CBFA2T3::GLIS2 pediatric acute megakaryoblastic leukemia is sensitive to BCL-X_L inhibition by navitoclax and DT2216

Acute megakaryoblastic leukemia (AMKL) is a developmentally restricted and early-onset cancer affecting children aged 0 to 4 years, including neonates,^1-4 pointing to a stem or progenitor cell of origin that is only transiently present in time.⁵ It represents ∼10% of pediatric acute myeloid leukemia (AML) cases and is a molecularly heterogenous and highly lethal leukemia, with dismal cure rates (<50%).⁶ Two-thirds of AMKL cases are driven by mutually exclusive chimeric fusion oncogenes including RBM15::MKL1,^7,8,KMT2A translocations,⁹ and cytogenetically cryptic rearrangements involving either NUP98^10,11 (NUP98r, specifically NUP98::KDM5A [N5A]¹² and NUP98::BPTF [NTF]¹³), HOX genes,^1,2,9 or the CBFA2T3::GLIS2 (CG2) fusion.^14-18 The mapping of functional dependencies of de novo pediatric AMKL is emerging,^19,20 and a repertoire of relevant preclinical models is needed to develop cell- and genotype-specific treatments for pediatric patients with leukemia.

CG2 and NUP98r AMKL are recognized as the highest relapse risk subgroups.^6,16,21 The CG2 fusion results from a cryptic inversion on chromosome 16, joining the N-terminal nervy homology regions (protein-interacting domains) of corepressor CBFA2T3 (ETO2) to the C-terminal zinc finger domains of transcription factor GLIS2, which allows for binding to GLI DNA consensus sequences.^14,15 CBFA2T3 is involved in hematopoietic stem cell maintenance and differentiation as well as in the regulation of erythro-megakaryocytic progenitors.^22-24 GLIS2 was primarily identified in adult kidneys, maintaining normal tissue architecture and function.²⁵ CG2-targeting treatments are actively explored, including Aurora A kinase inhibitors^14,26; interference with CG2 fusion oligomerization²⁷; and targeting of Neural Cell Adhesion Molecule 1 (NCAM1),¹⁷ the Folate Receptor Alpha (FOLR1),^18,28 or the JAK-STAT pathway with ruxolitinib.^12,29 

Resisting cell death is a long-recognized hallmark of cancer,³⁰ and agents that promote apoptosis entered the therapeutic arena for AML.³¹ The subsequent development of pharmacological inhibitors (BH3 mimetics) that target single or multiple prosurvival proteins^32,33 has sparked significant interest in oncology because these agents can induce mitochondrial apoptosis in multiple hematological malignancies,^34,35 with significant clinical responses,³⁶ including in children.^37,38 Evidence suggests that prosurvival factor dependency is both lineage-^39-42 and mutation–specific^43,44 in leukemia, and the integration of BH3 mimetics into current treatment protocols is under active investigation.

To define AMKL functional dependencies and overcome the paucity of pediatric patient samples, we have engineered CG2 AMKL models derived from cord blood (CB) hematopoietic stem and progenitor cells (HSPCs) that phenocopy the human disease. Across AMKL genotypes, CG2 leukemia distinctively upregulate the prosurvival factor B-cell lymphoma 2 (BCL2) yet are resistant to BCL2-specific inhibition with venetoclax or genetic knockdown (KD). In contrast, targeting the megakaryocytic lineage prosurvival factor BCL-X_L^45,46 with navitoclax⁴⁷ (binding to BCL2, BCL-X_L, and BCL-W) or BCL-X_L proteasomal degrader DT2216⁴⁸ induced apoptosis of CG2 and NUP98r AMKL cells, with minimal cross toxicity on normal CB CD34⁺ HSPCs. Our work highlights prosurvival protein BCL-X_L as a therapeutic vulnerability in pediatric AMKL and is timely aligned with a recent study suggesting that BCL-X_L is also a potential target of adult megaerythroid leukemia.⁴⁹ Addition of anti–BCL-X_L agents to cytarabine-based chemotherapy regimens in pediatric AMKL clinical trials is within reach and not exploited.

Extended methods are provided in supplemental Methods.

Pediatric AMKL samples (supplemental Table 1; and as previously reported¹²) were collected with approval from the research ethics boards of Centre Hospitalier Universitaire (CHU) Sainte-Justine, and informed patient consent in agreement with the Declaration of Helsinki. Biobanking was overseen by the Quebec Leukemia Cell Bank (BCLQ) (Montréal, Canada).

CG2 complementary DNA was amplified from the M07e cell line and cloned into the MNDU lentiviral expression vector^50,51 (supplemental Table 2). Umbilical CB units, collected with maternal consent, were distributed by Héma-Québec (Montréal, Canada) and manipulated as previously described.¹² 

Mouse experimental procedures were performed in accordance with the Canadian Council of Animal Care and with approval from the CHU Sainte-Justine research ethics board and animal institutional ethic committee (approval number 2020-2666). Xenotransplantation and characterization protocols are provided in the supplemental Methods.

Immunophenotyping antibodies are listed in supplemental Table 3. Apoptotic assays and intracellular staining are detailed in supplemental Methods.

Detailed protocols and sample descriptions are presented in supplemental Methods and supplemental Table 4.

RNA sequencing and data processing were performed at the Institute for Research in Immunology and Cancer’s Genomics Platform (Montréal, Canada) using the Illumina Nextseq500 with 150-cycle paired-end runs. Sequences were aligned to the reference human genome version GRCh38.

Single-cell suspensions were processed on the Chromium Controller according to the manufacturer’s protocol, and sequenced at the CHU Sainte-Justine Integrated Centre for pediatric clinical genomics (Montréal, Canada).

Synthetic models were maintained in optimized serum-free media. Navitoclax (ABT-263), venetoclax (ABT-199), DT2216, cytarabine and staurosporine were purchased from MedChemExpress (Monmouth Junction, NJ). Media composition, dose response curves, and functional assays are described in supplemental Methods.

Mice that underwent AMKL xenotransplantation were treated either daily via oral gavage with navitoclax (100 mg/kg; ABT-263⁴⁷), DT2216 intraperitoneal every 4 days (15 mg/kg),⁴⁸ or vehicle only. For drug combinations, mice were additionally treated with cytarabine (20 mg/kg, intraperitoneal) for the first 5 days. Detailed protocols are described in supplemental Methods.

To generate human leukemia models, we overexpressed CG2 in pooled CB CD34⁺ purified cells by transduction with a lentiviral construct carrying the fusion oncogene and a green fluorescent protein (GFP) reporter (Figure 1A). Gene transfer rates averaged 12.5% (Figure 1B), and 80% of each well was separately transplanted into 1 recipient NSG mouse. Leukemia penetrance was of 60%, with latencies between 9.9 and 36.1 weeks (Figure 1B). Expression of the CG2 fusion transcript was confirmed by reverse transcription polymerase chain reaction analysis (Figure 1C) and RNA sequencing of leukemic cells. Expression levels of the CG2 fusion transcript and wild-type CBFA2T3 were comparable between models and patient samples (supplemental Figure 8). At necropsy of leukemic mice, infiltration of hematopoietic organs by leukemic blasts was confirmed in the bone marrow (BM) and spleen (hCD45⁺GFP⁺ cells; Figure 1D-F; supplemental Figure 1; supplemental Table 5). No adverse outcomes were documented in the empty-vector control group (n = 10; Figure 1F-G). All 6 engineered leukemic models were able to robustly sustain serial transplantation with consistent disease phenotype and latencies (∼8-12 weeks), tested for up to 4 passages in vivo (Figure 1G; supplemental Figure 1; supplemental Table 5). As seen in patients with CG2, some mice exhibited neurological deficits of variable intensities indicative of leukemic paraspinal infiltration, which was confirmed by bioluminescent imaging of luciferase-transduced grafts (supplemental Figure 2).

The generated leukemia showed megakaryocytic morphology and lineage markers (hCD45⁺CD34^+/−CD41⁺CD61⁺) with significant CD56 (NCAM1) expression, a hallmark of CG2 AMKL^14,15,52 (Figure 1H-K; supplemental Figure 4; supplemental Table 5). The CG2-6 model was hCD45⁺CD71⁺CD117⁺ by immunophenotyping, with lower surface expression of CD41 (Figure 1E), CD34, CD33, and CD56 and absence of myelomonocytic markers (myeloperoxidase [MPO], CD68, lysozyme [LYZ], fucosyltransferase 4 [FUT4/CD15] and integrin subunit alpha M [ITGAM/CD11b]) by transcriptomic and flow cytometry (supplemental Table 5; supplemental Figures 3 and 4), pointing to an early megaerythroid progenitor state. AMKL development in vivo was retained after in vitro culture for 6 days, providing physiologically relevant cellular substrate for biochemical analyses (supplemental Figure 5).

AMKL leukemia initiating cell (LIC) frequency was assessed by limiting dilution assay using cells isolated from synthetic model of CG2 (mCG2-1, mCG2-2, and mCG2-6) recipients and approximated to 1 in 3445 cells, 1 in 14 938 cells, and 1 in 15 286 cells, respectively, close to the 1 in 10 300 LIC frequency assessed in a sample from a patient-derived NUP98r sample (Figure 1L-M; supplemental Table 6). The time to overt leukemia was proportional to the cell dose, with 4 of 8 mice of mCG2-1 succumbing to AMKL 20 to 30 weeks after injection of 250 cells (Figure 1L; supplemental Table 6). Overall, the engineered CG2 AMKL faithfully recapitulate human disease, reside in the megaerythroid differentiation space and harbor high LIC frequencies, as reported for patient samples.¹⁴ 

Transcriptomic profiling of the CG2 models closely correlate with genotype-matched patient samples (r = 0.64; n = 2, institutional cohort; n = 12, validation cohort⁹; Figure 2A), with high expression of NCAM1, BMP2 and ERG, and low levels of GATA1.^5,14,15,27 A CG2-specific gene signature of upregulated and downregulated markers was derived by comparing the expression profiles of CG2 models and patients (n = 10) against institutional NUP98r AMKL (n = 7) and CB CD34⁺ cells (n = 4), crossvalidated in a patient cohort comprising different AMKL genotypes⁹ (n = 73 total, with CG2 n = 12; Figure 2A; supplemental Table 7). CG2 AMKL is a non-HOXA/B, non-MEIS1/2 leukemia, which distinctively expresses GLIS2, whereas CBFA2T3 is expressed across AMKL genotypes (supplemental Figure 6B). Remarkably, CG2 AMKL upregulated prosurvival factor BCL2 (Figure 2A), pointing to a potential subtype-specific vulnerability. Furthermore, the engineered CG2 AMKL models cluster with patient-derived CG2 samples (n = 14) by expression profiling (Figure 2B) and apart from other AMKL genotypes and CB CD34⁺ cells. Principal component analysis revealed clustering of primary and tertiary xenografts based on expression profile (supplemental Figure 6A).

Akin to patient samples,^2,9,15 the genomic landscape of the engineered CG2 models was rather silent as assessed by whole-exome sequencing and comparative genomic hybridization (supplemental Figure 7), with no recurrent cooperating mutations, pathogenic single-nucleotide variants, or indels. Comparable with the reported genomic landscape of human disease,¹⁵ copy number variations were identified in 4 of 6 models, and chromosomal alterations remained stable over serial transplantation for 2 models analyzed (supplemental Figure 7). We excluded insertional mutagenesis events by long-read DNA sequencing (n = 4), and confirmed the (oligo)-clonal character of the engineered AMKL models (supplemental Figure 9). The CG2 AMKL models thus transcriptionally phenocopy human disease and appear genetically stable through serial transplantation.

We used a surface proteome (surfaceome) approach, as previously described,^12,53-55 to identify CG2-specific vs normal HSPC membrane proteins (CB CD34⁺, n = 11; CG2 AMKL, n = 2; and NUP98r AMKL, n = 2^12,13; Figure 2C; supplemental Figure 10; supplemental Table 9). Comparing proteomic data sets of CG2 and CB CD34⁺ cells confirmed that CG2 models express AMKL-associated markers CD151,⁵⁶ ITGA2B, PECAM1 (CD31)⁵⁷ and NCAM1 (CD56)^14,15,17 (supplemental Figure 10E). The 25 most abundant and differentially expressed proteins in CG2 vs CB CD34⁺ were crossvalidated against an external transcriptomic data set from patients with AMKL⁹ (Figure 2C; supplemental Figure 11B). Eight of these markers (CSF2RA, DPP10, NCAM1, NRP1, PCDH10, PCDH19, RECK, and SCN9A) were significantly associated with the CG2 genotype (highlighted in Figure 2D; supplemental Figure 11B). Five of the 25 surfaceome proteins could be validated using available flow cytometry antibodies, showing that PCDH19 and JAG1 were detected on a subset of cells across all patients and models, whereas CSF2RA, NRP1, and INSR were only detected in a subset of samples (Figure 2E; supplemental Figure 12). Similarly, the FOLR1 receptor, recently reported to be associated with CG2 AMKL,¹⁸ was found to be expressed on a significant proportion of cells across models, except for mCG2-6. Of note, the CD41^low mCG2-6 cells express CD31/CD151 but upregulate CD41/CD61 in culture (supplemental Figure 4), pointing to immature leukemic cells that are primed toward the megakaryocytic lineage. Pairwise correlation analyses of cell surface markers most associated with CG2 uncovered that the combination of NCAM1 with either PCDH10 or PCDH19 was most predictive of the CG2 genotype (Figure 2F; supplemental Figure 13; supplemental Table 10; P < .05). These markers can contribute to refine the panel of markers used for CG2 AMKL diagnosis and tracking of minimal residual disease in patients.⁵⁸ 

Engineered CG2 models (n = 4) and patient-derived samples (n = 2) were profiled by single-cell RNA sequencing to determine their cellular composition (Figure 3; supplemental Figures 14-17). Key upregulated genes of normal megakaryocyte-erythroid progenitors (MEPs), megakaryocyte progenitors (MKPs), and platelets were inferred from normal BM cell populations (Human Cell Atlas [HCA]⁵⁹; Figure 3A-B; supplemental Table 11). MEP/MKP/platelet-specific genes, for example ITGA2B (CD41), are universally expressed in CG2 AMKL cells but not in adult AML cells⁶⁰ (Figure 3C; supplemental Figures 15 and 16). The MKP-specific genes GP1BA (glycoprotein 1B platelet subunit α or CD42b) and GP9 (glycoprotein IX) are more expressed in CG2 than in healthy MEP cells, indicating that CG2 leukemic cells lie toward the primitive megakaryocytic differentiation space. Transferring the cell type assignment from the HCA to AMKL and AML data sets, confirmed the dominant MKP bias of CG2 models and patients compared with other AML samples (gray shaded samples in Figure 3D, top; supplemental Figure 14B-C). CG2 AMKL samples contain varying MEP proportions (Figure 3D, bottom), with the CG2-6 model partially projecting into the erythroid space (Figure 3D, blue shading in radar plot and CD34⁺ erythroid progenitors [ERP] fractions in the bottom panel). Although MKP genes are homogeneously expressed in CG2 samples, only small subsets express genes suggesting terminal differentiation such as CLEC1B (Figure 3E; supplemental Figures 15 and 16). These data indicate that CG2 models reside predominantly in the MKP differentiation space and recapitulate the lineage commitment in patients with CG2 AMKL.

Coembedding the data sets in the same uniform manifold approximation and projection (UMAP) revealed that CG2 models and patient samples lie in an area of the UMAP representation also populated by healthy BM cells (HCA) of the megakaryocytic lineage (supplemental Figure 17A), highlighting the low intersample heterogeneity, in contrast to the large cell heterogeneity of normal BM, as seen by the distribution overlap of the CG2 models and patient-derived samples (supplemental Figure 17B).

Furthermore, gene expression of leukemic CG2 MKP was compared with that of healthy MKPs to identify aberrantly expressed genes, such as NCAM1 and BCL2 (Figure 3F-G; supplemental Table 12), mirroring the CG2 bulk transcriptomic signature (Figure 2A). Although no other BCL2 family member was aberrantly expressed in CG2 MKP, BCL2L1 (BCL-X_L) is specifically expressed in normal MEP/MKP/platelet subsets as compared with other cell types (adjusted P = 7.5 × 10⁻¹¹¹; Figure 3B; supplemental Figure 14E), and BCL2L1 expression is maintained across CG2 models and patient-derived samples (Figure 3G; supplemental Figures 15 and 16). This provides a rationale to assess the functional dependency of CG2 cells to BCL-X_L, based on lineage ontogeny, and to BCL2, given its aberrant expression in CG2 AMKL.

We tested the sensitivity of AMKL cells to BCL2 inhibitor venetoclax (CG2, n = 6; N5A, n = 2¹²; and patient-derived xenograft [pdx]NTF, n = 1¹³) as well as in AML samples (KMT2A::MLLT3 [MLLr], n = 1; NUP98::NSD1 [NNSD1], n = 1; and N5A monocytic [CD68⁺LYZ⁺] AML models, n = 2¹²). All AMKL samples tested were resistant to venetoclax, with half-maximal inhibitory concentrations (IC₅₀s) of >10 μM, whereas AML samples were sensitive to venetoclax (Figure 4A; supplemental Table 13). In contrast, both CG2 and NUP98r AMKL were sensitive to navitoclax, which has a broader affinity to BCL2, BCL-X_L, and BCL-W as well as AML samples (Figure 4B; supplemental Table 13), whereas genotype-matched monocytic N5A AML cells were resistant, suggesting a lineage-specific vulnerability. AMKL cells exposed to navitoclax for 72 hours at concentrations spanning the leukemia-specific IC₅₀ values exhibited higher levels of apoptosis than control CB CD34⁺ cells (Figure 4C). In contrast, exposure to venetoclax did not induce apoptosis in AMKL cells at doses up to 10 μM (Figure 4D). Levels of mitochondrial superoxide radicals and loss of mitochondrial potential were higher in AMKL cells treated with navitoclax than in cells treated with venetoclax (Figure 4E-F; supplemental Figure 18A), suggesting mitochondrial dysfunction. These results imply that AMKL cells are sensitive to BCL2/BCLX_L/BCL-W inhibitor navitoclax with subsequent induction of mitochondrial apoptosis yet resistant to venetoclax that uniquely targets BCL2.

To dissect whether the induction of apoptosis in AMKL cells by navitoclax is mediated via BCL2, BCL-X_L, or BCL-W inhibition, expression levels of each gene were assessed using transcriptomic data sets. In accordance with the CG2 signature (Figure 2A), BCL2 was most abundant in CG2 compared with NUP98r AMKL cells (institutional cohort, Figure 5A) and all other AML genotypes (CG2, n = 7; patients total, validation cohort, n = 73⁹; Figure 5B). In contrast, BCL2L1 (BCL-X_L) is homogenously expressed across AMKL genotypes and at much lower levels in the N5A monocytic AML (n = 3). Intracellular protein levels of BCL-X_L were correspondingly increased in CG2 and NUP98r AMKL as compared with normal CB CD34⁺ cells, whereas the BCL2 protein was only detected in the CG2 genotype, matching gene expression (Figure 5C). Short hairpin(sh)–mediated KD against BCL2, BCL2L1 or BCL2L2 (BCL-W) was used to assess the functional impact of reduced prosurvival factors in 2 distinct CG2 models (gene transfer of >95% GFP⁺mCherry⁺ cells; Figure 5D-E; supplemental Figure 18B-C), with confirmation of efficient protein reduction of BCL2 and BCL-X_L (Figure 5F; supplemental Figure 18D). Increased apoptosis of AMKL cells (annexin V–positive cells; Figure 5D; supplemental Figure 18E) was observed only with KD of BCL-X_L, but not BCL2 or BCL-W. Additionally, AMKL and monocytic N5A AML models were treated with DT2216,⁴⁸ a selective BCL-X_L proteolysis-targeting chimera that targets BCL-X_L to the Von Hippel-Lindau E3 ligase (VHL) for degradation. VHL is abundantly expressed in our models of AMKL (Figure 5A) but not expressed in platelets, which are most vulnerable to BCL-X_L inhibition.⁴⁸ In a dose-response experiment, AMKL models were sensitive to DT2216, with IC₅₀s < 200 nM (Figure 5G; supplemental Table 13). In contrast, monocytic N5A AML models demonstrated resistance to DT2216 treatment, with IC₅₀ values > 10 μM (Figure 5G; supplemental Table 13). AMKL cells exposed to 100 nM and 1 μM of DT2216 showed significantly higher levels of apoptosis compared with normal CB CD34⁺ cells (Figure 5H) and higher mitochondrial stress, shown by increased superoxide levels and loss of mitochondrial membrane potential (supplemental Figure 18F-G). Furthermore, testing navitoclax, venetoclax, and DT2216 on several independent CB CD34⁺ pools revealed sample-dependent variability in drug response but overall lower toxicity with DT2216 in comparison to navitoclax (supplemental Figure 18H; supplemental Table 13). These results suggest that the sensitivity of AMKL cells toward navitoclax is due to the inhibition of BCL-X_L, as demonstrated by genetic and pharmacological inhibition.

To investigate the in vivo impact of BCL-X_L inhibition of navitoclax or DT2216 as single agent, 3 independent models of AMKL were tested (Figure 6A; supplemental Figures 19 and 20) as well as a chemotherapy refractory PDX of NTF¹³ (supplemental Figures 19 and 20). After a 3-week treatment cycle with navitoclax, a significant reduction of leukemic infiltration was noted in mCG2-1, as assessed based on spleen weight (Figure 6B) and abundance of GFP⁺hCD45⁺ AMKL cells in the blood, BM, and spleen (Figure 6C-D). Normal hematopoietic cell distribution and tissue architecture was restored in the BM and spleen of navitoclax-treated mice vs control mice (Figure 6E). Furthermore, navitoclax-treated mice showed reduced circulating blasts during treatment (week 5) as well as 1 week after treatment (week 7; Figure 6F), conferring a survival benefit over vehicle-treated mice (Figure 6G). For mCG2-2, treatment was stopped early because of rapid onset of hindleg paralysis. Nevertheless, a significant reduction of overall leukemic burden was detected after 2 weeks of treatment, measured by luciferase intensity (supplemental Figure 19A-B) as well as reduction of GFP⁺Ametrine⁺ cells in the spleen in comparison with vehicle controls (supplemental Figure 19C). Using in vivo imaging, reduced leukemic progression was seen in mice with pdxNTF xenografts treated with navitoclax, which was significantly lower than in vehicle-treated control mice (week 8; supplemental Figure 19D-E).

Furthermore, we tested DT2216 in 2 distinct models of CG2 (mCG2-1 and mCG2-6), presenting either as megakaryopoietic or megaerythroid leukemia, and in pdxNTF. Mice with xenografts were treated with DT2216 every 4 days for a total of 4 to 6 weeks. Treatment with DT2216 demonstrated reduced leukemic blasts in the blood (GFP⁺hCD45⁺ cells; Figure 6H) and significantly prolonged the survival of mice that received mCG2-1 transplantation in comparison with vehicle controls (Figure 6I). Mice that received CG2-6 transplantation were investigated by BM aspiration after 6 weeks of treatment and showed significant reduction of hCD45⁺GFP⁺CD31⁺CD151⁺ leukemic cells in the BM in comparison with vehicle controls (supplemental Figure 20A-B). Additionally, a 4-week treatment cycle of DT2216 significantly reduced the leukemic burden in mice with pdxNTF xenografts (supplemental Figure 20C). This work demonstrates that BCL-X_L degradation either with navitoclax or DT2216 could be exploited as a strategy in the treatment of high-fatality CG2 and NUP98r AMKL.

Cytarabine is a standard-of-care component of chemotherapy induction and intensification cycles for patients with AML. In this context, combinatorial treatment with BCL-X_L inhibitors and cytarabine was tested in CG2 AMKL. First, all CG2 models showed sensitivity toward cytarabine in vitro with IC₅₀s ranging from 1 to 7 nM (Figure 7A; supplemental Table 13). To assess the in vivo antileukemia activity of either navitoclax or DT2216 combined with cytarabine, NSG mice received a xenograft of mCG2-1 and were treated with low-dose cytarabine (20 mg/kg) for 5 consecutive days alongside navitoclax or DT2216 (Figure 7B; supplemental Figure 21A). Engraftment was verified for all groups before treatment by subjecting them to bleeding 3 weeks after transplantation (supplemental Figure 21B-C). At the end point, mice treated with the combination of cytarabine and navitoclax showed significantly lower leukemic infiltration in the blood, BM, and spleen as compared with mice treated with single agent alone or vehicle control (Figure 7C). In monotherapy, navitoclax or cytarabine performed equivalently in terms of leukemic burden reduction (Figure 7C). The same in vivo response was observed after combinatorial treatment with DT2216 and cytarabine (supplemental Figure 21D). Of note, although navitoclax shows a steeper reduction of leukemic burden as a single agent in comparison with DT2216, treatment with either navitoclax or DT2216 combined with cytarabine was equally effective against CG2 AMKL (Figure 7D-E; supplemental Figure 21F-E). These results highlight the translational potential of combining standard-of-care cytarabine with BCL-X_L inhibitors in the treatment of high-fatality pediatric AMKL.

We present 6 human models of CG2 AMKL that can sustain serial transplantation and phenocopy the clinical features, expression signature, genomic landscape, cellular ontogeny, and high LIC frequencies of the disease. Progress in the field is hampered by limited patient samples currently available for preclinical testing, a restricted pool of representative cell lines, and limited human models. Synthetic CG2 models contribute to map functional dependencies and capture the cellular and molecular heterogeneity of the leukemia, maturation block, concomitant genetic alterations, and surface marker expression. They represent faithful and scalable biomasses for biochemical studies, such as proteomics, to uncover new CG2 biomarkers that are either megaerythroid-lineage specific (CD151 and PECAM1), distinct from normal CB progenitors (JAG1), or most predictive of the CG2 AMKL genotype (NCAM1, PCDH10, and PCDH19), which can contribute to the optimization of flow cytometry antibody panels for CG2 AMKL diagnosis or minimal residual disease tracking.

Cell- and genotype-specific therapies are needed for pediatric AMKL to improve patient survival. Our models of CG2 highlight prosurvival factor BCL-X_L as a novel therapeutic target of high-fatality AMKL. In vitro pharmacological inhibition of BCL-X_L with navitoclax or DT2216 induces apoptosis in all CG2 AMKL models, accompanied by mitochondrial dysfunction. This was recapitulated by genetic KD of BCL-X_L but not BCL2 or BCL-W, suggesting that CG2 AMKL depend preferentially on BCL-X_L to resist apoptosis. Similarly, NUP98r AMKL cells (n = 3) were also sensitive to navitoclax or DT2216 treatment, whereas genotype-matched monocytic AML cells were resistant, pointing to a lineage-specific dependency. Indeed, prosurvival factor dependencies are cell context–specific, with the megakaryocytic lineage relying predominantly on BCL-X_L^45,61,62 for proper maturation.

Importantly, after in vivo treatment with either navitoclax or DT2216, we note a reduction in tumor burden, and a survival benefit in leukemic mice, using 3 CG2 models (mCG2-1, -2, and -6) and a PDX of NUP98r AMKL, the 2 highest fatality AMKL subgroups. The mCG2-6 model resides in a distinct megaerythroid differentiation space (CD41^lowCD31⁺CD151⁺), reflecting the cellular heterogeneity of CG2 AMKL, yet demonstrates a common BCL-X_L dependency across this cellular context. Our findings are aligned with a recent publication reporting that adult leukemia of the megaerythrocytic lineage is sensitive to BCL-X_L–specific inhibitor A-1331852, including the HEL erythrocytic line in vivo.⁴⁹ 

Expression of BCL2 on its own is not considered a predictor of drug response to venetoclax⁶³ and may explain the relative resistance of CG2 leukemic cells to BCL2 inhibition, as suggested by our work and that of others.^64,65 This is an important finding, because venetoclax (potent BCL2 inhibitor) is gaining unprecedented momentum for leukemia treatment,^37,38 including in children. Given that CG2 cells do express higher baseline BCL2 levels than NUP98r AMKL and normal MEP/MKPs, combinatorial strategies that trigger apoptosis such as cytotoxic or targeted chemotherapy, in addition to BCL2 inhibition, may ultimately induce apoptosis in CG2 cells that express BCL2, as recently suggested with either azacytidine⁶⁶ or prosurvival protein myeloid cell leukemia 1 (MCL1) inhibitor S63845.⁶⁴ 

Moving forward, our work demonstrates that combining BCL-X_L inhibitors (navitoclax or DT2216) and low-dose cytarabine is more potent against AMKL in vivo than either single agent alone, with significant translational potential given the current use of venetoclax and high-dose cytarabine in the treatment of relapsed or refractory pediatric AML.^37,66,67 However, thrombocytopenia is a dose-limiting toxicity of navitoclax due to BCL-X_L inhibition in the megakaryocytic lineage.⁶⁸ Platelets express BCL-X_L at higher levels than MEP/MKPs (supplemental Figure 16), in line with the rapid drop in platelets observed in patients exposed to navitoclax.⁶⁹ The BCL-X_L degrader DT2216, developed to mitigate the on-target thrombocytopenia of navitoclax, interacts with the Von Hippel-Lindau E3 ligase, which is weakly expressed in platelets. When gauging treatment toxicity in normal cells, we found that DT2216 was less toxic than navitoclax to CB CD34⁺ cells, underscoring the translational potential of BCL-X_L degraders for treatment of hematological malignancies of the megaerythrocytic lineages. Although monotherapy with navitoclax may seem slightly more effective than DT2216 against AMKL in vivo (Figure 7D), the DT2216 intermittent dosing scheme can be further optimized in future studies, and, importantly, both agents are equally effective in combination with cytarabine. Altogether, these preclinical studies suggest that integration of selective and potent BCL-X_L inhibitors to existing treatment regiments may improve disease eradication in pediatric AMKL.

The authors thank the patients and their families and members of the pediatric Hematology-Oncology Team of CHU Sainte-Justine for their outstanding support; the members of the Quebec Leukemia Cell Bank (BCLQ) and the Hematology-Oncology Division of CHU Sainte-Justine for their contribution to the biobanking initiative; Elie Haddad, Kathie Béland, and the members of the humanized mouse core facility (CHU Sainte-Justine); Ines Boufaied of the flow cytometry facility (CHU Sainte-Justine) for cell sorting; Virginie Saillour, Charles Privé, and René Allard of the Integrated Centre for pediatric clinical genomics at CHU Sainte-Justine for whole-exome sequencing and single-cell RNA sequencing studies; France Léveillé, Frédérique Tihy, Géraldine Mathonnet, and Emmanuelle Lemyre of the Medical Genetics Service at CHU Sainte-Justine for comparative genomic hybridization analysis; Elke Küster-Schöck of the microscopy facility at CHU Sainte-Justine; Raphaëlle Lambert from the Genomics Core Facility (Institute for Research in Immunology and Cancer) for quantitative reverse transcription polymerase chain reaction analysis and RNA sequencing analysis; Simon Mathien and Jean Duchaine from the High Throughput Screening Core Facility (Institute for Research in Immunology and Cancer) for dose-response experiments; Marianne Issac from the Histology Platform (Institute for Research in Immunology and Cancer) for processing of histological samples; Carolina Marmolejo, Anne-Cécile Soufflet, Emma Rose Cheetham, Nadia Emely Chauca Torres, and Françoise Couture for their help with experiments; Sami Ayachi for bioinformatic analysis; Keith Humphries, Donald B. Kohn, and Connie Eaves for providing lentiviral vectors; and Michael A. Lieberman for offering the CHRF-288-11 cell line.

This study was supported by seminal Cole Foundation transition awards (S. Cellot and V.-P.L.), Terry Fox Research Institute new investigator grants, a Canadian Cancer Society Research Institute /Cole Foundation impact grant (S. Cellot and B.T.W.), an operating grant from the Leukemia and Lymphoma Society of Canada (S. Cellot), an operating grant from the Richard and Edith Strauss Foundation (S. Cellot and B.T.W.), an operating grant from the Canadian Institutes of Health Research (S. Cellot and H.B.) (FRN 178326), funding from the CHU Sainte-Justine Foundation (Fonds d’Innovation Thérapeutique), the Fondation Charles-Bruneau (operating grants), and the Canada Foundation for Innovation (John R. Evans Leaders Fund) (S. Cellot). This research work was made possible thanks to the financial support of the Oncopole, which receives funding from Merck Canada Inc, the Fonds de recherche du Québec-Santé (FRQS), Génome Québec, the Cancer Research Society, and the Institute for Research in Immunology and Cancer-Commercialization of Research (B.T.W., F.B., and S. Cellot). V.P.L. is supported by a FRQS Junior 1 Clinical Research Scholarship and by the CHU Sainte-Justine and Charles-Bruneau Foundations. S. Cellot is recipient of a Clinician scientist senior award from the FRQS. V.G., M.R., B.K., and L.T. are recipients of Cole Foundation fellowships. N.E.-H. is supported by a postdoctoral scholarship from the Institute for Data Valorization. The Quebec Leukemia Cell Bank is supported by the Cancer Research Network of the FRQS. The humanized mouse core and the flow cytometry facilities at CHU Sainte-Justine are supported by the Fondation Charles-Bruneau.

Contribution: V.G. and M.R. designed and performed the experiments, analyzed the data, generated the figures, and wrote the manuscript; M.B. and S. Cardin led projects, performed experiments, analyzed data, and wrote the manuscript; L.B., L.L., and F.F. assisted with experiments and data collection; N.E.-H. performed bioinformatic analysis, generated figures, and wrote the manuscript; A.R., E.A., P.G., L.J., and S.J. performed bioinformatic analysis; V.L., B.K., and A.F. performed and analyzed single-cell RNA sequencing and wrote the manuscript; V.-P.L. supervised bioinformatics and single-cell experiments and analyses and wrote the corresponding manuscript sections; L.T., L.A., and P.P.R. performed and analyzed the cell surface proteomic experiments; E.B. performed mass spectrometry analysis; B.P. and S.M.S. performed nanopore genome sequencing and bioinformatic analysis under the supervision of M.A.S.; T.A.G. shared the AMKL validation cohort data set; T.H.T., R.S., M.D., H.B., P. Teira, V.-P.L., and S. Cellot oversaw patients and contributed to the biobanking; J.H. supervised the biobanking of patient specimens at the BCLQ; J.C., G.S., F.B., B.T.W., P. Thibault, and H.B. provided guidance in designing the experiments and analyses; S. Cellot conceived the experiments, supervised the biobanking of patient specimens at the BCLQ, and wrote the manuscript; and all authors reviewed the manuscript.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Sonia Cellot, Pediatric Hematology-Oncology Division, Centre Hospitalier Universitaire (CHU) Sainte-Justine Research Center, 3175 Côte Sainte-Catherine, Room 6-17-007, Montréal, QC H3T 1C5, Canada; email: sonia.cellot@umontreal.ca.