Mutant nucleophosmin and cooperating pathways drive leukemia initiation and progression in mice.

Acute myeloid leukemia (AML) is a molecularly diverse malignancy with a poor prognosis whose largest subgroup is characterized by somatic mutations in NPM1, which encodes nucleophosmin. These mutations, termed NPM1c, result in cytoplasmic dislocation of nucleophosmin and are associated with distinctive transcriptional signatures, yet their role in leukemogenesis remains obscure. Here we report that activation of a humanized Npm1c knock-in allele in mouse hemopoietic stem cells causes Hox gene overexpression, enhanced self renewal and expanded myelopoiesis. One third of mice developed delayed-onset AML, suggesting a requirement for cooperating mutations. We identified such mutations using a Sleeping Beauty transposon, which caused rapid-onset AML in 80% of mice with Npm1c, associated with mutually exclusive integrations in Csf2, Flt3 or Rasgrp1 in 55 of 70 leukemias. We also identified recurrent integrations in known and newly discovered leukemia genes including Nf1, Bach2, Dleu2 and Nup98. Our results provide new pathogenetic insights and identify possible therapeutic targets in NPM1c+ AML.

Nucleophosmin, has roles in several cellular processes including ribosome biogenesis and centrosome duplication5-6, for which it relies on its ability to shuttle between the nucleolus, nucleus and cytoplasm using subcellular localization signals7. This ability is impaired in 35% of AMLs, as a result of NPM1c mutations1 which disrupt the N-terminal nucleolar localization signal of Nucleophosmin and generate a nuclear export signal in its place1,8. NPM1c mutations are mutually exclusive of fusion genes found in other types of AML9, but frequently co-occur with activating mutations in FLT31 or RAS10.

NPM1c is known to bind to and alter the subcellular distribution of several proteins including HEXIM1, Fbw7γ, p19Arf and NF-kappaB8, however the relevance of these interactions to AML is unknown8. Selected transgenic mouse lines overexpressing NPM1c in myeloid progenitors display an increased incidence of mild myeloproliferative syndromes, but the significance of this observation is unclear as these mice do not develop AML11.

To study the hemopoietic effects of NPM1c, we generated a conditional knock-in mouse model of the commonest form of NPM1c mutation, type A1. We confirmed that the human (NPM1cA) and “humanized” mouse (Npm1cA) type A mutant proteins (Supplementary figure 1) displayed the same sub-cellular localization (Figure 1a) and proceeded to modify the Npm1 locus in mouse embryonic stem (ES) cells. The conditional allele, Npm1, was designed to minimize interference with the native locus, as this could itself be leukemogenic12, yet switch to the mutant allele, Npm1, after Cre-loxP recombination (Figure 1b). We confirmed that after Cre-loxP recombination Npm1 ES cells expressed the Npm1cA mutant mRNA and protein (Figure 1c,d), and established the Npm1 allele in mice which were born at Mendelian ratios. However, the Npm1 allele was incompatible with normal embryonic development as crosses between Npm1 mice and mice heterozygous for Stella-Cre, which mediates Cre-loxP recombination in the early embryo (PL, unpublished), gave no double transgenic live offspring (0/80) or embryos at E8 (n=11), E10 (n=10) and E12 (n=23). By contrast, the offspring of Npm1 − Mx1-Cre crosses were born at Mendelian ratios (Figure 1e).

Figure 1

Conditional mouse model of type A NPM1c mutation

a, N-terminal GFP-fusions of type A human (NPM1) and “humanized” mouse (Npm1) mutants show identical sub-cellular distribution. b, The conditional Npm1 allele interferes minimally with the native locus and converts to mutant Npm1 allele by Cre (humanized mutant exon 11* in red; mouse exon 11 is homologous to human exon 12). c-d, Npm1 RNA and protein detected in Post-Cre ES cells by RT-PCR (c) and Western Blot (d) respectively. e, Universal lethality of Npm1 (** p=0.0001) versus mendelian ratios for Npm1 mice.

P, PstI site; PuΔTK, Puro-Delta-TK cassette; K562-Npm1, K562 cells transfected with Npm1 or Npm1 cDNA; RT, Reverse Transcriptase, RT-PCR primers: R1, forward; R2, Npm1 reverse; R3, Npm1 reverse (red).

To study the hemopoietic effects of Npm1, 5- to 8-week-old Npm1 and Npm1 mice (hereafter collectively referred to as Npm1) and Npm1 mice (hereafter referred to as Npm1), were treated with Polyinosinic-Polycytidylic acid (pIpC) and analyzed 8 weeks later. Cre-loxP recombination was seen in >90% of bone-marrow-derived hemopoietic colonies from Npm1 mice (Supplementary Figure 2), reflecting efficient recombination in hemopoietic stem cells (HSCs) and Npm1cA protein was detectable in Npm1 hemopoietic tissues (Figure 2a). Gene expression profiling of Npm1 compared to Npm1 lineage negative marrow progenitors (Lin−), showed differential overexpression of HoxA5, HoxA7, HoxA9, HoxA10 and HoxB5 genes (Figure 2b and Supplementary Table 1). Similar gene expression changes were seen with total bone marrow nucleated cells (BMNC) and B220+ cells, but not Gr1+/Mac1+ cells (Supplementary tables 2-4). For Npm1 BMNC HoxA9, HoxA7 and HoxA5 were the 3 most significantly overexpressed mRNAs in the mouse genome and expression of several lymphoid-specific genes was reduced (Supplementary table 2). Total Npm1 mRNA expression was unaltered (Supplementary tables 1-4 and Supplementary figure 3).

Figure 2

Hematopoietic changes and incidence of AML in Npm1 mice

a, Hemopoietic expression of Npm1cA protein in Npm1 mice. b, Hox-overexpression in Npm1 vs Npm1 lineage negative hemopoietic progenitors. c,d, No significant differences in white cell (WCC), hemoglobin (Hb) or platelet counts (Plts), but higher mean red cell (MCV) and platelet (MPV) volumes in Npm1 mice. e,f, No differences in marrow stem and progenitor cell compartment sizes. g,h, Expansion of mature myeloid (Gr1+/Mac1+) and reduction in late (B220+/CD19+) B-cells in Npm1 vs Npm1 marrow. i, Summary of hemopoietic changes in Npm1cA/+ mice. j, Increased serial re-plating of Npm1cA/+ myeloid progenitors. k, Decreased survival of Npm1 mice due to excess AML. l, Example of AML showing splenomegaly due to infiltration with myeloperoxidase-positive blasts (Sp), also infiltrating the liver (Li) and blood (Bld).

Error bars show the standard error of the mean; *p<0.05; B,T, B or T-cell leukemia/lymphoma; Non-Hem, Non-hematological.

Npm1 and Npm1 blood counts did not differ (Figure 2c); although Npm1 mice had increased mean red cell (MCV) and platelet (MPV) volumes (Figure 2d). Bone marrow histo-morphological examination revealed no detectable abnormalities/differences (Supplementary Figure 4a). Flow cytometric analysis of marrow cells showed no significant differences in stem or progenitor cell compartment sizes (Figure 2e-f and Supplementary Figure 4b). However, Npm1 mice had increased numbers of Gr1+/Mac1+ myeloid cells and decreased numbers of late B-cells (B220+/CD19+) (Figure 2g-i). Additionally, Npm1cA/+ cells showed increased serial re-plating ability in methylcellulose, an in vitro surrogate for self-renewing potential13 (Figure 2j). There were no differences in levels of Gr1+/Mac1+ cell apoptosis or DNA damage (Supplementary Figure 4c-e).

To study the leukemogenicity of Npm1cA, we aged 43 Npm1 and 44 Npm1 mice. Npm1 mice had a shorter overall survival compared to Npm1 animals (617 vs 769 days, p=0.018) as a result of excess deaths due to AML (13 vs 0, p=0.0001) (Figure 2k-l). Morphologically AMLs showed maturation (“myeloid leukemia with maturation”14) and all 5 AMLs tested were Gr1+/Mac1+. Of 13 Npm1 mice with AML, 6 were found to have coincidental non-extensive B-lymphoid tumors. Compared to Npm1cA/+ mice dying of non-hematological causes, mice with AML had enlarged livers (2.6g vs 1.8g, p<0.001) and spleens (1.3g vs 0.3g, p<0.001) and higher blood leukocyte counts (77.1 vs 7.0 ×106/l, p=0.006)(Supplementary table 5). Also 2/2 AMLs tested were transplantable into sublethally irradiated syngeneic mice (Supplementary table 6). Finally, Npm1 and Npm1 mice developed lymphoid tumors at the expected rate for their age/strain15 and had similar rates of non-hematological mortality (Figure 2k and Supplementary table 5).

The above data show that Npm1c can initiate AML, however the long latency reflects the need for additional mutations. To identify cooperating mutations, we subjected Npm1 mice to insertional mutagenesis with Sleeping Beauty (SB)3. We generated transgenic mice carrying approximately 80 copies of GrOnc, a novel bi-functional PiggyBac/SB transposon16 capable of both gene activation and disruption (Figure 3a-b and Supplementary Figure 5). GrOnc harbors the Graffi1.4 murine leukemia virus LTR, which preferential promotes myeloid rather than lymphoid leukemia in predisposed backgrounds, an effect attributed to the LTR17.

Figure 3

Npm1 and the GrOnc transposon synergize to cause AML

a, The GrOnc transposon carrying gene-activating and inactivating elements flanked by repeats for the Sleeping Beauty (SB) and PiggyBac (PB). b, Blood-metaphase FISH showing the genomic location of the GrOnc donor locus on chromosome 19. c, Acceleration of leukemogenesis in Npm1 compared to Npm1 mutagenised mice. d, Marked increase in the proportion of AMLs and absence of T cell tumors in Npm1 compared to Npm1 mice. e, Southern blot showing clonal transposon integrations in mouse leukemias. Endogenous En2 (arrowhead) and GrOnc donor locus concatamer (dots) bands indicated. f, Morphology and immunohistochemistry (anti-myeloperoxidase) from an Npm1 AML.

Gr1.4 LTR, Graffi1.4 MuLV long terminal repeat; SD, Lun splice donor; En2-SA and βA-SA, Engrailed 2 and Carp β-actin splice acceptor; pA, adenovirus polyadenylation signal; *p<0.00001; B, T and U, B-cell, T-cell and undifferentiated leukemia/lymphoma; Bld, blood; Sp, spleen; Sar, myeloid sarcoma; Ki, kidney; Li, liver.

We crossed Npm1, GrOnc, Mx1-Cre and conditional Rosa26 transposase (Supplementary Figure 6) mouse lines as per Supplementary Figure 7a to generate 125 Npm1 (Npm1 mutagenesis cohort), and 45 Npm1 (Npm1 mutagenesis cohort). All mutagenised mice developed aggressive leukemia/lymphoma within one year of pIpC induction (Figure 3c). However, mean survival was significantly shorter in the Npm1 compared to Npm1 mice (99 vs. 150 days, p=0.001) (Figure 3c). The first 121 mice (87 Npm1 and 34 Npm1) were studied in detail and tumors classified by histological type (Supplementary Table 6), confirmed by flow cytometry in 18 leukemias. Compared to Npm1, Npm1 mice had a significantly higher fraction of AMLs (80.5% vs. 26.5%, p=0.00000002) and no T-cell leukemias (0% vs. 17.6%, p=0.00006) (Figure 3d-f). These findings indicate strong synergy between Npm1 and GrOnc in promoting myeloid leukemogenesis. In some mice a coincidental second tumor was identified (leukemia/lymphoma in 19.8% and angiosarcoma in 3.3%) (Supplementary Table 7).

In order to identify the genetic mutations collaborating with Npm1c to cause AML, we mapped the transposon integration sites in hemopoietic tumors from 87 Npm1 and 34 Npm1 mice using barcoded splinkerette PCR and 454-sequencing. A total of 219,556 high-quality reads mapped evenly throughout the genome although, as expected for SB3-4, we observed local hopping around the GrOnc donor site at Megabase (Mb) 30 on chromosome 19 (Supplementary Figure 8). CISs for Npm1 and Npm1 tumors were identified using the kernel convolution method18 and those mapping within 2Mb of the donor site discarded. The only other CIS on chromosome 19 involved Pten located at 32.8Mb, which makes its significance uncertain. The CIS sets in Npm1 and Npm1 tumors included several known and many putative novel leukemia genes (Figure 4a, Supplementary Figures 8 and 9). The two sets overlapped but were clearly different, indicating that distinct molecular pathways operated in the two groups.

Figure 4

Common integration sites in transposon-derived leukemias

a, Common integration sites in Npm1cA/+ and Npm1+/+mice show some overlap (regular font) but are mostly different (bold). b-d, Directional GrOnc integrations at the Csf2 locus identified in 42 of 70 AMLs, were associated with the formation of two LunSD-Csf2 fusion mRNA splice variants (sv1 and sv2) and marked overexpression of Csf2 mRNA (note break in Y axis). e, GrOnc integrations in three individual myeloid blast colonies from one of these AMLs shared only 2 insertions, involving Csf2 and the myeloid oncogene Myst4(Moz) suggesting that these were “driver” insertions. e, Directional activating integrations in intron 9 of Flt3. g, Bi-directional integrations in Nup98.

In the Npm1 group, the most striking finding was the identification of activating integrations upstream of Csf2, the gene for Gm-csf, in 48.3% of leukemias, associated with LunSD-Csf2 fusion transcripts, Csf2 mRNA overexpression (Figure 4b-d) and increased levels of Gm-csf in leukemic cell supernatants (p=0.0007, Supplementary Table 8). We studied transposon integrations in single-cell derived methylcellulose colonies from three Csf2+ AMLs and found that over 90% of CFU colonies (11/12, 12/12 and 19/20) carried Csf2 insertions. We went on to map transposon integrations in 3 such colonies from the AML 36 and found that individual colonies carried 67-90 independent integrations each, yet only two integrations were shared between them, those involving Csf2 and the myeloid oncogene Myst4 (Moz) (Figure 4e). This is the first study of the subclonal make-up of transposon-derived malignancies and shows that, if the transposase remains active, transposons are continuously re-mobilized generating a highly heterogeneous population; bar for a small number of key insertions, possibly the leukemia “drivers”. We went on to study transposon integrations in leukemias developing in irradiated recipients transplanted with AMLs 26 and 38. We found that Csf2 insertions persisted in all 5 recipient mice studied (Supplementary Table 5), whilst the vast majority (>98%) of other insertions were lost. These data strongly indicate that Csf2 insertions are critical to the growth of these leukemias and thus persist during leukemic evolution/propagation, whilst “passenger” insertions are lost. Csf2 can behave as a myeloid oncogene in mice when activated by endogenous retroviruses19, whilst autocrine Gm-csf production has been reported to drive the growth of human AML blasts in-vitro 20-21 and this is usually, albeit not always, inhibited by anti-GM-CSF antibodies.

Other important CISs included Flt3 (Figure 4f), a gene frequently co-mutated in human NPM1c+ AML and Nup98 (Figure 4g), a key component of the nuclear pore and a leukemogenic fusion gene partner not been previously identified as a retroviral or transposon CIS. NUP98 fusion proteins can directly interfere with nucleo-cytoplasmic transport of NPM1c22, raising the possibility that Nup98 insertions may operate in a similar way hence their specificity to Npm1cA/+ AMLs. Other CISs included inactivating GrOnc insertions in signal transduction inhibitors such as Nf1, Ptpn1 and Ptptn2 and insertions in transcription factors including Bach2, Cnot1, Pax5 and Zfp521 (Figure 5a and Supplementary figure 10).

Figure 5

A model for Npm1cA/+-driven leukemogenesis

a, Co-occurrence table of CIS genes in 70 NPm1cA/+ AMLs reveals mutually exclusive insertions in Csf2, Flt3, Rasgrp1 and Kras. b, Persistence of Hox overexpression in Npm1 vs Npm1 AMLs indicating that this effect of Npm1cA persists in leukemic cells(* p≤0.03). c. Model for Npm1cA - driven leukemogenesis supported by our data: isolated Npm1 increases self renewal but leukemic transformation only occurs upon activation of defined proliferative pathways (Type I mutation), facilitated by a permissive transcription factor (TF) mutation.

Error bars show the standard error of the mean. Localization of Npm1 protein indicated in green.

Co-occurrence analysis of CIS genes from Npm1 AMLs, showed that Csf2 integrations were mutually exclusive of integrations in Flt3 (p=0.002) and Rasgrp1 (p=0.008) (Figure 5a and Supplementary figure 11). This was also true for Kras but did not reach statistical significance (p=0.12). These observations suggest that these genes provide alternative proliferation signals that complement Npm1c in promoting cellular transformation. Interestingly Rasgrp1, previously associated with T-lymphoblastic leukemia in retroviral mutagenesis screens23, was recently identified as a drug resistance gene in another mouse model of AML24. Hox overexpression persisted in Npm1c/+ transposon-derived AMLs, irrespective of transposon insertions, suggesting that the molecular effects of Npm1c mutations remained operative in leukemic cells (Figure 5b). HoxA7 and HoxA9 are frequently targeted by retroviruses in BXH2 myeloid leukemias25 and mediate leukemic transformation by MLL oncoproteins26. Also, in line with our findings, HoxA9 mice exhibit defects in maturing myeloid and B-lymphoid cells, but not early progenitors27. Overall, this makes it highly plausible that Hox overexpression mediates the leukemogenic effects of NPM1c, particularly as expression levels seen in our mice are comparable to those in human AMLs.

Taken together, our data demonstrate that Npm1c are AML-initiating mutations which cause Hox overexpression, impart increased self-renewal to and prime hemopoietic stem/progenitor cells to leukemic transformation by activation of a narrow set of pro-proliferative molecules/pathways, usually in combination with mutations in transcriptional regulators (Figure 5c, Supplementary Tables 1-3). Observations from whole genome sequencing studies are generating data in support of such a model in this and other subgroups of AML with a normal karyotype10,28. Also, our data explain other important features of human NPM1c+ AML, including a consistent negativity for the primitive marker CD341 (effects of Npm1cA most noticeable on later progenitors) and a failure to observe NPM1c mutations in the human germline (embryonic lethality). Finally, our approach to define the effects of a mutation in isolation and proceed to map its collaborative oncogenic pathways provides a model for the study of other human cancer-associated mutations and can be used to complement and help decipher recent and impending whole cancer genome sequencing studies.

Methods

Primers/Oligonucleotides

For all sequences see Supplementary table 9

Transient transfections

Amino-terminal GFP-NPM1/GFP-Npm1 fusion constructs were generated and K562 cells photographed 24hrs after electroporation.

Npm1 ES cells and mice

A 9.8kb was retrieved from BAC bMQ-282D14 using oligonucleotide recombineering29 and modified to the Npm1 targeting construct (duplicated region: NCBIM37-chr11:33051153-33053141). Npm1 (Post-Cre) ES cell clones were derived under FIAU selection30. Npm1 and Npm1 mice were crossed with Stella-Cre for embryonic viability and Mx1-Cre31 mice for hemopoietic studies. For Mx1-Cre activation 500μg pIpC x6 was used.

Stella-Cre mice

Generated by WW and PL by insertion of IRES-Cre into the 3'UTR of Stella, they show 100% Cre-loxP recombination at several loci in embryos and germline.

GrOnc mice

Generated by pronuclear injection of linearized GrOnc DNA.

Rosa transposase mice

Generated as per Supplementary Figure 6.

Mouse genotyping

We used PCR assays with primers P1-P16 (e.g. Supplementary Figures 2, 6, 7).

Flow Cytometry & Cell Sorting

Mouse femurs and spleens were processed as described32. For BM lineage analysis we stained with Gr1-PE and CD11b(Mac1)-FITC, or B220-APC-Alexa750 and CD19-PE. For HSC/progenitor analyses, we depleted lineage-positive cells using MACS columns and stained with cKit-APC, ScaI-PB, CD34-FITC and either Flt3-PE (HSC) or FcRIII-PE (progenitors). Antibodies from BD Pharmingen except CD11b-FITC (Caltag). We used a CyAn-ADP analyzer and FlowJo. B220+ and Gr1+/Mac1+ cells were sorted using a MoFlo sorter. For leukemia phenotyping we used combinations of the above.

Gene expression profiling

Global profiling was done using Illumina mouse-6 expression beadchip version 2. Data were quantile normalised33 and analyzed using the bioconductor (http://www.bioconductor.org/), lumi (http://www.bioconductor.org/packages/2.0/bioc/html/lumi.html) and limma34 packages, then p-value adjusted for multiple testing35. Microarray data and description of experimental design were deposited to Array Express with accession number E-MEXP-3113.

Hemopoietic colonies

BM or spleen cells were plated in M3434 (Stem Cell Technologies) as described33. Colonies were counted/picked after 10-12 days. For re-plating assays 50,000 BM cells from 6xNpm1cA/+ and 6xNpm1WT mice were plated and counted after 8 days with 30,000 cells re-plated etc.

Southern Blots

Npm1 targeting: PstI-digested ES cell DNA was hybridized with a PCR-generated-probe (Figure 1b). GrOnc copy number and clonal integrations: SacI-digested tail DNA or BamHI-digested leukemic DNA respectively, were hybridized with the XbaI-SacII GrOnc fragment (Supplementary Figure 5).

RT-PCR

For Npm1 and Npm1 cDNAs we used allele-specific primers R1-R3 and for GrOnc-Csf2 primers Graffi1.4-LTR_F and Csf2_exon3_R.

qPCR/qRT-PCR

For GrOnc copy number En2-SA and β-Actin were quantified from tail DNA. En2-SA/β-Actin ratio was normalized against wild-type mice (2 copies of En2-SA). For Csf2 qRT-PCR we designed specific primers/probe and normalized against Gapdh. For Hox genes, we used standardized assays (Applied Biosystems), normalized against Gapdh and expressed values relative to Npm1.

Anti-Npm1cA specific antisera and Western Blots

Anti-Npm1cA-specific rabbit antisera, were generated and affinity-purified using peptide Hydrazine-IQDLCLAVEEVSLRK and used at 1/100. Mouse monoclonal anti-Actin (Sigma) was used at 1/5000.

FISH analysis

The location of the 80-copy GrOnc donor site, was identified using Texas Red-labeled GrOnc DNA on blood metaphases.

Hematological measurements

Blood counts were performed on a VetABC analyzer (Horiba ABX).

Splinkerette PCR and sequencing

These were done as described36 using SB-specific primers and 454-sequenced.

Identification of common integration sites (CISs)

ssaha2 was used to map GrOnc insertions to mouse genome. Sequences were filtered to contain the SBcommon-Sp2-454R primer then the end of the SB repeat, followed by genomic sequence starting with TA and ending at GATC (MboI restriction site). Sequences <17 nucleotides or with <85% maximal genomic identity were discarded. For sequences matching several loci, we calculated a normalized score difference (NSD) comparing the best hit to the second best hit, whereby NSD= ((score of best hit) − (score of second-best hit)) / query-length*100. We mapped 5000 random mouse genomic fragments to the genome and found that 96.5% of correctly and only 1.5% of wrongly mapped reads had NSD ≥4 and removed reads with NSD <4 from analysis. Finally, redundant sequences from the same tumor and same location were “collapsed” to one integration. To identify CISs, non-redundant insertions were analyzed using a Gaussian Kernel Convolution-based framework37 for 10Kb, 30Kb, 60Kb and 120Kb windows and all windows merged to compile CIS lists38. To identify previously described CISs we searched the retroviral tagged cancer gene database (RTCGD, http://rtcgd.ncifcrf.gov)36 and recent insertional mutagenesis publications3-4,39-40.

Histopathology

Formalin-fixed, paraffin-embedded sections were stained with H&E and anti-CD3, anti-B220 and anti-myeloperoxidase, detected by immunoperoxidase. All material was examined by two experienced histopathologists (PW and MA) blinded to mouse genotypes. A primary diagnosis was established and hemopoietic tissues examined for additional malignancies.

Gm-csf ELISA

Cryopreserved AML cells were plated in duplicate at 100,000 per 100ul RPMI/ 10%FCS. Supernatants were collected after 2 days and assayed (EMGMCSF, Thermo).

Leukemia Transplants

Syngeneic mice were sublethally irradiated (500rad) and injected with 1×106 cryopreserved AML cells into the tail vein. Mice were culled when unwell or after 180 days.

Statistics

We used Fisher's exact test for 2×2 comparisons and mutual exclusivity testing, χ2-test for nominal data and t-test for continuous data unless normality test failed, when Mann-Whitney Rank Sum tests were used (Sigma Stat). SEMs are represented in error bars.