Research Article

Mapping the cellular origin and early evolution of leukemia in Down syndrome

See allHide authors and affiliations

Science  09 Jul 2021:
Vol. 373, Issue 6551, eabf6202
DOI: 10.1126/science.abf6202

Down with leukemia

Down syndrome is a congenital disorder caused by the trisomy of chromosome 21, and it is associated with a greatly increased risk of leukemia with origins in fetal development. Infants with Down syndrome are often born with a preleukemic condition, which later resolves in most cases. By using gene-edited human cells implanted into mouse models, Wagenblast et al. recapitulated the development of preleukemia and leukemia in the context of Down syndrome (see the Perspective by Roberts and Vyas). A specific mutation triggered a preleukemic condition in the context of trisomy 21 as expected, but progression to full-blown leukemia required a different genetic path and was not dependent on trisomy 21.

Science, abf6202, this issue p. eabf6202; see also abj3957, p. 155

Structured Abstract

INTRODUCTION

Leukemia is the most common cancer in children, with the first genetic alterations often occurring during fetal development. These initiating events generate preleukemic cells, which are the evolutionary ancestors of leukemia that arises after birth. Because of our inability to directly access human fetal preleukemia, the identity of the cell of origin and the steps of leukemia evolution remain largely unknown. Down syndrome leukemogenesis represents a disease setting to study human preleukemia and the evolutionary steps that lead to fully transformed leukemia. Up to 30% of children with Down syndrome (trisomy 21) exhibit a preleukemic transient abnormal myelopoiesis (TAM) and, overall, have a 150-fold increased risk of developing myeloid leukemia within the first 5 years of life. However, the mechanism by which an extra copy of chromosome 21 predisposes to preleukemia and leukemia remains unclear.

RATIONALE

Understanding Down syndrome leukemogenesis requires a humanized model that faithfully recapitulates the full developmental spectrum of premalignant and malignant stages of Down syndrome leukemia. Using CRISPR/Cas9–mediated gene editing in human disomic and trisomic fetal liver–derived hematopoietic stem and progenitor cells and xenotransplantation, we developed a model with which to characterize the genetic events and cellular contexts underlying the preleukemic and leukemic phases of Down syndrome leukemogenesis.

RESULTS

Trisomy 21 hematopoietic stem and progenitor cells (HSPCs) showed reduced proliferation in vitro and generated smaller grafts in xenotransplanted mice, with reduced serial transplant ability, as compared with that of disomic HSPCs. Preleukemia was initiated in trisomy 21, but not disomic, long-term hematopoietic stem cells (LT-HSCs) when mutations in the erythroid-megakaryocyte transcription factor GATA binding protein 1 (GATA1) were introduced, which led to exclusive expression of the short isoform (GATA1s). Subsequent leukemic progression could occur in multiple stem and progenitor populations, was independent of trisomy 21, and induced through deletion of cohesin genes, including STAG2 (STAG2ko). Serial engraftment in mice showed that GATA1s-induced preleukemia underwent spontaneous resolution, which contrasted with the persistent ability of the GATA1s/STAG2ko–induced leukemia to engraft serially in mice. Leukemic progression was developmentally restricted to fetal and early postnatal stages; adult-derived bone marrow HSPCs were unable to undergo GATA1s/STAG2ko-induced leukemic transformation. We identified a molecular mechanism by which three chromosome 21 microRNAs (miRNAs) contributed to the predisposition toward preleukemia initiation. Simultaneous overexpression of miR-99a, miR-125b-2, and miR-155 in normal disomic LT-HSCs recapitulated a trisomy 21–like hematopoietic state, as assessed through comparable lineage differentiation, reduced self-renewal capacity, and similar gene expression and open chromatin accessibility profile. Removal of these miRNAs in trisomy 21 LT-HSCs inhibited GATA1s-induced preleukemia development. Using secondary xenotransplantations of defined cell populations, we identified CD117+/KIT proto-oncogene (KIT) as a marker of disease-driving cells. Pharmacological KIT inhibition targeted preleukemic stem cells, both in GATA1s-induced preleukemia and in primary Down syndrome preleukemia patient samples.

CONCLUSION

Collectively, our results provide insight into how human preleukemia and leukemia evolve in fetal life and early childhood. We were able to identify distinct cellular origins and effects of trisomy 21 for preleukemia initiation and leukemia progression. Predisposition to preleukemia in Down syndrome is affected by overexpression of distinct chromosome 21 miRNAs, specifically in the preleukemic LT-HSC cell of origin. Our study reveals the relevance of the cellular and developmental status of the cell of origin during leukemogenesis, which begins to explain why genetic drivers can be distinct between pediatric and adult acute myeloid leukemia. KIT inhibitors targeted preleukemic stem cells, providing proof of principle for early prevention strategies in childhood leukemia that may be able to inhibit leukemia progression, and these results encourage further preclinical and clinical assessment.

Cell of origin in Down syndrome leukemogenesis.

Down syndrome preleukemia originated in long-term hematopoietic stem cells (LT-HSCs) through mutations in GATA1, leading to the expression of the short isoform GATA1s. Progression toward leukemia occurred in various stem and progenitor cells through mutations in cohesin factors such as STAG2. Predisposition to preleukemia was affected by chromosome 21 miRNAs, and pharmacological inhibition of KIT targeted preleukemic stem cells.

Abstract

Children with Down syndrome have a 150-fold increased risk of developing myeloid leukemia, but the mechanism of predisposition is unclear. Because Down syndrome leukemogenesis initiates during fetal development, we characterized the cellular and developmental context of preleukemic initiation and leukemic progression using gene editing in human disomic and trisomic fetal hematopoietic cells and xenotransplantation. GATA binding protein 1 (GATA1) mutations caused transient preleukemia when introduced into trisomy 21 long-term hematopoietic stem cells, where a subset of chromosome 21 microRNAs affected predisposition to preleukemia. By contrast, progression to leukemia was independent of trisomy 21 and originated in various stem and progenitor cells through additional mutations in cohesin genes. CD117+/KIT proto-oncogene (KIT) cells mediated the propagation of preleukemia and leukemia, and KIT inhibition targeted preleukemic stem cells.

Children with Down syndrome [trisomy 21 (T21)] have a 150-fold increased risk of developing acute myeloid leukemia, which is called myeloid leukemia associated with Down syndrome (ML-DS), in the first 5 years of life (1). However, the mechanism by which an extra copy of chromosome 21 (Chr21) predisposes and cooperates with genetic events in Down syndrome leukemogenesis is not known. In pediatric leukemia, the initiating genetic events occur before birth and generate preleukemic cells (2), which are the evolutionary ancestors of leukemia that arises after birth. However, characterizing human fetal preleukemia is challenging because of our inability to directly access it, rendering the identity of the cell of origin and the steps of leukemia evolution largely unknown. Up to 30% of newborns with Down syndrome exhibit transient abnormal myelopoiesis (TAM), a preleukemic phase characterized by a clonal proliferation of immature myeloid cells (mostly megakaryoblasts) that carry somatic mutations in the erythroid-megakaryocyte transcription factor GATA binding protein 1 (GATA1) (fig. S1A) (35). Mutations in GATA1 occur in utero, beginning at 21 weeks of gestation (6, 7), and lead to the expression of a truncated isoform [GATA1-short (GATA1s)]. The preleukemia resolves spontaneously in the majority of newborns; however, in 20% of cases ML-DS evolves within 4 years from the GATA1s-mutated preleukemic clone through acquisition of additional mutations, predominantly in genes of the cohesin complex or CCCTC Binding Factor (CTCF) (810). Comprehensive sequencing studies have shown that mutations in the cohesin subunit stromal antigen 2 (STAG2) are most frequently implicated in ML-DS development (11, 12). On the basis of these observations, it is hypothesized that the evolution of Down syndrome leukemia requires at least three distinct genetic events: T21, GATA1s, and additional mutations such as STAG2. However, the identity of the human hematopoietic cell type that acquires GATA1s and originates preleukemia and the cell type in which subsequent mutations accumulate to generate leukemia are unknown. Furthermore, the cellular origin of preleukemic mutations in pediatric leukemia in general is currently unclear, although twin studies of B cell acute lymphoblastic leukemia (ALL) point to a primitive cellular origin (13). Down syndrome leukemogenesis offers a disease setting to uncover generalized principles regarding this phenomenon.

Currently, there are no effective strategies to predict the subset of individuals at increased risk for progression from preleukemia to ML-DS. Life-threatening symptoms associated with preleukemia are treated with cytarabine (14, 15); however, this treatment does not prevent subsequent development of leukemia. Despite the generally favorable response of ML-DS to standard chemotherapy, outcomes are dismal for those with refractory or relapsed disease, with an overall survival rate of less than 20% (1618). Therefore, therapeutic targeting of preleukemic clones could represent a general concept to prevent development of leukemia. Experimentally, the mechanistic study of T21 and Down syndrome preleukemia has been challenging, primarily because of the fetal origin of the disease and the overall lack of suitable in vivo models (1921). To circumvent these limitations, we used disomic and trisomic hematopoietic cells that were isolated from primary human fetal livers, the major hematopoietic organ during prenatal development, to investigate the mechanisms underlying preleukemia and leukemia.

Here, we describe a model that faithfully recapitulates the full spectrum of premalignant and malignant stages of Down syndrome leukemia by use of CRISPR/Cas9 methodology optimized for single human hematopoietic stem cells (HSCs) (22) in disomic and trisomic primary human fetal liver-derived HSCs and downstream progenitors (23). Using this tool, we uncover insights into the genetic events and cellular contexts underlying the preleukemic and leukemic phases of Down syndrome leukemogenesis and provide proof of concept for targeting the preleukemic stage of the disease.

Results

GATA1s induces a megakaryocytic bias in hematopoietic stem and progenitor cells

To study the initiating events in Down syndrome preleukemia, we first sorted hematopoietic stem and progenitor cells (HSPCs) from normal (disomic) and T21 human fetal livers (N-FL and T21-FL, respectively) obtained at 16 to 19 weeks of gestation and carried out phenotypic analysis of the HSPC hierarchy (Fig. 1A and fig. S1B). T21 karyotype of sorted HSPCs was confirmed by means of droplet digital polymerase chain reaction (PCR) with a set of probes against Chr21 (fig. S1C). In addition, error-corrected targeted sequencing at 7500× coverage did not reveal any preexisting mutations of GATA1 exon 2 in T21-FL–derived HSPCs (table S1); although rare, preexisting mutations in exon 3 would have been missed (24). The impact of T21 on the HSPC hierarchy of fetal liver revealed a 30% increase in the percentage of total phenotypic long-term HSCs (LT-HSCs) and a simultaneous decrease in short-term HSCs (ST-HSCs) compared with that in N-FL. In addition, an expansion of megakaryocyte-erythroid progenitors (MEPs) was seen in T21-FL compared with N-FL, as previously reported (figs. S1, D to E, and S2A) (25). Quantification of GATA1 and its isoforms revealed comparable expression in N-FL and T21-FL HSPC subpopulations, with a gradual increase in expression upon differentiation commitment (fig. S2B).

Fig. 1 GATA1s induces a megakaryocytic bias in hematopoietic stem and progenitor cells.

(A) Experimental overview of in vitro single-cell differentiation/proliferation assay and near-clonal xenotransplantation. (B) Western blot assay of GATA1 in combined N-FL CMP and MEP cells, which were CRISPR/Cas9–edited with control and GATA1s gRNAs (n = 2 experiments). (C) Western blot assay of STAG2 in combined N-FL CMP and MEP cells, which were CRISPR/Cas9–edited with control and STAG2ko gRNAs (n = 2 experiments). (D) Proliferation capacity assessed by the overall number of CD45+ cells from in vitro single-cell assay of individual CRISPR/Cas9–edited cells for N-FL. Numbers of single-cell colonies with appropriate positive genotype are indicated for each condition (n = 2 experiments). (E) Proliferation capacity described in (D) for T21-FL (P < 0.0001 for combined T21-FL versus N-FL, n = 2 or 3 experiments). (F) Proliferation capacity assessed as the number of CD41+ megakaryocytic cells in all megakaryocyte-containing colonies from the single-cell assays described in (D) and (E). (G) Lineage output from in vitro single-cell assay described in (D) [P < 0.05 for Meg in GATA1s versus control, P = 0.12 for Meg in GATA1s/STAG2ko versus control, and P < 0.01 for (E) in GATA1s versus control among all cell types, n = 2 experiments]. (H) Lineage output from in vitro single-cell assay as described in (G) for T21-FL [P < 0.001 for Meg in GATA1s versus control, P < 0.01 for Meg in GATA1s/STAG2ko versus control, and P = 0.16 for (E) in GATA1s versus control among all cell types, n = 2 or 3 experiments]. Unpaired Student’s t test: *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; error bars represent standard error of the mean.

To examine the roles of GATA1s and STAG2 in leukemogenesis individually or in combination, we performed CRISPR/Cas9 editing of sorted HSPC subpopulations to express the short isoform of GATA1 (GATA1s) under its endogenous promoter and/or to delete the cohesin subunit STAG2 (STAG2ko). Using our optimized methodology (22), editing efficiency exceeded 75% (fig. S2C). Karyotyping analysis of N-FL HSPCs revealed no structural abnormalities after CRISPR/Cas9 editing (fig. S2D), and whole-genome sequencing in N-FL LT-HSCs at 30× coverage revealed either very rare or no off-target indels at sites that were similar to the guide RNA (gRNA) sequence (table S2). In the few cases in which off-target indels were detected, the allelic depth was 6% or lower. Western blot assays of CRISPR/Cas9–edited common myeloid progenitors (CMPs) and MEPs showed exclusive expression of GATA1s and undetectable STAG2 protein (Fig. 1, B and C, and fig. S2, E and F), confirming CRISPR/Cas9 editing of the respective genes at the protein level.

To elucidate the functional consequences of GATA1s and STAG2ko in different HSPC subpopulations, N-FL and T21-FL LT-HSCs, ST-HSCs, CMPs, and MEPs were purified; CRISPR/Cas9–edited for control, GATA1s, STAG2ko, or GATA1s/STAG2ko; and placed into single-cell in vitro differentiation and proliferation assays by using erythro-myeloid-megakaryocytic–promoting medium (26). The phenotype and genotype of all ~3000 single-cell–derived colonies used in our study were determined (fig. S2, G and H). Consistent with a previous report (25), colony-forming efficiency was higher in CRISPR/Cas9–edited T21-FL LT-HSCs compared with N-FL (fig. S3, A and B). The single-cell CRISPR/Cas9 editing efficiency was above 80% for both control and STAG2ko colonies (fig. S3, C and D). For GATA1s, the CRISPR/Cas9 efficiency was ~40% because only colonies with confirmed complete excision of exon 2 were included in the analysis. No preexisting GATA1 mutations in exon 2 were observed in any of the analyzed ~900 T21-FL control–edited colonies, confirming the results of the error-corrected targeted sequencing of T21-FL–derived HSPCs (fig. S3E).

Proliferation measured by total CD45+ cell output was lower in control-edited T21-FL HSPC subpopulations as compared with N-FL (Fig. 1, D and E). However, there was an increase in cell numbers observed in T21-FL GATA1s and GATA1s/STAG2ko subpopulations compared with T21-FL control colonies. A similar pattern was noticed in edited N-FL progenitor subpopulations but to a lesser extent when comparing N-FL GATA1s and GATA1s/STAG2ko colonies with N-FL control colonies. The increase in proliferative capacity in GATA1s and GATA1s/STAG2ko LT-HSCs was accompanied by a significant increase in the production of CD41+ megakaryocytes within all megakaryocyte-containing colonies (P < 0.05) (Fig. 1F). To investigate whether the decreased proliferative capacity of control-edited T21-FL was related to a change in the number of cycling or quiescent cells, we performed cell-cycle analysis. T21-FL HSPC subpopulations contained a lower percentage of cells in S phase and a higher frequency of cells arrested in G0 or G1 phase compared with those in N-FL (fig. S3F). No difference was observed in the ratio of quiescent G0 to G1 cells between N-FL and T21-FL (fig. S3G). Thus, despite an increased proportion of the LT-HSC compartment in T21-FL (fig. S2A) and higher colony-forming capacity (fig. S3, A and B), T21-FL cells exhibited significantly reduced proliferative capacity compared with their N-FL counterparts (P < 0.0001) (Fig. 1, D and E). However, the acquisition of a GATA1 mutation increased their proliferative capacity (Fig. 1, D and E), possibly providing a selective advantage to T21-FL HSPCs.

Phenotypic analysis was performed on single-cell–derived colonies, which were cultured in high-cytokine medium that forces terminal differentiation of HSPCs. Colonies derived from single N-FL HSPCs revealed a shift toward megakaryocytic differentiation and a concomitant decrease in erythroid lineage output in GATA1s and GATA1s/STAG2ko colonies compared with controls (Fig. 1G). A subset of these GATA1s and GATA1s/STAG2ko colonies expressed the early erythroid marker CD71 but not the mature erythroid marker GlyA, potentially indicating a hindrance toward erythroid differentiation (fig. S3H). Similar to N-FL, T21-FL GATA1s and GATA1s/STAG2ko HSPC subpopulations displayed a significant megakaryocytic bias (P < 0.01) and a decrease in erythroid output compared with controls (Fig. 1H and fig. S3I). By contrast, both N-FL and T21-FL STAG2ko cells exhibited an increase in erythroid output compared with controls. Collectively, these in vitro results indicate that exclusive expression of GATA1s with or without STAG2ko resulted in increased megakaryocytic output in all HSPC subpopulations, with no major differences in terminal differentiation between N-FL and T21-FL.

T21 is required for preleukemia initiation but dispensable for leukemia development

To evaluate the functional effects of GATA1s and STAG2ko in vivo, we used LT-HSCs because these are the only cells that have the ability to permanently repopulate the entire hematopoietic system after transplantation (23). We carried out xenotransplantation assays using NOD.Cg-PrkdcscidIl2rgtm1Wjl/SzJ (NSG) and NOD.Cg-PrkdcscidIl2rgtm1WjlKitem1Mvw/SzJ (NSGW41) recipients. NSGW41 mice were added to the study for their ability to engraft human cells without the need of irradiation and their increased ability to support the growth of erythroid and megakaryocytic lineages (27). The repopulating cell frequency of N-FL LT-HSCs injected into mice evaluated at 20 weeks was approximately 1 in 300 (fig. S3, J and K). We therefore transplanted N-FL and T21-FL control, GATA1s, STAG2ko, and GATA1s/STAG2ko LT-HSCs at cell doses of 300 to 400 into mice to obtain near-clonal grafts. After 20 weeks, human engraftment was analyzed in the bone marrow (BM), and extramedullary hematopoiesis was assessed in the spleen. Only mice bearing confirmed CRISPR/Cas9–edited grafts were used in the subsequent analysis (fig. S4, A to C). To evaluate the clonality of xenografts, BM cells of engrafted mice were plated in methylcellulose colony assays. Sanger sequencing of CRISPR/Cas9–mediated indels in individual colonies showed predominantly clonal engraftment in mice (fig. S4, D and E), validating our in vivo experimental approach.

On average, the human CD45+ engraftment level in BM was ~25% for mice transplanted with N-FL LT-HSCs and lower for T21-FL LT-HSCs, with the exception of mice transplanted with GATA1s/STAG2ko LT-HSCs, which displayed engraftment of ~30% (Fig. 2, A and B). Lineage marker analysis revealed increased myeloid and decreased lymphoid lineage cells in T21-FL control grafts compared with N-FL (Fig. 2, C and D, and fig. S4, F to R). The proportion of human CD41+CD45 megakaryocytic lineage cells was at least threefold higher in GATA1s and GATA1s/STAG2ko grafts as compared with control for both N-FL and T21-FL, which is consistent with the results observed in the in vitro single-cell assays. Moreover, immunohistochemistry (IHC) staining for the megakaryocytic marker CD61 in bone sections of humeri revealed an increase in cells expressing megakaryocytic markers in mice engrafted with GATA1s and GATA1s/STAG2ko cells from both N-FL and T21-FL (Fig. 2E and fig. S5, A to D). Engraftment patterns were similar in NSGW41 and NSG recipients of CRISPR/Cas9–edited N-FL and T21-FL (fig. S6, A to I). However, several differences were observable in GATA1s grafts from N- and T21-FL LT-HSCs. Overall, GATA1s LT-HSCs from T21-FL, but not from N-FL, were able to engraft in mice more efficiently than were their control-edited counterparts (fig. S6, J and K). This was further confirmed through enhanced repopulation of GATA1s LT-HSCs when a mixture of control and GATA1s LT-HSCs from T21-FL were transplanted into NSG mice for 6 weeks (fig. S6L). Immunophenotypic analysis of the HSC hierarchy of engrafted mice at 20 weeks after transplantation revealed a distorted LT-HSC/ST-HSC composition in T21-FL GATA1s xenografts but not in N-FL–derived grafts (fig. S6M). Last, T21-FL GATA1s grafts showed increased infiltration of myeloid lineage cells into the spleen compared with N-FL GATA1s (fig. S4R).

Fig. 2 T21 is required for preleukemia initiation but dispensable for leukemia development.

(A) Engraftment of N-FL LT-HSC grafts in NSG mice. Engraftment was assessed on the basis of human CD45+ expression in BM (only mice with >1% of CD45+ cells in BM and >90% CRISPR/Cas9 efficiency were taken for the analysis; n = 3 cohorts). (B) Engraftment as described in (A) for T21-FL (n = 4 cohorts). (C) Lineage marker distribution based on cell surface markers in N-FL grafts in NSG mice. (D) Lineage marker distribution as described in (C) for T21-FL. P < 0.05 for T21-FL control myeloid cells versus N-FL control myeloid cells, and P < 0.05 for T21-FL control lymphoid cells versus N-FL control lymphoid cells. (E) Hematoxylin and eosin (H&E) and IHC stainings for human CD45 and human megakaryocytic marker CD61 in humeri of T21-FL grafts. Scale bar, 50 μm. (F) Morphological analysis of human cells in primary xenografts of N-FL and T21-FL grafts. Human cells were prepared by using cytospin and stained with Giemsa (100× magnification). Scale bar, 10 μm. (G) Quantification of cell morphology as seen in (F) (n = 400 cells per condition). (H) Percent expression of cell surface markers within the CD45+ blast population in N-FL grafts in NSG mice. Data are from pooled samples of multiple xenografts. (I) Percent expression of cell surface markers as described in (H) for T21-FL. Data are from pooled samples of multiple xenografts. (J) Survival curve of N-FL LT-HSC grafts in NSGW41 mice (n = 5 mice per condition). (K) Survival curve as described in (J) for T21-FL (n = 5 mice per condition). Unpaired Student’s t test: *P < 0.05; ***P < 0.001; ****P < 0.0001; error bars indicate standard deviation.

To investigate whether the observed lineage shifts and engraftment patterns were associated with development of preleukemia or malignant transformation to full leukemia, we assessed xenografts for the presence of immature blast cells as assessed from cytomorphology. There was a dramatic increase in blasts to ~30 to 40% in T21-FL GATA1s but not in N-FL GATA1s xenografts (Fig. 2, F and G, and fig. S6N). This was further confirmed through histology, which showed active blast infiltration within the BM of T21-FL GATA1s grafts but not in N-FL GATA1s xenografts (fig. S5D). Higher blast percentages of ~50 to 80% were observed in both N-FL and T21-FL GATA1s/STAG2ko grafts. Subsequently, we carried out a detailed flow cytometric analysis of lineage markers on large nongranulated cells in the blast gate (fig. S7A). N-FL and T21-FL control and N-FL GATA1s grafts had no enrichment of this gated population. The blast population of T21-FL GATA1s grafts expressed the primitive stem cell markers CD34 and CD117 (KIT), megakaryocytic marker CD41, erythroid markers CD71 and GlyA, and myeloid marker CD33 and also aberrantly expressed lymphoid markers CD4 and CD7 compared with control and N-FL GATA1s grafts. This immunophenotype accurately recapitulates the clinical phenotype seen in patients with preleukemic TAM and meets clinically defined criteria (Fig. 2, H and I) (14, 2830). Blasts in both N-FL and T21-FL GATA1s/STAG2ko grafts had immunophenotypes nearly identical to those of T21-FL GATA1s grafts, which is in keeping with the clinical observation that blasts from patients in the preleukemic and leukemic stages are often indistinguishable (30, 31). The blast immunophenotype of grafts generated in NSGW41 mice followed a comparable pattern (fig. S7, B and C).

We next assessed the survival of NSGW41 mice transplanted with 1300 N-FL or T21-FL control, GATA1s, STAG2ko, or GATA1s/STAG2ko LT-HSCs. No effect on overall survival was found in mice transplanted with control, GATA1s, or STAG2ko LT-HSCs from N-FL and T21-FL during the observation period of 210 days. By contrast, mice transplanted with either N-FL or T21-FL GATA1s/STAG2ko cells had a shorter median survival of 120 and 88 days, respectively (Fig. 2, J and K), highlighting an important difference between the preleukemic and leukemic disease in this model. Thus, in our model, we defined preleukemia on the basis of the GATA1s genotype, characterized by elevated blast counts (>10%) with megakaryocytic features, which is consistent with clinical guidelines (30, 32), whereas leukemia was defined on the basis of the GATA1s/STAG2ko genotype, increased blast counts (>20%) with megakaryocytic features (30, 33), and lethality in humanized mice. Overall, our findings demonstrate that T21 is necessary for preleukemia development driven by GATA1s but dispensable for leukemic progression upon acquisition of STAG2ko.

CD117 marks preleukemia- and leukemia-initiating cells, which possess a more MEP-like chromatin accessibility landscape

To assess the self-renewal properties of T21 GATA1s–induced preleukemia and GATA1s/STAG2ko–induced leukemia, we carried out secondary xenotransplantation assays. Because CD34 expression is absent in some Down syndrome leukemia cases (30), we sorted all primitive CD34+ and CD117+ cells from primary xenografts and transplanted them at defined doses into secondary NSGW41 recipients (Fig. 3A). We observed differences in self-renewal as measured by the ability of N-FL versus T21-FL GATA1s cells to propagate hematopoiesis in secondary recipients (Fig. 3B). Whereas secondary grafts originating from N-FL GATA1s cells were phenotypically similar to control grafts after 12 weeks, preleukemic T21-FL GATA1s secondary grafts contained characteristic blast populations equivalent to those seen in primary recipients (fig. S7, D and E), although with a lower preleukemia-initiating cell frequency of ~1/150,000. Both N-FL and T21-FL STAG2ko grafts had higher initiating-cell frequencies compared with that of controls, albeit lower in T21-FL as compared with N-FL (Fig. 3B), which is consistent with the previously reported increase in HSC self-renewal in a mouse model in which STAG2 was deleted (34). Both N-FL and T21-FL GATA1s/STAG2ko cells from primary grafts were able to generate secondary leukemic grafts containing characteristic blast populations (fig. S7, D and E), with initiating-cell frequencies of ~1/45,000 and ~1/90,000, respectively.

Fig. 3 CD117 marks preleukemia and leukemia initiating cells.

(A) Experimental overview of secondary xenotransplantation experiments. Flow cytometry plots of sorted human fractions from primary grafts are depicted. (B) Stem cell frequencies based on secondary xenotransplantations as described in (A). Limiting dilution analysis was used to assess normal, preleukemia-initiating, and leukemia-initiating cell frequencies (>0.1% CD45+ cells in BM was defined as engraftment, n = 2 to 5 mice for each condition and dose, total 113 mice). (C) Experimental overview of secondary xenotransplantations using sorted fractions of CD34+CD117, CD34+CD117+, and CD34CD117+ cells from primary grafts. Flow cytometry plots of sorted human fractions are shown, and highlighted cells were transplanted at defined doses into NSG mice. A blue plus sign indicates engraftment with CD45+ cells, and a red minus sign indicates no engraftment in secondarily transplanted mice (n = 2 to 5 mice for each condition and dose, in total 333 mice) (stem cell frequencies are provided in table S3). (D) Tertiary xenotransplantations of N-FL and T21-FL grafts in NSG mice for 12 weeks (>0.1% CD45+ cells in BM was defined as engraftment, n = 5 mice per condition). (E) CIBERSORTx analysis to computationally quantify cell type lineage in the sorted fractions from primary N-FL grafts (n = 3 replicates per condition). (F) CIBERSORTx analysis as described in (E) for T21-FL (n = 3 replicates per condition).

To evaluate the relevance of CD34 and CD117 expression independently, cells from primary xenografts were further sorted into CD34CD117+, CD34+CD117+, and CD34+CD117 fractions and transplanted at defined doses into secondary NSG recipients (Fig. 3C and table S3). For both N-FL and T21-FL controls, only cells from the CD34+CD117+ fraction were able to generate serial grafts at 12 weeks, with T21-FL showing a lower stem cell frequency compared with that of N-FL (Fig. 3C and table S3). Similar to control grafts, only CD34+CD117+ cells from preleukemic T21-FL GATA1s primary grafts were able to generate secondary grafts, with a low preleukemia-initiating cell frequency of ~1/380,000. For STAG2ko primary grafts, both CD34+CD117+ and CD34+CD117 cells were able to engraft in secondary recipients. For leukemic N-FL and T21-FL GATA1s/STAG2ko primary grafts, cells from both CD34+CD117+ and CD34CD117+ fractions propagated engraftment in secondary recipients, indicating that CD117 might be a better marker than CD34 for leukemia-initiating cells in Down syndrome leukemia. Both preleukemic and leukemic engraftments were confirmed by the appearance of characteristic blast populations (fig. S7, F to H). N-FL and T21-FL GATA1s/STAG2ko grafts but not T21-FL GATA1s grafts could be serially reproduced in tertiary mice (Fig. 3D). These findings highlight the transient nature of GATA1s-mediated preleukemia versus leukemia induced by GATA1s/STAG2ko and reflect the spontaneous remission that occurs in most affected individuals with TAM.

To investigate the mechanism by which GATA1s and STAG2 deficiency contribute to leukemogenesis, specifically within the propagating CD34/CD117 cell fractions from primary xenografts, we carried out transcriptional and epigenetic profiling by means of RNA-sequencing (RNA-seq) and assay for transposase-accessible chromatin with high-throughput sequencing (ATAC-seq). CIBERSORTx was used to computationally infer the cell lineage contribution from bulk ATAC-seq (35). For this, a signature matrix was generated from normalized read counts over a set of sites specific to individually sorted N-FL HSPC subpopulations (figs. S1D and S8, A and B), including F1, F2, and F3 subgroups of MEPs and CMPs sorted according to CD71 and BAH-1 expression (26). Engrafting fractions of T21-FL GATA1s LT-HSCs and N-FL and T21-FL GATA1s/STAG2ko LT-HSCs exhibited an increased MEP-like signature compared with control, with the MEP F3 subgroup being the most prominent (Fig. 3, E and F). Furthermore, these fractions showed enrichment of GATA-binding motifs at promoters (fig. S8C and table S4) associated with an increase in gene expression at these sites (fig. S8D). Gene set enrichment analysis of differentially expressed genes between preleukemic versus control and leukemic versus control populations (table S5) revealed down-regulation of pathways implicated in translation, ribosome biogenesis, and interferon signaling in preleukemic and leukemic populations (fig. S8, E and F, and table S6). Up-regulated genes in T21-FL GATA1s fractions were enriched in Down syndrome leukemia blasts from primary patient samples, whereas down-regulated genes in preleukemic and leukemic fractions were enriched in the stem cell–rich CD34+CD38 population of human FL (fig. S8G) (36). Our results demonstrate that the preleukemic and leukemic propagating populations possess an open chromatin landscape that is largely driven by GATA1s-mediated transcriptional activation. Both of these propagating populations are identifiable from CD117 expression, which suggests that it could potentially serve as a therapeutic target.

Combined GATA1s and STAG2ko drive leukemic progression in progenitors

The originating cell type in leukemogenesis is increasingly recognized as playing an essential role in the resulting malignancy (3739). To determine whether progeny downstream of LT-HSCs are able to initiate preleukemic or leukemic transformation, we introduced GATA1s and/or STAG2ko into functionally defined subpopulations of ST-HSCs, CMPs, and MEPs and transplanted them into NSGW41 mice at a dose of 1000 cells (Fig. 4A). No consistent human CD45+ engraftment was detected after 12 weeks in mice transplanted with control, GATA1s, or STAG2ko cells from either N-FL or T21-FL progenitors (Fig. 4, B and C), although limited engraftment of early erythroid lineage cells was observed in mice transplanted with N-FL GATA1s cells (fig. S9, A and B). Even higher doses of 5000 control, GATA1s, or STAG2ko progenitors from T21-FL did not produce any CD45+ engraftment (fig. S9, C and D). This is consistent with the limited self-renewal and repopulation potential of progenitors but also highlights the inability of T21-FL GATA1s progenitor cells compared with LT-HSCs to initiate preleukemia. By contrast, human CD45+ engraftment was observed in mice transplanted with N-FL or T21-FL GATA1s/STAG2ko cells, regardless of the differentiation stage of the progenitors (Fig. 4, B and C). Only progenitors and stem cells could initiate leukemic engraftment because CD34 mature cells from T21-FL failed to initiate any CD45+ grafts upon GATA1s/STAG2ko, even when transplanted at a high dose of 125,000 cells for 12 weeks (fig. S9E). All grafts generated by N-FL and T21-FL GATA1s/STAG2ko progenitors contained high proportions of CD117+ blasts (Fig. 4, D and E, and fig. S9F) accompanied by other phenotypic markers typical of Down syndrome leukemia (fig. S9, G and H). Last, cells harvested from grafts generated by N-FL and T21-FL GATA1s/STAG2ko progenitors were able to propagate the leukemia in secondary recipients (fig. S9, I and J). Taken together, our results, on the basis of the assessment of these defined HSPC subpopulations, suggest that the GATA1s preleukemia-initiating event likely occurs in LT-HSCs and not downstream progenitors. However, subsequent STAG2 mutations are not limited to LT-HSCs but can be acquired further downstream in the expanded pool of GATA1s-primed progenitor cells, highlighting that the preleukemic and leukemic events could occur in distinct cells of origin.

Fig. 4 Combined GATA1s and STAG2ko drive leukemic progression in progenitors.

(A) Experimental overview of sorting N-FL and T21-FL derived progenitor cells for CRISPR/Cas9 editing and transplanting into NSGW41 mice. (B) Engraftment of N-FL ST-HSC, CMP, and MEP grafts in NSGW41 mice (all mice are shown regardless of CD45+ engraftment, n = 4 or 5 mice per condition). (C) Engraftment as described in (B) for T21-FL (n = 5 mice per condition). (D) Quantification of cell morphology of human cells prepared by using cytospin in N-FL and T21-FL GATA1s/STAG2ko grafts in NSGW41 mice from (B) and (C) (n = 400 cells per condition). (E) Flow cytometry plots depicting the blast population out of CD45+ cells in primary xenografts of T21-FL GATA1s/STAG2ko progenitors in NSGW41 mice, as described in (C). The CD34/CD117 profiles out of the CD45+ blast populations are depicted below. (F) Experimental overview of sorting T21-FL CMPs and MEPs to conduct a loss-of-function screen to identify genes that endow leukemic progression in combination with GATA1s. (G) Result of screen described in (F) showing the number of mice with leukemic phenotypes based on average CD45+ engraftment in BM and blast appearance (>1% CD45+ in BM, n = 5 mice per condition). (H) Flow cytometry plots of blast populations out of CD45+ cells in grafts of T21-FL CMPs and MEPs edited with GATA1s and candidate gene gRNAs as described in (F).

To verify whether leukemic transformation can be induced in a stepwise transplantation setting and to confirm whether progenitor-like cells could be responsible for leukemic progression, we sorted CD34+CD38+ progenitor-enriched and CD34+CD38 stem cell–enriched HSPCs from T21-FL GATA1s primary xenografts, induced STAG2ko, and transplanted them into NSGW41 secondary recipients for 12 weeks (fig. S10A). Stepwise introduction of GATA1s followed by STAG2ko in both subpopulations of T21-FL HSPCs elicited leukemic transformation, which was evident by the higher percentage of blasts as compared with that of the GATA1s-edited preleukemic control (fig. S10, B to E), confirming that progenitor-like cells with acquired STAG2 mutations could drive leukemic progression. Furthermore, to understand whether leukemic transformation with GATA1s and STAG2ko is developmentally restricted, we introduced GATA1s/STAG2ko in normal disomic FL-, postnatal umbilical cord blood (CB)–, or adult BM–derived CD34+ enriched HSPCs and transplanted them into NSGW41 mice for 12 weeks (fig. S11A). CD34+ cells from only FL and CB, but not BM, were able to induce leukemic transformation as assessed from blast accumulation and their characteristic immunophenotype (fig. S11, B to G). Therefore, the potential for leukemic transformation is developmentally restricted to a time window during fetal and early postnatal development.

To explore whether mutations in genes other than STAG2 can drive leukemic transformation, we carried out a focused loss-of-function screen to evaluate the effects of deleting seven additional genes in T21-FL GATA1s CMPs and MEPs, including four additional cohesin genes and three genes encoding epigenetic regulators that are frequently mutated in Down syndrome leukemia (11, 12). For each gene, four gRNAs were individually introduced into T21-FL progenitor cells together with GATA1s, and cells were pooled after CRISPR/Cas9 editing and transplanted at a dose of 20,000 cells into NSGW41 mice (Fig. 4F and table S7). After 12 weeks, all five cohesin gene mutations, each in combination with GATA1s, drove leukemic engraftment in mice (average level of CD45+ engraftment 2 to 20%), with STAG2, RAD21, and NIPBL being the most consistent (Fig. 4G). Of the three targeted epigenetic regulators, mutations in only KANSL1 drove leukemic transformation with GATA1s, implying that additional events are needed in the case of CTCF and EZH2 mutations. As expected, control-edited T21-FL progenitor cells with GATA1s did not produce any CD45+ grafts. All leukemic grafts contained CD117+ blasts with varying degrees of CD34+ expression (Fig. 4H). The blast immunophenotype in the leukemic grafts was similar regardless of the underlying mutation (fig. S12A), suggesting that the mutations converge on a common pathway for leukemic transformation. By contrast, a previous loss-of-function screen in a mouse model of TAM did not identify cohesin mutations as drivers of Down syndrome leukemogenesis (11), implying marked differences between mouse and human systems in their susceptibility to particular mutations. Altogether, our results show that specific cell types within a particular developmental time window are susceptible to GATA1s-induced preleukemia and GATA1s- and cohesin mutation–induced leukemia, underscoring the importance of the cellular and developmental context during leukemogenesis.

Chr21 microRNAs predispose to preleukemia

To investigate the mechanism underlying the cooperation between T21 and GATA1s in driving preleukemia development, we analyzed the binding occupancy of GATA1. To do this, we performed Cut&Run assays (40) to profile genome-wide GATA1 binding sites to quantify binding changes upon GATA1s editing in N-FL and T21-FL CD34+-enriched HSPCs. GATA1s retained many of the binding sites of full-length GATA1, as evidenced by the large number of shared peaks (fig. S12B), which is consistent with previously reported findings in a mouse cell line and a mouse model of Gata1s (41, 42). GATA-binding motifs were highly enriched in these peaks, as were motifs for ETS family members (fig. S12C), suggesting binding cooperativity with GATA1. Pathway enrichment analysis of GATA1s-specific peaks in T21-FL compared with either control-edited full-length GATA1 peaks in T21-FL or GATA1s peaks in N-FL revealed a 13-fold enrichment of promoter sites of genes involved in microRNA (miRNA) loading (fig. S12, D and E, and table S8), which was confirmed through gene expression of AGO1, AGO2, TARBP2, and ADAR (fig. S12F). These results suggest that GATA1s binding to these miRNA biogenesis genes increases their respective expression in the T21 context, possibly further influencing miRNA-mediated silencing and posttranscriptional regulation.

To explore this idea further, we investigated miRNA expression in T21-FL. We profiled miRNAs from N-FL and T21-FL CD34+-enriched HSPCs with next-generation sequencing. Differential expression of miRNAs on Chr21 was not observed (fig. S12, G and H). However, when Chr21 miRNAs were profiled by means of quantitative PCR in highly purified LT-HSCs, as compared with bulk CD34+ cells, differences were found (Fig. 5A). MiR-99a, miR-125b-2, miR-155, and let-7c were up-regulated in T21-FL LT-HSCs compared with N-FL, with the first three having the greatest differential expression.

Fig. 5 Chr21 miRNAs predispose to preleukemia.

(A) Relative expression of Chr21 miRNAs in T21-FL LT-HSCs compared with N-FL as measured with reverse-transcription quantitative PCR (n = 3 replicates per condition). (B) Overview of lentiviral mediated overexpression of Chr21 miRNAs in N-FL LT-HSCs used for primary xenotransplantation into NSG and NSGW41 mice. (C) Engraftment of transduced control and Chr21 miRNAs in N-FL LT-HSCs transplanted into NSG mice (only mice with >1% CD45+ cells in BM were analyzed, n = 9 or 10 mice per condition). (D) Lineage marker distribution based on cell surface markers of engrafted NSG mice in (C). (E) Quantification of cell morphology of human cells prepared by using cytospin in grafts described in (C) (n = 400 cells per condition). (F) Flow cytometry plots out of CD45+ cells in primary xenografts described in (C). (G) Experimental overview of sorting T21-FL LT-HSCs for CRISPR/Cas9 editing with miR-99a, miR-125b-2, and miR-155 gRNAs and subsequently with GATA1s and STAG2 gRNAs for primary xenotransplantation into NSG mice. (H) Engraftment of control knockout (KO) and Chr21 miRNAs KO in T21-FL LT-HSCs transplanted into NSG mice. Each subgroup was additionally edited with control, GATA1s, and GATA1s/STAG2 gRNAs (only mice with >1% CD45+ cells in BM and >90% CRISPR/Cas9 efficiency are depicted, n = 5 to 10 mice per condition). (I) Quantification of cell morphology of human cells prepared by using cytospin in transplanted NSG mice described in (H) (n = 400 cells per condition). (J) Flow cytometry plots showing blast populations out of CD45+ cells in primary xenografts described in (H). (K) Quantification of CD117+CD45+ blasts of transplanted NSG mice described in (H) (§§§ indicate significance in relation to GATA1s control KO). Unpaired Student’s t test: *P < 0.05; **P < 0.01; ***/§§§P < 0.001; ****P < 0.0001; error bars indicate standard deviation.

To investigate whether our observed T21-specific phenotypes could be recapitulated upon enforced expression of these differentially expressed Chr21 miRNAs in N-FL LT-HSCs, we used lentiviral transduction to overexpress miR-99a, miR-125b-2, and miR-155 (Chr21 miRNAs) together with the fluorescent marker mOrange. Transduced cells were transplanted into NSG and NSGW41 mice, and lineage output was assessed at 12 weeks (Fig. 5B and fig. S12I). Cells transduced with Chr21 miRNAs generated twofold higher engraftment in BM and spleen as compared with control-transduced cells (Fig. 5C). Chr21 miRNA grafts displayed a significant bias toward increased myeloid and decreased lymphoid differentiation in the transplanted BM (P < 0.0001) (Fig. 5D and fig. S13, A to E), similar to the lineage output of control CRISPR/Cas9–edited T21-FL cells in transplanted NSG mice (Fig. 2D). Similar grafts were seen in NSGW41 recipients of Chr21 miRNAs N-FL LT-HSCs (fig. S13, F and G). No elevated or abnormal blast populations were detected in any of these grafts with morphologic or flow cytometric analysis (Fig. 5, E and F, and fig. S13H). Immunophenotypic analysis of the HSC hierarchy of engrafted mice revealed reduced LT-HSCs in Chr21 miRNA grafts, which resulted in a lower ability of transduced CD45+ cells to engraft in secondary NSG mice (fig. S13, I and J). ATAC-seq and RNA-seq analysis of Chr21 miRNA LT-HSCs cultured in vitro and compared with analogously cultured T21-FL control-edited LT-HSCs showed similar chromatin accessibility profiles and enrichment of similar down-regulated genes (fig. S13, K and L, and table S9). These results demonstrate that simultaneous overexpression of miR-99a, miR-125b-2, and miR-155 in N-FL LT-HSCs recapitulates features of a T21-like hematopoietic state.

Next, to examine the role of Chr21 miRNAs in preleukemic initiation and leukemic transformation, we first deleted Chr21 miRNAs in T21-FL LT-HSCs and then followed with CRISPR/Cas9 editing for GATA1s, with or without STAG2ko, and transplanted into NSG mice (Fig. 5G and fig. S13 M to O). Deletion of Chr21 miRNAs combined with GATA1s resulted in a significant reduction in the blast population, including CD117+CD45+ blasts, at 20 weeks after transplantation (P < 0.001) (Fig. 5, H to K). However, leukemic engraftment or blast accumulation in mice with T21-FL GATA1s/STAG2ko grafts with or without deletion of Chr21 miRNAs were similar. Taken together, these results suggest that Chr21 miRNAs seem to play a major role in preleukemic initiation but are dispensable for leukemic progression.

CD117/KIT inhibition targets preleukemic-initiating cells and inhibits leukemic progression

Currently, there are no effective strategies to prevent progression from preleukemia to leukemia in individuals with Down syndrome. Our results indicate that CD117/KIT expression marked the cells that mediated the propagation of the GATA1s-induced preleukemia and GATA1s/STAG2ko–induced leukemia. Thus, it is possible that both the preleukemia and leukemia are dependent on KIT signaling for maintenance and progression.

CD117/KIT is a receptor tyrosine kinase that regulates HSC proliferation, maintenance, and survival after binding to its ligand, stem cell factor (43). To analyze KIT expression in normal hematopoiesis, FL-, CB-, and BM-derived LT-HSCs were immunophenotypically profiled for CD117 expression. N-FL and T21-FL LT-HSCs contained distinct CD117-low and CD117-high populations (Fig. 6A). By contrast, LT-HSCs from N-CB and N-BM showed a single population of cells with a continuum of low to high CD117 expression. After transplantation at limiting cell dose into NSG mice, only the CD117-high populations and not the CD117-low populations from N-FL and T21-FL LT-HSCs were able to generate grafts at 20 weeks (Fig. 6, B and C). By contrast, both CD117-low and CD117-high N-CB LT-HSCs were able to generate long-term engraftment, albeit with different lineage outputs (Fig. 6D). These results suggest that KIT signaling plays an essential role in LT-HSC function during fetal development. On the basis of the engraftment results of disomic CB (Fig. 6, B and C), KIT signaling may be less dependent, at least transiently, for LT-HSC function after birth.

Fig. 6 KIT inhibition targets preleukemic-initiating cells and inhibits leukemic progression.

(A) Immunophenotypic profile of CD117 and CD34 expression of isolated LT-HSCs from N-FL and T21-FL, normal disomic CB (N-CB), and normal disomic BM (N-BM). (B) CD117-low and CD117-high LT-HSCs were transplanted at defined doses into NSG mice for 20 weeks. Resulting stem cell frequencies are depicted (>0.1% CD45+ cells in BM was defined as engraftment, n = 4 or 5 mice per condition, total 67 mice). (C) Engraftment of N-FL CD117-high, T21-FL CD117-high, N-CB CD117-high, and N-CB CD117-low LT-HSCs transplanted into NSG mice (n = 4 to 8 mice per condition). (D) Lineage marker distribution based on cell surface markers of engrafted NSG mice from (C). (E) Experimental overview of control-, GATA1s-, and GATA1s/STAG2ko–edited T21-FL LT-HSCs transplanted into NSG mice, which were subsequently treated twice daily with small-molecule inhibitors against KIT for 2 weeks. (F) Engraftment of T21-FL LT-HSCs transplanted into NSG mice treated with vehicle, imatinib, dasatinib, or ripretinib (n = 4 or 5 mice per condition). (G) Quantification of cell morphology of human cells prepared by using cytospin in transplanted NSG mice described in (F) (n = 400 cells per condition). (H) Flow cytometry plots showing blast populations out of CD45+ cells in primary xenografts described in (F). (I) Quantification of CD117+CD45+ blasts in transplanted NSG mice described in (F). (J) Quantification of CD117+CD45+ blasts in NSGW41 mice transplanted with primary sample TAM 17003 and NSG mice transplanted with primary sample TAM 17041 and treated with vehicle, imatinib, dasatinib, or ripretinib (n = 5 mice per condition). Unpaired Student’s t test: *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; error bars indicate standard deviation.

To investigate whether pharmacological inhibition of KIT can target and eliminate preleukemic and leukemic blasts, mice engrafted with 1300 T21-FL control, GATA1s, or GATA1s/STAG2ko LT-HSCs were treated with first-, second-, and third-generation KIT inhibitors (50 mg/kg imatinib, 20 mg/kg dasatinib, or 7.5 mg/kg ripretinib) starting 10 weeks after transplantation with twice-daily dosing for 2 weeks (Fig. 6E) (4446). KIT inhibition did not have a significant effect on the overall level of CD45+ engraftment for any group, except for dasatinib-treated mice bearing T21-FL GATA1s preleukemic grafts (P < 0.01) (Fig. 6F). KIT inhibition had no effect on the blast population of leukemic grafts generated by T21-FL GATA1s/STAG2ko cells (Fig. 6G). GATA1s preleukemic grafts from mice treated with any of the KIT inhibitors contained significantly lower proportions of blasts as compared with that of vehicle-treated mice (P < 0.001), with reduction of CD117+CD45+ blast populations to levels seen in controls (Fig. 6, H and I, and fig. S14A). KIT inhibitor–treated mice revealed an increase in granular CD33+, CD11b+, and CD13+ myeloid cells, suggesting differentiation of blast cells toward more mature myeloid cells (fig. S14, B and C). Because some residual CD117+ blasts remained detectable in mice with preleukemic GATA1s grafts (Fig. 6H), cells harvested from primary mice were serially transplanted at defined doses into secondary NSG recipients to determine whether preleukemia-initiating cells were affected. Cells from vehicle-treated mice showed a 32-fold higher ability to generate secondary grafts at 8 weeks compared with cells from KIT inhibitor–treated mice (fig. S14D). This lies in contrast to the secondary grafts generated from KIT inhibitor–treated GATA1s/STAG2ko–induced leukemia, which did not show any difference in their ability to generate secondary grafts (fig. S14E). To further validate the sensitivity of KIT inhibition in GATA1s-induced preleukemia, two primary TAM samples were phenotypically characterized (fig. S14, F and G, and table S10), subsequently transplanted into mice, and treated with KIT inhibitors starting at 6 weeks after transplantation for 2 weeks (fig. S15A). KIT inhibition resulted in reduced CD45+ engraftment with significantly reduced CD117+CD45+ blast populations (P < 0.001) and increased granular CD33+ myeloid cells (Fig. 6J and fig. S15, B to H). However, we cannot rule out that the effects of pharmacological KIT inhibition could also be mediated, at least partly, through other receptor tyrosine kinases. Altogether, our results demonstrate as a proof of principle that KIT inhibition targets preleukemic expansion (fig. S15I), supporting further clinical evaluation of the concept that preleukemic intervention could inhibit progression to leukemia.

Discussion

Our study provides insight into the cellular and molecular mechanism of Down syndrome leukemogenesis, from atypical hematopoiesis associated with T21 to preleukemia initiation and ultimately to leukemic progression. We confirmed that the T21-FL hematopoietic system exhibits an altered phenotypic HSPC hierarchy as previously described (25, 47, 48). Although earlier reports have proposed that T21 enhances self-renewal in vitro (48), our functional studies revealed the opposite; individual T21 HSPC subpopulations exhibited reduced proliferation in vitro and generated smaller grafts in xenotransplanted mice with myeloid and megakaryocytic bias and reduced serial transplant ability. These are likely cell-autonomous effects and may be the basis for the higher incidence of hematopoietic abnormalities such as isolated cytopenias, myelodysplasia, and BM failure seen in adults with Down syndrome (49). Despite the reduced proliferative capacity of T21 LT-HSCs, our data demonstrate that preleukemia can be initiated in this cellular compartment, contrary to previous hypotheses that megakaryocytic-erythroid progenitor cells are the cell of origin for preleukemia, a prediction derived from their expansion in the HSPC hierarchy of T21-FL (50). The reduced proliferative capacity of T21 LT-HSCs is offset by the acquisition of GATA1 mutations, providing a possible explanation for the observed selection of GATA1 mutations in the context of T21. However, the increased function provided by GATA1 mutation comes at a cost, including the development of preleukemia and a hindrance in erythrocyte maturation—a result consistent with the anemia seen in GATA1-deficient mice and humans (5154). However, because we based our study on well-characterized functional HSPC subpopulations, we cannot exclude the possibility that undefined populations may also be able to initiate preleukemia (55). In contrast to the LT-HSC origin for preleukemia, leukemic progression can occur in multiple types of downstream progenitors in addition to LT-HSCs. The overall pool of progenitors is vastly expanded owing to GATA1s-priming, providing a large reservoir for acquisition of secondary mutations in genes such as STAG2 and thereby increasing the probability of leukemic progression. Enhanced self-renewal mediated by STAG2 deficiency could explain why it is subsequently selected for during leukemic evolution. Selection could also arise from STAG2 deficiency resulting in a temporary increase in mature erythroid output, as seen in our in vitro single-cell differentiation assays. Erythroid cells make up the vast majority of the numerical daily output of the blood system, raising the possibility that there may be strong evolutionary pressure for the erythropoiesis-defective GATA1s-mutated clones to reacquire erythroid potential through additional STAG2 mutation. Further, this leukemic transformation can only occur during fetal and early postnatal development but not in adult BM stem cells. Thus, our study reveals how critical it is to understand the identity and the developmental stage of the cell type that acquires genetic drivers during leukemogenesis. Moreover, genetic drivers of leukemia are typically distinct between pediatric and adult acute myeloid leukemia (56), so our findings uncover that the basis for the differential leukemic potential could possibly be the developmental status of the cell of origin.

Our findings establish that initiation of GATA1s-induced preleukemia is dependent on T21, which exerts its effects at least in part through up-regulation of Chr21 miRNAs—specifically, miR-99a, miR-125b-2, and miR-155—exclusively within the LT-HSC compartment. This result refines previous suggestions that deregulated expression of many Chr21 genes contributes to Down syndrome leukemogenesis and extends them to noncoding RNAs (57). These three miRNAs are highly expressed in leukemia-initiating populations of adult acute myeloid leukemia (58), and miR-125b-2 has been shown to be a potential oncomiR (59). Children with Down syndrome are also at high risk for developing B cell ALL (60), and it will be important to determine whether this mechanism of predisposition also affects ALL development. Although preleukemic initiation is dependent on T21, we made the unexpected finding that progression to leukemia is independent of T21 and can be induced by deficiency of STAG2 in combination with GATA1s. We found that preleukemic and leukemic populations were similar with respect to expression of lineage markers on blasts, enrichment of GATA1-binding sites at their promoters, and down-regulated pathways compared with T21 controls. Nevertheless, the addition of STAG2ko to GATA1s-bearing cells led to enhanced self-renewal, as evident in the increased frequency of leukemia-initiating cells enriched in CD117+ cells. Similar leukemic transformation was observed upon induced deficiency of other cohesin genes in combination with GATA1s, and thus, it is possible that the effects of these mutations converge on an increase in self-renewal and stemness programs in general. LSC17 stemness signature being strongly associated with survival across a wide spectrum of acute myeloid leukemia patients irrespective of genetic drivers (61) lends support to the concept of convergence. Last, although our results demonstrate that Down syndrome leukemogenesis can be modeled by a sequence of cell-intrinsic mechanisms, we cannot exclude an additional role for the T21 microenvironment in leukemic evolution in individuals with Down syndrome.

Through analysis of individual preleukemic and leukemic cell fractions isolated on the basis of primitive stem cell markers and propagated in serial transplants, we identified CD117/KIT as a marker of disease-driving cells. Further, pharmacological KIT inhibition targeted preleukemic stem cells, and this proof of concept could open an approach for therapy at the preleukemic stage. This would limit the pool of GATA1s-primed progenitors that can acquire additional mutations and ultimately inhibit progression to ML-DS. Such a scenario could foreseeably work in combination with refined clinical scores capable of identifying patients with TAM who have risk of progression toward leukemia subtypes that have poor outcome (62). However, this concept requires further in-depth preclinical and clinical assessment. Our findings not only provide insight into Down syndrome leukemogenesis but also have implications for pediatric leukemia in general. Sequencing data from newborns indicated that the first genetic alterations for many subtypes of childhood leukemia occur during fetal development (6365). Our results potentially suggest that the cellular origin of preleukemic mutations in other pediatric leukemia may also be LT-HSCs, which is supported by that it can take years after birth until leukemia is diagnosed (66). Early targeting of preleukemia during the newborn phase could become a clinical paradigm for pediatric leukemia. Last, numerical and structural alterations of Chr21 are extremely common in hematological malignancies in both children and adults (67). Gains of Chr21 are seen in up to 35% of cases in several types of acute leukemia (68). Thus, it will be important to identify mechanisms including but not limited to dysregulation of Chr21 miRNAs that are shared among those cases.

Materials and methods summary

All mouse experiments were approved by the University Health Network (UHN) Animal Care Committee, and we confirm that all experiments conform to the relevant regulatory and ethical standards. All xenotransplantations were performed in 8- to 12-week-old female NOD.Cg-PrkdcscidIl2rgtm1Wjl/SzJ (NSG) mice (JAX) that were sublethally irradiated with 225cGy, 24 hours before transplantation, or in 8- to 12-week-old female NOD.Cg-PrkdcscidIl2rgtm1WjlKitem1Mvw/SzJ (NSGW41) mice that were not irradiated. Human fetal liver samples were obtained from elective pregnancy terminations at Mount Sinai Hospital with informed consent in accordance to guidelines approved by the Mount Sinai Hospital Research Ethics Board (18-0093-E) and the UHN Research Ethics Board (02-0763). Fetal liver samples of normal disomic karyotype and T21 were collected at 16 to 19 weeks gestation from either sex. Fetal liver samples were processed within 1 to 3 hours, and CD34+ cells were isolated by use of the human CD34 MicroBead kit (Miltenyi Biotec) according to the manufacturer’s protocol (table S11). For all in vitro and in vivo experiments, the full stem and progenitor hierarchy was used to sort LT-HSCs, ST-HSCs, CMPs, and MEPs (26). CRISPR/Cas9 RNP electroporations were carried out by using chemically synthesized gRNAs (IDT), recombinant Cas9 nuclease (IDT), and the 4D-Nucleofector (Lonza) as previously described (22). Single-cell in vitro assays were performed in erythro-myeloid-megakaryocytic promoting medium. Xenotransplantations of CRISPR/Cas9–edited LT-HSCs were performed at cell doses of 300 to 400 per mouse, unless otherwise specified, to obtain near-clonal xenotransplantations.

Supplementary Materials

science.sciencemag.org/content/373/6551/eabf6202/suppl/DC1

Materials and Methods

Figs. S1 to S15

Tables S1 to S9

References (6987)

MDAR Reproducibility Checklist

References and Notes

Acknowledgments: We thank D. Curovic, R. Kelly, J. Law, and M. Niit at the Research Centre for Women’s and Infants’ Health Biobank (Mount Sinai Hospital) for sample coordination; B. Chow at the Pathology and Laboratory Medicine (Mount Sinai Hospital) for assistance with pathology; M. DSouza and R. Lopez at the Animal Resources Centre (UHN) for support with mouse work; M. Bergeret, S. Boddeda, S. Ng, A. Srinath, O. Subedar, A. AuYeung, and S. Zhao at the SickKids-UHN Flow and Mass Cytometry Facility for assistance with flow cytometry; B. Apresto at The Centre for Applied Genomics (SickKids) for sanger sequencing; K. Ho at the Centre for Applied Genomics (SickKids) for next-generation sequencing; A. Smith at the Cancer Cytogenetics Laboratory (UHN) for karyotyping; K. Asoyan, C. Cimafranca, J. Mouatt, M. Peralta, and Y. Yang at the Pathology Research Program (UHN) and N. Law at the Sttarr Innovation Centre (UHN) for assistance with histology; M. Bartolini at the Advanced Optical Microscopy Facility (UHN) for slide scanning; J. Moffat, K. Brown, and C. Ross at the Donnelly Centre for supplying the gRNA sequences for the in vivo screen; S. Henikoff at the Fred Hutchinson Cancer Research Center for supplying the Protein A-Micrococcal nuclease fusion protein for the Cut&Run assay; L. Shultz at the Jackson Laboratory for providing NSGW41 mice; the laboratories of S. Chan and F. Notta for sharing equipment; N. Mbong and A. Mitchell for technical assistance; and H. Hasle for sharing unpublished work. We thank C. Jones, A. Tikhonova, and members of the Dick laboratory for comments on the manuscript. Funding: This work was supported by funds from Human Frontier Science Program (LT-000601); Alex’s Lemonade Stand Foundation (19-16679); The Leukemia & Lymphoma Society and The Leukemia & Lymphoma Society of Canada (3404-21); Portuguese Foundation for Science and Technology (SFRH/BD/136200/2018); Princess Margaret Cancer Centre Foundation; Ontario Institute for Cancer Research through funding provided by the Government of Ontario; Canadian Institutes for Health Research (Foundation: 154293, Operating Grant 130412, Operating Grant 89932, and Canada-Japan CEEHRC Teams in Epigenetics of Stem Cells 127882); International Development Research Centre, Canadian Cancer Society (703212); Terry Fox Research Institute Program Project Grant; University of Toronto’s Medicine by Design initiative, which receives funding from the Canada First Research Excellence Fund; and a Canada Research Chair. Author contributions: E.W., J.E.D., and E.R.L. conceived the project, supervised research, and wrote the paper. J.A., O.I.G., G.K., and J.C.Y.W. edited the paper. E.W., J.A., O.I.G., and E.R.L. analyzed experiments. E.W., J.A., S.K.C., M.A., S.A.S., and B.A.G. performed in vitro and in vivo experiments. O.I.G and E.R.L. assisted with mouse work. J.A. performed morphological analysis. O.I.G. assisted with single-cell assays. A.M. analyzed ATAC-seq and RNA-seq data. G.K. performed Western blot assays and Cut&Run assays and prepared miRNA libraries. J.L.M. assisted with intrafemoral injections. S.A.M. and D.D.D.C. performed Cut&Run analysis. M.G. and L.S. analyzed miRNA sequencing data. J.J.F.M. generated smMIP libraries. S.M., J.C., and J.K.H. supplied primary TAM samples. M.C.-S.-Y. performed CRISPR/Cas9 off-target analysis. L.G.-P. assisted with ATAC-seq library preparations. S.A. performed smMIP analysis. M.A. performed histopathological analysis. K.C., M.R., P.S., and D.C. coordinated patient consent and sample collection. J.C.Y.W. and J.K.H. provided study consultation. J.E.D. secured funding for this study. Competing interests: D.D.D.C.: Pfizer and Nektar Therapeutics, research funding; DNAMx, cofounder and shareholder. J.E.D.: Celgene, research funding; Trillium Therapeutics, advisory board. All other authors declare no competing interests. Data and materials availability: Raw sequence data are available at European Genome-phenome Archive (EGAS00001004780) and processed data are available at Gene Expression Omnibus (GSE160096). All other data are available in the manuscript or the supplementary materials.

Stay Connected to Science

Subjects

Navigate This Article