACS)

   Quantitative flow cytometry and FACS analyses of Lin^–CD34^+ cells were
   performed using previously described staining
   protocols[264]^56^,[265]^57 and antibodies against CD2, FITC, RPA-2.10,
   BD Biosciences, 555326; CD3, FITC, SK7, BD Biosciences, 349201; CD4,
   FITC, S3.5, Thermo Fisher, MHCD0401; CD7, FITC, 6B7, BioLegend, 343104;
   CD11b, FITC, ICRF44, Thermo Fisher, 11-0118-42; CD14, FITC, MφP9, BD
   Biosciences, 347493; CD19, FITC, SJ25C1, BD Biosciences, 340409; CD20,
   FITC, 2H7, BD Biosciences, 555622; CD33, FITC, P67.6, Thermo Fisher,
   11-0337-42; CD56, FITC, B159, BD Biosciences, 562794; CD235a, FITC,
   HIR2, BD Biosciences, 559943; CD34, BV421, 581, BD Biosciences, 562577;
   CD38, APC, HIT2, BioLegend, 303534, as we described previously.[266]^58

   Samples used for flow cytometry and FACS were acquired with a BD LSR
   Fortessa and a BD Influx Cell Sorter (BD Biosciences), respectively.
   The cell populations were analyzed using FlowJo software
   ([267]https://www.flowjo.com). All experiments included single-stained
   controls and were performed at the South Campus Flow Cytometry &
   Cellular Imaging Facility at MD Anderson Cancer Center.

Multiplex imaging assay

   BM core biopsies were used for multiplex immunofluorescence assessment.
   We optimized and validated a multiplex immunofluorescence panel using
   antibodies against CD3e, CD4, CD8, CD14, CD56, and CD68. Each antibody
   was assessed by multiplex immunofluorescence using the Opal 9 kit
   (catalog #NEL797001KT; Akoya Biosciences, Marlborough, MA), according
   to the following clones and dilutions: CD3e (clone D7A6E(AM), Cell
   Signaling Technology, 1:100), CD4 (clone EPR6855, Abcam, 1:200), CD8
   (clone C8/144B, Thermo Scientific, 1:25), CD14 (clone SP192, Abcam,
   1:100), CD56 (clone 123C3, Dako, 1:25), and CD68 (clone PG-M1, Dako,
   1:50). The slides were imaged using the Vectra Polaris spectral imaging
   system (Akoya Biosciences, Marlborough, MA) using the fluorescence
   protocol at 10 nm λ from 420 nm to 720 nm. Both germinal center and
   interfollicular areas from lymph nodes with reactive lymphoid
   hyperplasia were used as positive controls. Each marker was analyzed at
   the single-cell level, and a supervised algorithm for phenotyping was
   built for each marker. Cell density for each marker and combinations of
   phenotypes were consolidated using Spotfire software (TIBCO Spotfire).
   The nearest neighbor analysis was performed using R version 4.2.1.

Western blot

   BM CD34^+ cells were enriched from BM MNCs using magnetic sorting with
   the CD34 Microbead Kit (Miltenyi Biotec). Cells were resuspended in
   Mammalian Cell & Tissue Extraction Kit buffer (BioVision Incorporated)
   and incubated on ice for 10 min. Lysates were then collected after
   centrifugation at 12,000 rpm at 4°C for 20 min. The amount of protein
   was quantified using the Qubit Protein Assay Kit and a Qubit
   Fluorometer (Thermo Fisher Scientific). SDS-PAGE and Western blotting
   were performed following standard protocols. Blotted membranes were
   incubated with primary monoclonal antibodies against human MCL1
   (#4572S; 1:750 dilution; Cell Signaling Technology) and vinculin
   (hVIN-1; 1:2,000 dilution; Sigma-Aldrich). Membranes were developed
   using the SuperSignal West Pico PLUS Chemiluminescent Substrate (Thermo
   Fisher Scientific) in a KwikQuant Imager (Kindle Biosciences). Vinculin
   was used as a loading control, and lysates from the myeloma cell line
   JJN3 were used as positive controls.

T cell and NK cell cytokine secretion assays

   NK cells were isolated from BM MNCs obtained from HDs, and RAS pathway
   mutant or wild-type CMML patients by negative magnetic selection using
   the NK Cell Isolation Kit (Miltenyi Biotec). NK cells or BM MNCs were
   mixed with the human erythroleukemia cell line K562 at a
   target-to-effector ratio of 1:1. Cells were incubated for 4h at 37°C in
   5% CO[2] in the presence of a protein transport inhibitor cocktail
   (eBioscience 00-4980-93) for 4 h. After incubation, cells were
   harvested, washed with PBS, and stained with the viability dye Zombie
   UV. Cells were washed with PBS and resuspended in the presence of an Fc
   receptor-binding inhibitor antibody (ThermoFisher) for 20 min. NK cells
   and BM MNCs were stained with antibodies against CD3 (AF700,
   BioLegend). NK cells were further stained with antibodies against CD56
   (PE-Dazle 594) and CD16 (PerCP-Cy5.5), whereas BM MNCs were stained
   with antibodies against CD4 (PE-Dazzle 594) and CD8 (PerCP-Cy5.5).
   Cells were then washed, fixed, and permeabilized using the
   Cytofix/Cytoperm kit (BD Biosciences) and intracellularly stained with
   antibodies against IFN-γ (APC, BD Biosciences) and perforin (BV711,
   BioLegend). Samples were acquired with a BD Fortessa (BD Biosciences),
   and cell populations were analyzed using FlowJo software (version
   10.7.1, Ashland, OR).

scRNA-seq analysis and bioinformatic pipeline

   ScRNA-seq analysis was performed as we described previously.[268]^58
   Live Lin^–CD34^+ cells and live MNCs were isolated by FACS. Sample
   preparation and sequencing were performed at the Advanced Technology
   Genomics Core at MD Anderson Cancer Center. Sample concentration and
   cell suspension viability were evaluated using a Countess II FL
   Automated Cell Counter (Thermo Fisher Scientific). Samples were
   normalized for input onto the Chromium Single Cell A Chip Kit (10X
   Genomics), and single cells were lysed and barcoded for reverse
   transcription. Equal amounts of each uniquely indexed sample library
   were pooled together. Pooled libraries were sequenced using a
   NovaSeq6000 SP 100-cycle flow cell (Illumina). After sequencing, raw
   reads were aligned to the human genome (hg38), and the digital
   expression matrix was generated using cellranger count. Individual
   samples were merged to generate the digital expression matrix using
   cellranger aggr. The Seurat package in R was used to analyze the
   digital expression matrix. Cells with less than 100 genes and less than
   500 unique molecular identifiers detected were not analyzed further.
   The Seurat function NormalizeData was used to normalize the raw counts.
   Variable genes were identified using the FindVariableGenes function.
   The Seurat ScaleData function was used to scale and center expression
   values in the dataset for dimensional reduction. Default parameters
   were used for the Seurat functions. When needed, samples were
   integrated using the Seurat functions FindIntegrationAnchors and
   IntegrateData. Principal component analysis and Uniform Manifold
   Approximation and Projection (UMAP) were used to reduce the dimensions
   of the data, and the first 2 dimensions were used in plots. To cluster
   the cells and determine the marker genes for each cluster, we used the
   FindClusters and FindAllMarkers functions, respectively. Differential
   expression analysis of the samples was performed using the FindMarkers
   function and the Wilcoxon rank-sum test. The Benjamini-Hochberg
   procedure was applied to adjust the false discovery rate. Functional
   enrichment analysis was performed using the Metascape software
   ([269]https://metascape.org/gp/index.html#/main/step1).[270]^59 The
   human hallmark gene set was used. Analyses were performed using gene
   annotation available in 2020–2023.

   CellphoneDB (v2.0.0)[271]^17 was used to analyze the ligand–receptor
   interactions. Briefly, each cell type was separated by disease
   classification, and a separate run was performed for each disease
   classification. The connectome web was plotted using the igraph package
   in R.

scATAC-seq analysis and bioinformatic pipeline

   ScATAC-seq analysis was performed as we described previously.[272]^58
   The scATAC-seq Low Cell Input Nuclei Isolation protocol (10X Genomics)
   was used to isolate nuclei from FACS-purified cells. Extracted nuclei
   were used for the consecutive steps of the scATAC-seq library
   preparation protocol following 10X Genomics guidelines. Equal molar
   concentrations of uniquely indexed samples were pooled together. Pooled
   libraries were sequenced using a NextSeq500 150-cycle flow cell
   (Illumina). Reads were aligned to human (hg38) genomes, and peaks were
   called using the cellranger-atac count pipeline. Individual samples
   were merged using the cellranger-atac pipeline to generate the
   peak-barcode matrix and TF-barcode matrix. To identify specific TF
   activity for each cell cluster, we used the R package Seurat to analyze
   the TF-barcode matrix. The raw counts were normalized by the sequencing
   depth for each cell and scaled for each TF using the NormalizeData and
   ScaleData functions. Principal component analysis and UMAP were applied
   to reduce the dimensions of the data, and the first 2 dimensions were
   plotted. The FindClusters function was used to cluster the cells. The
   FindAllMarkers function was used to determine the TF markers for each
   cluster. Differential analysis of TF activity in the samples was
   performed using the FindMarkers function and the Wilcoxon rank-sum
   test. Cluster identity was determined based on the activity of master
   regulators of lineage commitment, as we[273]^60 and
   others[274]^33^,[275]^61 described previously. Cluster-specific peaks
   were determined using the FindAllMarkers function, and differentially
   accessible peaks between the samples were determined using the
   FindMarkers function. Each peak was associated with a specific gene
   based on its distance to that gene’s transcription start site (TSS).
   Peaks overlapping with a promoter region (−1,000 bp, +100 bp) of any
   TSS were annotated as peaks in promoters, whereas peaks not in promoter
   regions but within 200 kb of the closest TSS were annotated as peaks in
   the distal elements. Peaks not mapped in either the promoters or distal
   elements were annotated as peaks in intergenic regions.

scDNA and protein-seq analysis

   Simultaneous analyses of DNA mutations and the cell-surface
   immunophenotype (scDNA and protein-seq) were performed as we described
   previously[276]^62 and according to the Mission Bio protocol using the
   custom-designed 37-gene myeloid panel kit and 48 oligo-conjugated
   antibodies against all major BM cell types (Biolegend). Briefly,
   cryopreserved BM MNCs were thawed, quantified, and then stained with
   the pool of the oligo-conjugated antibodies. Stained cells were washed
   and loaded onto the Tapestri machine for single-cell encapsulation,
   lysis, and barcoding. DNA libraries were extracted from the droplets
   followed by the purification using Ampure XP beads (Beckman Coulter).
   Then, the supernatant was incubated with biotinylated oligonucleotides
   (Integrated DNA Technologies) to capture the antibody tags, followed by
   purification using streptavidin beads (Thermo Fisher Scientific).
   Purified DNA and antibody-tagged libraries were indexed and then
   sequenced on the Illumina NovaSeq 6000 or NextSeq 500 systems with
   150 bp paired-end multiplexed runs.

   The resulting files containing DNA and protein data were visualized
   using the Mission Bio Mosaic library version 1.8. Only manually curated
   and whitelisted variants were used. Variants were filtered using the
   below setting: min_dp = 5, min_gq = 0, min_vaf = 21, max_vaf = 100,
   min_prct_cells = 0, min_mut_prct_cells = 0, and min_std = 0. Protein
   reads were normalized by centered log ratio, and subsequently underwent
   dimensionality reduction and clustering using Mosaic ‘run_pca’
   (components = 15), ‘run_umap’ (attribute = 'pca', n_neighbors = 20,
   metric = 'cosine', min_dist = 0), and ‘cluster’ (attribute = 'umap',
   method = 'graph-community', k = 150). Default parameters were used
   unless otherwise specified, and randomness was controlled in all steps.
   Heatmaps were separately visualized in R using the ComplexHeatmap
   package
   ([277]https://bioconductor.org/packages/release/bioc/html/ComplexHeatma
   p.html).

Primary cell culture assays

   FACS-purified Lin^–CD34^+ HSPCs were resuspended in cytokine-free
   sterile RPMI medium supplemented with 10% FBS, 1%
   penicillin-streptomycin, and 0.1% amphotericin B and plated in 48-well
   plates previously seeded with low-passage (p ≤ 4) healthy BM-derived
   human mesenchymal cells. Co-cultures were incubated at 37°C in a 5%
   CO[2] atmosphere. After treatment with vehicle or AMG-176 (20 nM) for
   48 h, cells were harvested and stained for quantitative flow cytometric
   analysis using the antibody panel described above and with AccuCheck
   Counting Beads (Thermo Fisher Scientific) added to each tube.

Quantification and statistical analysis

   Statistical analysis of flow cytometry data was performed using Prism 8
   software ([278]https://www.graphpad.com). The figure legends include
   the statistical test(s) used in each experiment. Statistical
   significance was represented as ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001,
   and ∗∗∗∗p < 0.0001. In all analyses involving human samples,
   investigators were blinded to sample annotations and patient outcomes.
   For replicated experiments, the number of replicates is indicated in
   the figure legends. No statistical method was used to predetermine the
   sample size. No data were excluded from the analyses. The experiments
   were not randomized. Statistical analysis for blast progression and
   survival in the clinical cohort was performed as specified in the
   “[279]clinical data analysis” section ([280]method details).

Acknowledgments