Induction and Therapeutic Targeting of Human NPM1c⁺ Myeloid Leukemia in the Presence of Autologous Immune System in Mice

Acute myeloid leukemia (AML) is a malignancy of myeloid precursor cells that manifests as uncontrolled proliferation in the bone marrow (BM), leading to progressive marrow failure (1, 2). Old age (>65 y), diagnoses of myelodysplasia, and prior treatment with chemotherapeutic drugs are among the risk factors for developing AML. Most patients are treated with decades-old chemotherapy regimens often combined with hematopoietic stem cell transplantation. Patients who are able to withstand the intensive chemotherapy can achieve a complete response; however, a majority, especially the elderly, experience relapse and die of the disease (3, 4). Relapse is usually caused by minimal residual disease in the BM resulting from leukemic stem or initiating cells (LSC) that are refractory to standard therapies (3). Therefore, there is a need to develop therapies that can completely eliminate the leukemic burden.

At a molecular level, AML is a heterogeneous disease. The most commonly recurring genetic alterations in AML fall into distinct categories, including DNA methylation enzymes, transcription factors, and proteins involved in signaling cascades (5). Mutations in the nucleophosmin (NPM1) gene form a distinct subset and are present in ∼30% of all adult AML cases (6). Mutations in NPM1 occur in exon 12 and result in the loss of a nuclear localization signal (6, 7). Wild-type NPM1, which has a nucleocytoplasmic distribution, is involved in a multitude of cellular processes (7). Mutant NPM1, also referred to as NPM1c because of its predominantly cytoplasmic localization, has been shown to destabilize the p19 (Arf) tumor suppressor (8) and prevent the degradation of Myc (9, 10). NPM1c mutation is postulated to be a driver mutation because of its presence in all leukemic cells, including LSCs, the stable nature of the mutation throughout disease (detected at relapse), and its occurrence prior to genetic lesions in other genes such as internal tandem duplications in FMS-like kinase 3 (FLT3-ITD) (11, 12).

Based on recent successes of cancer immunotherapies, enormous effort is being poured into the development of immune-based targeted therapies for the treatment of cancer, including AML. However, one major hurdle is the lack of representative preclinical models. Ideally, such models should have stable reconstitution of human leukemic cells and immune cells, including T cells, NK cells, and macrophages, that mediate the cytotoxic effect of immunotherapeutics. Over the years, many small animal models have been developed for AML, including transplantable xenograft models, chemically and virally induced murine leukemic models, and genetically engineered mouse models (13, 14). A major limitation of these models is the lack of a matching human immune system because of the requirement for human immune cells in cancer cell elimination. Several groups have attempted to circumvent this problem by introducing non-HLA–matched human PBMCs, but the survival of these mice is very short because of induction of graft-versus-host disease. With respect to NPM1c-induced AML, introduction of human NPM1c into the corresponding mouse locus does not result in robust development of AML (15, 16). Vassiliou et al. (15) restricted the expression of NPM1c in mouse hematopoietic cells and observed AML development in 30% mice with a long disease latency. Although these models have facilitated our understanding of NPM1c in leukemogenesis, they are not suitable for testing biologics, which are often human-specific and require the human immune system to function.

In this article, we report a model of de novo human AML with an autologous human immune system in immunocompromised mice. In this model, AML is driven by enforced expression of NPM1c in human hematopoietic stem/progenitor cells (HSPCs) and results in a disease that resembles human NPM1c⁺ AML in presentation, phenotype, and transcriptional profile. Transcriptome analysis identifies upregulation of Myc and HOX signature genes in leukemic cells. Importantly, the nontransduced, normal HSPCs give rise to a functional human immune system in the same mice. The de novo AML also produces CD123⁺ LSCs in the BM, which can be depleted with a bispecific Fab conjugate (BFC) targeting CD3 and CD123 in a T cell–dependent manner. This model is uniquely positioned as a platform for studying early events in leukemogenesis in human and as a preclinical tool for testing immunotherapies.

Human CD34⁺ HSCPs were purified from fetal livers as previously described (17). Briefly, tissue was dissected into 5 mm³ pieces in digestion buffer containing DNase I and collagenase D, incubated at 37°C for 30 min, and homogenized. Following top layering with Ficoll–Paque (GE Healthcare), interphase containing immune cells and CD34⁺ HSPCs was collected and washed with PBS. EasySep Human CD34 Positive Selection Kit (StemCell Technologies) was then used to purify CD34⁺ HSPCs. For viral transduction, lentivirus was produced by transient transfection of 293 cells with plasmids encoding VSVG, delta8.9, and pGL3-derived lentivirus plasmids encoding GFP or GFP and NPM1c. HSPCs were propagated in Stem Span media supplemented with Angiopoietin-like 5 (Angptl5; Abnova), human stem cell factor, human fibroblast growth factor (Invitrogen), insulin-like growth factor binding protein 2 (IGFBP2; R&D systems), heparin (Sigma-Aldrich), and thrombopoietin (R&D systems). Cytokines were reconstituted in PBS + 0.1% BSA. The use of human tissue in this study was approved by the Institutional Review Board at Massachusetts Institute of Technology.

Lentivirus-transduced CD34⁺ HSPCs (2 × 10⁵) were engrafted via intracardiac injection into NOD-scid ILR2γ^−/− (NSG) neonates irradiated with 0.7 Gy. Eight weeks postengraftment, mice were serially bled every 2 wk, and leukocytes were analyzed for human CD45 and GFP expression. For secondary transplantation, 6- to 8-wk-old NSG mice were hydrodynamically injected with 100 μg DNA plasmids encoding human IL-3 and GM-CSF, as previously described (18). Ten to fourteen days later, mice were irradiated with 2.7 Gy, followed by tail vein injections with leukemic cells from the primary mice. Four weeks postengraftment, mice were bled to monitor disease development by flow cytometry. All mouse work was approved by the Institutional Animal Care and Use Committee.

Mice were bled via the tail vein to assess human cell reconstitution. Terminal mice were sacrificed with CO₂, followed by cardiac puncture for blood collection. BM cells were collected from the femurs. Blood samples and BM cells were incubated with ACK lysis buffer (Lonza) to lyse RBCs. Leukocytes were resuspended in PBS, stained with the appropriate Abs on ice for 20 min, and washed. The following Abs, specific for mouse CD45.1 (clone A20), human CD45 (clone HI30), CD13 (clone WM15), CD33 (clone WM53), CD38 (clone HIT2), CD47 (clone CC2C6), CD11b (clone M1/70), CD14 (clone M5E2), CD34 (clone 581), CD123 (clone 6H6), CD3 (clone HIT3a), CD56 (clone HCD56), CD19 (clone HIB19), CD45RA (clone HI100), and CD45RO (clone UCHL1), were purchased from BioLegend. Stained cells were resuspended in PBS + 10% FBS with DAPI and filtered. At least 10,000 events were collected on an LSR II flow cytometer (Becton Dickinson). Data were analyzed with FlowJo software. Cell sorting was performed on an Aria3 machine (Becton Dickinson). For secondary transplant experiments, sorted cells were resuspended in PBS for immediate tail vein injections. For mRNA processing, sorted cells were pelleted, snap frozen, and stored at −80°C.

For in vitro T cell killing assays, autologous T cells and BM cells from NPM1c⁺ mice were harvested. T cells were purified from the spleens or blood of mice with an EasySep CD3 enrichment kit (StemCell Technologies). Briefly, single-cell suspension was prepared from spleens and blood, and the resulting cell suspension was ACK lysed and resuspended in appropriate medium for purification, as per the manufacturer’s protocol. T cells with a purity of >90% were used for in vitro killing assays. BM cells were harvested and processed as described above. Cells were counted and stained with allophycocyanin-conjugated anti-CD123 and PE-conjugated anti-CD33 Ab to determine absolute cell numbers. For in vitro killing assays, T cells and target cells were resuspended in RPMI 1640 + 10% FBS and incubated at the indicated ratios in 96-well plates. Five microliters of anti-CD107a Ab and 1 μg of BFC was added to the cell cultures and incubated for 4–48 h at 37°C. For JQ1 pretreatments, BM cells were treated with JQ1 for 12–16 h and the drug was washed out. BM cells were then incubated with T cells, BFC, and anti-CD107a Ab as indicated. At the end of the incubation period, cells were washed with PBS + 0.1% BSA and stained with Live/Dead Aqua (Invitrogen), followed by staining with the indicated Abs. Samples were processed on a BD LSRII cytometer, and data were analyzed with the FlowJo software. For JQ1 treatments, a fixed number of cells were treated with 1 μM of JQ1 for 48 h in 96-well plates. At the end of the incubation period, cells were washed and viable cells were counted with trypan blue stain.

The mIgG2a anti-CD3 clone OKT3 (Orthoclone OKT3; Janssen), mIgG2a anti-CD123 clone 7G3 (554526; BD Pharmingen), and null arm control mIgG1 anti–keyhole limpet hemocyanin (KLH) (MAB002; R&D Systems) were digested to F(ab′)₂ using immobilized pepsin or ficin (44988 and 44980; Pierce, Thermo Fisher Scientific) following the manufacturer’s instructions. Undigested parental Ab was removed from F(ab′)₂ using Protein AG columns (89950; Pierce, Thermo Fisher Scientific). F(ab′)₂ was reduced by addition of TCEP (77720; Pierce, Thermo Fisher Scientific) to a final concentration of 2.5 mM, and buffer was exchanged to 100 mM phosphate buffer, 150 mM NaCl (pH 8). Anti-CD3 and null control Fab′ were modified with equimolar maleimido trioxa-4-formyl benzamide (MTFB; S-1035-105; Solulink). Anti-CD123 Fab′ and additional null control Fab′ were modified with equimolar 3-N-maleimido-6-hydraziniumpyridine hydrochloride (MHPH; S-1009-010; Solulink). To generate BFC, equimolar Fab-MTFB and Fab-MHPH were conjugated overnight at 4°C in the presence of 10% (v/v) aniline catalyst (Solulink S-2006-105) and a 2-fold molar excess of Ellman’s reagent (ALX-400-034-G005; Enzo Life Sciences) at pH 6. Unreacted Fab monomers were removed from BFC conjugation products using Sephadex G-100 (G10050-10G; Sigma-Aldrich) gravity gel filtration in PBS. BFC was diluted in PBS, and mice were dosed with 1 μg of BFC for seven consecutive days via tail vein injections.

For quantitative RT-PCR (qRT-PCR) analysis, RNA was isolated with TRIzol (Invitrogen), as per the manufacturer’s instructions. cDNA was generated with SuperScript First-Strand (Invitrogen), and quantitative PCR was performed using LightCycler 480 SYBR green mix (Roche). Primer sequences are as follows: tubulin forward (FWD): 5′-CCAGATCTTTAGACCAGACAAC-3′, tubulin reverse (RVS): 5′-CAGGACAGAATCAAC CAGCTC-3′; human CD34 FWD: 5′-GAGACAACCTTGAAGCCTAG-3′, human CD34 RVS: 5′-CTGAGTCAATTTCACTTCTCTG-3′; HOXA9 FWD: 5′-CTTGTGGTTCTCCTCCAGTTG-3′, HOXA9 RVS: 5′-CATGAAGCCAGTTGGCTGCTG-3′; HOXA5 FWD: 5′-GCAAGCTGCACATA AGTC-3′, HOXA5 RVS: 5′-CCAGATTTTAATTTGTCTCTCGG-3′; HOXA6 FWD: 5′-GTTTAC CCTTGGATGCAGC-3′, HOXA6 RVS: 5′-GTAGCGGTTGAA GTGGAACTC-3′; Myc FWD: 5′-CTCCAGCTTGTACCTGCAGGATCTGAG-3′, Myc RVS: 5′-GAGCCTGCCTCTTTTCCACAG-3′. For RNA sequencing, frozen pellet of sorted cells was thawed on ice and RNA was extracted with an RNeasy Micro Kit (Qiagen) and analyzed on a BioAnalyzer. Due to the limited amount of RNA obtained, the Ovation RNA-Seq system from NuGEN was used for library preparation. Adapters were ligated on the amplified library and sequenced with an Illumina HiSeq 2000. Raw sequences are deposited in the database of Gene Expression Omnibus with accession identifier GSE124538 and URL: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE124538. Quality control was performed on the data followed by RSEM analysis using the Bowtie2 option. Raw counts were aligned to the human transcriptome (hg19 reference database). EdgeR was then used for paired analysis with a p value cutoff of 0.01.

Sorted cells were resuspended at a concentration of 1 × 10⁶ cells per ml and fixed with 70% ethanol for 2 h. Fixed cells were centrifuged at 300 × g for 5 min. Ethanol was removed, and cells were rinsed with ice-cold HBSS (containing Mg²⁺ and Ca²⁺). Cells were resuspended in HBSS at a density of 1 × 10⁶ cells per ml. Cell suspensions were mixed 1:1 ice-cold Pyronin Y Hoechst 33342 solution and incubated for at least 10 min in the dark. Cell fluorescence was measured using a BD LSR II cytometer.

Where indicated, two-tailed t tests were performed to compute statistical significance between datasets. The p values were as follows: *p < 0.05, **p < 0.01, and ***p < 0.001.

To develop human AML in mice, we introduced the human NPM1c mutation (6) into HSPCs and engrafted transduced cells in NSG mice. To identify transduced cells, GFP was coexpressed with NPM1c in equal stoichiometry using the 2A self-cleaving peptide. Lentiviral vectors were constructed to express either GFP alone (control) or GFP plus NPM1c, all under the control of the ubiquitous phosphoglycerate kinase 1 (PGK) promoter (Fig. 1A). Human CD34⁺ HSPCs were transduced with either lentivirus with a transduction efficiency of 5–20%. Mixtures of transduced and untransduced HSPCs were engrafted into NSG neonates by intracardiac injection. Mice were monitored for human leukocyte reconstitution and GFP expression in the peripheral blood starting 8 wk postreconstitution. Mice engrafted with HSPCs transduced with GFP or GFP plus NPM1c lentivirus are referred to as control mice and NPM1c⁺ mice, respectively.

Control and NPM1c⁺ mice had similar levels of human leukocyte reconstitution in the peripheral blood 9 wk postengraftment (Fig. 1B). Whereas control mice had ∼ 15% GFP⁺ human leukocytes, NPM1c⁺ mice had ∼22% GFP⁺ human leukocytes (Fig. 1C). Notably, a higher fraction the GFP⁺ leukocytes was CD33⁺ myeloid cells in NPM1c⁺ mice than in control mice (Fig. 1D). All NPM1c⁺ mice died within 14–27 wk postengraftment, whereas control mice lived regular lifespan of more than a year (Fig. 1E). The age at which NPM1c⁺ mice died seemed to correlate with the level of HSPC transduction: the higher the percentage of lentiviral transduction, the younger the recipient mice died. In moribund NPM1c⁺ mice, blasts cells were readily detected in the blood and BM (Fig. 1F). Moribund NPM1c⁺ mice had visibly fewer RBCs in blood smear and pale femurs with significantly reduced cellularity but enlarged spleens (Fig. 1F–1H), indicating suppressed erythropoiesis in the BM. Infiltration of leukemic cells was often detected in the liver (Fig. 1I) and lung of moribund NPM1c⁺ mice. Immunohistochemistry confirmed GFP expression in BM sections of both control and NPM1c⁺ mice but cytoplasmic expression of NPM1c only in NPM1c⁺ mice (Fig. 1J).

Disease development in NPM1c⁺ mice was further examined by flow cytometry. By the time NPM1c⁺ mice started to lose weight and become sick, there was a marked increase in the percentage and number of human GFP⁺CD33⁺ myeloid cells in the BM that were also CD13⁺, whereas GFP⁻CD33⁺CD13⁺ myeloid cells did not expand (Fig. 2A, 2B). The GFP⁺CD45⁺ human cells in NPM1c⁺ mice were positive for CD13, CD33, CD47, and CD38, modest for CD11b and CD14, and low or negative for CD34 (Fig. 2C). A small but distinct fraction of GFP⁺CD45⁺ human leukocytes in the BM of NPM1c⁺ mice expressed the LSC markers CD123 and CD38 but minimal CD34 (Fig. 2C, 2D). A higher fraction of GFP⁺CD123⁺CD34⁺ cells were in G0/G1 phase as compared with GFP⁺CD123⁻CD34⁻ cells (Supplemental Fig. 1).

The shorter life span of NPM1c⁺ mice suggests that the human myeloid cells are aggressive leukemic cells. To test this, we adoptively transferred total BM cells from moribund NPM1c⁺ mice into 6- to 8-wk-old NSG recipients. Because human IL-3 and GM-CSF are known to enhance AML engraftment, we expressed these cytokines in sublethally irradiated NSG recipients prior to engraftment (14). As with AML cells from patients, secondary mice became sick and died only when recipient mice were irradiated and expressed human IL-3 and GM-CSF, and not all secondary mice developed disease (Table I). Flow cytometry and histology analyses showed that most of the cells in the spleen and BM of the diseased secondary mice were human GFP⁺CD33⁺CD13⁺ leukemic cells (Fig. 2E, 2F). Because a small fraction of human T cells, B cells, and NK cells were also GFP⁺, we isolated GFP⁺CD123⁻CD34⁻ cells from the BM of moribund NPM1c⁺ mice and transferred them into NSG recipients, but no engraftment or disease was observed (Table I). Together, these data show that enforced expression of NPM1c in human CD34⁺ HSPCs is sufficient to drive the development of myeloid leukemia in mice.

We assayed reconstitution of human immune cells in NPM1c⁺ mice. By 9 wk of age, CD3⁺ (GFP⁻) T cells were detected in the peripheral blood of both control and NPM1c⁺ mice (Fig. 3A). In moribund NPM1c⁺ mice, both CD3⁺ T cells and CD19⁺ B cells were detected in the spleen and BM at significant levels (Fig. 3B, 3C). CD56⁺CD3⁻ NK cells were also detected in the spleen and BM but at lower levels. Most of the T cells, B cells, and NK cells were GFP⁻, indicating their development from nontransduced, normal HSPCs. Thus, NPM1c-transduced HSPCs give rise to myeloid leukemia; the nontransduced HSPCs give rise to autologous human immune cells in the same mice.

NPM1c⁺ mice with human AML and autologous immune system are ideally suited to evaluate the efficacy and mechanism of action of immune-based therapies. We tested a BFC in which one arm binds to CD3 and the other arm binds to CD123, therefore redirecting T cells to kill CD123⁺ LSCs. NPM1c⁺ mice were given 1 μg of BFC daily for 7 d and bled 2 d before BFC treatment (day −1) and 1 (day 8) and 10 (day 17) d after BFC treatment (Fig. 4A). The level of GFP⁺CD123⁺ LSCs and hCD45⁺CD3⁺ T cells in the blood of each mouse were analyzed by flow cytometry and normalized to pretreatment levels in the same mouse. As shown in Fig. 4B, the percentage of human CD45⁺GFP⁺CD123⁺ cells was unchanged in NPM1c⁺ mice following PBS injection. In contrast, the percentage of human CD45⁺GFP⁺CD123⁺ cells decreased significantly (∼2-fold) on day 8 following BFC treatment. The percentage of human CD45⁺GFP⁺CD123⁺ cells was still lower on day 17, although not statistically significant. There was no significant change in the percentages of human CD45⁺CD3⁺ T cells following either PBS or BFC injection (Fig. 4C).

We investigated the requirement for T cells in mediating the effect of BFC. Administration of an anti-CD3 Ab, OKT3, 2 d before BFC treatment led to a complete depletion of T cells on day 8 in the blood and abolished BFC-mediated depletion of human CD45⁺GFP⁺CD123⁺ cells (Fig. 4B, 4C). We also enhanced T cell reconstitution by expressing human IL-7 10 d before BFC treatment (Fig. 4C). As a result, a more severe depletion (∼5-fold) of human CD45⁺GFP⁺CD123⁺ cells was detected on day 8 following BFC treatment (Fig. 4B). When BM cells were analyzed on day 8, significantly fewer CD45⁺GFP⁺CD123⁺ LSCs were detected in BFC-treated NPM1c⁺ mice than in mice treated with a control CD3/KLH BFC, in which one BFC arm binds to CD123 and the other to KLH (Fig. 4D). Although the percentages of CD3⁺ T cells in the blood did not change significantly following BFC treatment, the proportions of CD45RA⁺ naive T cells were significantly decreased whereas the proportions of CD45RO⁺ effector/memory CD8 T cells were increased (Fig. 4E), consistent with previous reports (19, 20). Thus, CD123/CD3 BFC can activate T cells to eliminate CD123⁺ LSCs in NPM1c⁺ mice.

To further determine if the BFC redirects T cells to eliminate tumor cells, we performed in vitro killing assays. Naive T cells were purified from the peripheral blood and spleens of NPM1c⁺ mice and incubated with autologous BM cells, of which a majority were leukemic cells, at an E:T ratio of 5:1 in the presence or absence of CD123/CD3 BFC. No significant change in viability of GFP⁺CD123⁺ LSC was observed when BM cells were incubated with T cells in the absence of BFC (Fig. 4F). However, in the presence of BFC, a significant reduction in human CD45⁺GFP⁺CD123⁺ cells was detected. Correspondingly, an increase in expression of CD107a, a marker for T cell degranulation, was seen, which was further enhanced at 48 h (Fig. 4G). These data suggest direct killing of leukemic cells by T cells in the presence of BFC and validates the functionality of T cells in NPM1c⁺ mice.

To understand the molecular mechanisms underlying NPM1c-mediated tumorigenesis, we performed transcriptome analysis on bulk leukemic cells and LSCs. GFP⁺CD33⁺ bulk leukemic cells and GFP⁺CD123⁺CD33⁺ LSCs were purified by cell sorting from the BM of three NPM1c⁺ mice generated with three different human CD34⁺ donor cells (Supplemental Fig. 2A). RNA was isolated from the sorted cell populations (>85% purity), converted into cDNA, and sequenced. Each sample yielded 36–65 million reads with <1.4% reads from rRNAs (Supplemental Fig. 2B), indicating high quality of RNA sequencing (RNAseq). Unsupervised hierarchical clustering showed that LSC populations were similar to each other and clustered together. Similar clustering was observed for leukemic cells (Fig. 5A). Paired analysis uncovered 486 genes that were upregulated 2-fold or more in LSCs (p < 0.05) and 465 genes that were upregulated 2-fold or more in bulk leukemic cells (p < 0.05) (Supplemental Tables I and II). Gene set enrichment analysis showed that genes upregulated in bulk leukemic cells were enriched in those involved in cell cycle and DNA replication (Fig. 5A), consistent with previous reports (21).

To assess the similarity between the transcriptomes of patient AML and de novo AML from two published datasets (22, 23) with leukemic cells from NPM1c⁺ mice, we assessed the similarity between genes upregulated in each of the published datasets with each other and with bulk leukemic cells from our analysis. Supplemental Fig. 2C lists genes whose transcript level was found to be upregulated by 2-fold or more from LSCs to bulk leukemic cells in either the Verhaak or Alcaley dataset. Among the upregulated genes in the Verhaak dataset, 31% were upregulated in the Alcalay dataset, and 65% were upregulated in bulk leukemic cells from NPM1c⁺ mice (Supplemental Fig. 2C). Among the upregulated genes in the Alcalay dataset, 50% were upregulated in the Verhaak dataset, and 81% were upregulated in bulk leukemic cells from NPM1c⁺ mice. Eight genes were upregulated in all three datasets including HOXA9, which is part of the HOX gene signature of NPM1c⁺ AML in humans. Gene set enrichment analysis confirmed the upregulation of several HOX family members in leukemic cells from NPM1c⁺ mice (Supplemental Fig. 2D). qRT-PCR analysis validated the significant upregulation of HOXA5, HOXA6, and HOXA9 in GFP⁺CD33⁺ BM cells from NPM1c⁺ mice as compared with control mice (Fig. 5B). Thus, the strong similarities in transcription profiles suggest that the de novo generated myeloid leukemic cells in NPM1c⁺ mice are similar to NPM1c⁺ AML in patients.

The transcript for the Myc oncogene was more highly expressed in bulk leukemic cells than in LSCs (Supplemental Table II). Myc has been implicated in NPM1c⁺ AML (9) but its role is not well understood. We confirmed a higher level of Myc transcript in purified bulk leukemic cells than in LSCs from NPM1c⁺ mice using qRT-PCR (Fig. 5C). Both GFP⁺CD33⁺ cells and GFP⁺CD123⁺CD33⁺ cells from NPM1c⁺ mice expressed a higher level of Myc than the corresponding cell populations from control mice (Supplemental Fig. 2E). We tested the sensitivity of these cell populations to Myc inhibition using the indirect Myc inhibitor JQ1 (24). Following incubation with 1 μM of JQ1 for 48 h, only ∼15% of GFP⁺CD123⁺CD33⁺ cells and ∼40% of GFP⁺CD33⁺ cells from NPM1c⁺ mice were viable (Fig. 5D), whereas ∼90% of these cells were viable from control mice. To test if JQ1 sensitizes leukemic cells to CD123/CD3 BFC treatment, BM cells from NPM1c⁺ mice were cultured with or without JQ1 overnight and then incubated with autologous T cells with or without CD123/CD3 BFC. The viability of GFP⁺CD123⁺ LSCs was decreased by 70% following BFC treatment alone (Fig. 5E). When BM cells were pretreated with 1 μM JQ1, the viability decreased further to ∼80% without affecting CD107a degranulation of T cells (Fig. 5F). These data suggest that upregulation of Myc may play a role in the survival of leukemic stem cells from NPM1c⁺ mice, although it is not clear whether JQ1 enhances BFC or is just additive.

We report de novo induction of human myeloid leukemia in mice in the presence of an autologous human immune system. In humanized mice with NPM1c⁺ leukemia, referred to as NPM1c⁺ mice, disease is driven by the expression of a frequently found human mutation, NPM1c, in human HSPCs. The de novo induced AML in mice recapitulates features of the human disease: anemia, perturbed hematopoiesis, presence of leukemic blasts in the BM and blood, and infiltration of leukemic cells into other organs. It is notable that the numbers of human CD45⁺ cells in the BM were significantly reduced in moribund NPM1c⁺ mice as compared with age-matched control mice (Fig. 1H). This could result from the large size of leukemic blasts in the BM (Fig. 1F) or the poor supporting environment of the diseased BM. Leukemic cells are CD33- and CD13-positive with minimal expression of CD34 and share similar transcriptional profile as patient NPM1c⁺ AML. A distinct fraction of leukemic cells also express LSC markers CD123 and CD38. In addition, NPM1c⁺ mice express a full complement of human immune cells, including CD4⁺ and CD8⁺ T cells, B cells, and NK cells. The de novo development of human AML with an autologous immune system makes this model unique for studying mechanisms of human leukemogenesis and tumor-immune system interaction. The proof-of-principle approach should open the possibility to model other human hematologic and solid tumors with an autologous immune system.

In this model, enforced expression of NPM1c alone in human HSPCs leads to rapid development of myeloid leukemia with 100% penetrance, although there was variation in the survival of NPM1c⁺ mice (an average survival of ∼100 d following HSPC engraftment) generated using different donor HSPCs and from different lentiviral transductions. The difference seems to correlate with the level of HSPC transduction; mice engrafted with higher percentages of NPM1c-transduced HSPCs became moribund and died at younger age than mice engrafted with lower percentages of NPM1c-transduced HSPCs. These observations are consistent with induction of leukemogenesis by enforced expression of NPM1c rather than stochastic insertional mutagenesis. In our study, NPM1c-transduced HSPCs were engrafted into NSG recipient mice, which are NSG without transgenic expression of human cytokines. Their engraftment into MISTRG recipients (25), which are immunodeficient mice (Rag2^−/− ILR2γ^−/−) expressing human cytokines M-CSF, IL-3, GM-CSF, and thrombopoietin required for myeloid cell development, would be expected to lead to even more rapid induction of AML. In contrast, systemic expression of human NPM1c in mice leads to myeloproliferation (26) and restricted expression of human NPM1c in the mouse hematopoietic compartment leads to AML and B cell malignancies with poor penetrance and long latencies (15). Disease latency is significantly shortened by knocking in the FLT3-ITD, another commonly occurring genetic lesion in AML (27). Thus, expression of human oncogenic lesion NPM1c in human but not mouse HSPCs is sufficient to drive AML development, suggesting intrinsic differences in cellular context between humans and mice.

Our transcriptional analysis further sheds light on how the enforced expression of NPM1c alone in human HSPCs drives development of AML. Most of the genes upregulated in bulk leukemic cells are involved in cell cycle and DNA replication, consistent with a tumorigenic phenotype. As observed in NPM1c⁺ AML in patients, a characteristic HOX gene signature is observed in the de novo NPM1c⁺ AML reported in this article. In addition, we found that Myc is upregulated in bulk leukemic cells. NPM1c is known to prevent the degradation of Myc (9), which in turn positively regulates (endogenous) NPM1 transcription (10). The positive feedback regulation could have contributed to leukemogenesis in our model. Consistently, NPM1c⁺ LSC and leukemic cells are more sensitive to Myc inhibition. Although detailed mechanisms of leukemogenesis by enforced NPM1c expression have yet to be elucidated, our de novo AML model is ideal for exploring early events of human leukemogenesis.

We demonstrate as a proof-of-principle the efficacy of a CD123/CD3 BFC in eliminating CD123⁺ LSCs both in vivo and in vitro in a T cell–dependent manner. These data demonstrate the functionality and responsiveness of human T cells in this system and provide support that humanized mouse models can be used to dissect the mechanisms of action of immunotherapies. Although T cells from humanized mice, including those that develop in the presence of human thymic grafts (28, 29), may not be fully functional, our data suggest that CD3-dependent activation and target cell elimination by T cells occur reproducibly in humanized mice. The relatively short-term effect (i.e., significant decrease of GFP⁺CD123⁺ LSCs at day 8 but not day 17 and an absence of any prolonged survival by NPM1c⁺ mice following BFC treatment) is likely due to the short half-life of BFC, which lacks an Fc domain. In summary, our de novo AML model, in which a normal human immune system and leukemia coexist, should facilitate the assessment of efficacy of immunotherapeutics prior to clinical testing.

We thank Amanda Hanson and Starsha Kolodziej for technical assistance. We also thank the Koch Institute Swanson Biotechnology Center and core facilities for assistance with flow cytometry data acquisition, histology and immunohistochemistry (Dr. Roderick Bronson for analysis of histology slides), and RNAseq data acquisition and analysis.

The authors have no financial conflicts of interest.