To the Editor

Mutation of NPM1 results in cytoplasmic sequestration of NPM1C+ protein that maintains a unique expression pattern of HOXA/B genes and MEIS1 oncogene by directly binding to chromatin target sites to facilitate AML leukemogenesis [1,2,3]. However, molecular mechanisms underlying NPM1C+-driven gene regulatory networks and leukemogenesis remain elusive. Since wild-type (WT) NPM1/NPM1C+-associates with CTCF and may influence CTCF-mediated loops by anchoring for CTCF cluster [4, 5], it has been long speculated that NPM1C+ mislocalization may alter CTCF-mediated gene expression. We test whether mislocalization of NPM1C+ to cytoplasm alters CTCF-mediated three-dimensional (3D) topologically associated domains (TADs) by Hi-C comparing mouse bone marrow (BM) Lin-cKit+ (LK) stem- and progenitor-enriched cells purified from WT and NPM1C+-KI animals induced for NPM1C+ expression for 4-months (Table S1 listed kits, constructs, cells, antibodies). Mice conditionally expressing NPM1C+ in the BM develop a myeloproliferative neoplasm/myelodysplasia-like condition with a long latency characterized by enlarged spleens, leukocytosis, and thrombocytopenia [6, 7]. We confirmed that NPM1C+ cKI aberrantly activated Hoxa/Hoxb and Meis1 genes that coincided with a gain of chromatin accessibility as seen by RNA-seq and ATAC-seq (Fig. S1A, B) [1, 8]. NPM1C+ expression led to an increase of 46 TADs and a decrease of 31 TADs (Fig. 1A). The increased TADs encompassed 434 genes of which approximately 18% (78 genes) overlapped with genes increased in expression upon NPM1C+-cKI measured by RNA-seq (Fig. 1B, top). Theses overlapping genes are mainly NPM1C+-signature genes involved in pathways regulating transcription, anterior/posterior patterning, homeobox genes, and cell cycle (Fig. 1C, top). Top TADs strengthened by the NPM1C+-cKI included Hoxa/b, and Pbx3 loci (Fig. 1D–F). Many of these genes are specifically involved in unique NPM1C+-driven leukemic transcription program [1, 8]. In contrast, the decreased TADs included cell cycle kinase inhibitors (CDKIs), such as Cdkn1a (p21) gene (Fig. 1G). The altered TADs or sub-TADs were defined by CTCF binding when we integrated with public available mouse CTCF ChIP-seq dataset (GSM918744), whereby CTCF bound to the altered TAD or sub-TAD boundaries (Fig. 1D-G, Red arrowheads). As expected, chromatin accessibility in Hoxa/b and Pbx3 loci were coordinately increased (Figs. S1B, 1H). In light of NPM1C+ directly interacting with CTCF and binding to its chromatin targets [2, 3, 5], NPM1C+ and CTCF may coordinate to remodels NPM1C+-signature gene TADs resulting in altered chromatin accessibility and oncogenic gene expression signatures.

Fig. 1: NPM1C+ KI alters hematopoietic TAD resulted in NPM1C+ signature gene expression profile.
figure 1

A Overlapping of mouse BM HS/PC TADs and NPM1C+ KI HS/PC TADs identified by Hi-C assay using WT and NPM1C+ cKI mouse BM LK cells of which NPM1C+ was induced at 8 weeks old animals for 4 months. B Overlapping of enhanced or reduced TAD encompassed gene and upregulated or downregulated genes by RNA-seq upon NPM1C+ KI in BM HS/PCs, respectively. C GO term analysis of overlapped enhanced or reduced TAD encompassed gene and upregulated or downregulated genes by RNA-seq upon NPM1C+ KI in BM HS/PCs, respectively. NPM1C+ was induced at 8 weeks old animals for 4 months. D Hi-C interacting maps in part of mouse chromosome 6 regions containing Hoxa locus comparing WT and NPM1C+ KI mouse BM HS/PCs. Publicly available CTCF ChIP-seq data (GSM918744) in MEL cells was integrated on top with Hi-C data. Note: CTCF binding at the boundaries of altered TADs or sub-TADs (red arrow heads). E Hi-C interacting maps in part of mouse chromosome 11 regions containing Hoxb locus comparing WT and NPM1C+ KI mouse BM HS/PCs. Publicly available CTCF ChIP-seq data (GSM918744) in MEL cells was integrated on top with Hi-C data. Note: CTCF binding at the boundaries of altered TADs or sub-TADs (red arrow heads). F Hi-C interaction map at the Pbx3 locus comparing WT and NPM1C+ KI in BM HS/PCs. Publicly available CTCF ChIP-seq data (GSM918744) in MEL cells was integrated on top with Hi-C data. Note: CTCF binding at the boundaries of altered TADs or sub-TADs (red arrow heads). G Hi-C interaction map at the Cdkn1a locus comparing WT and NPM1C+ KI in BM HS/PCs. Publicly available CTCF ChIP-seq data (GSM918744) in MEL cells was integrated on top with Hi-C data. Note: CTCF binding at the boundaries of altered TADs or sub-TADs (red arrow heads). H ATAC-seq chromatin accessibility of Pbx3 gene locus compared WT and NPM1C+ KI in BM HS/PCs.

Full size image

We next transfected OCI-AML3 cells with vector caring WT NPM1 cDNA to restore nuclear NPM1 (NPM1OE), or treated OCI-AML3 cells with 50 nM XPO1 inhibitor (XPO1i), selinexor, for 3 days to block the NPM1C+ cytoplasmic export, and test if restoration of nuclear NPM1 reversed NPM1C+-associated HOX expression signature and TAD formation. Immunofluorescent staining confirmed that XPO1i or NPM1OE resulted in nuclear localization of NPM1 and enhanced its colocalization with CTCF (Fig. 2A). Nuclear retention of NPM1 reduced MYC, MEIS1, and HOX gene expression, while increased CDKIs, CDKN1A and CDKN2A genes (Fig. 2B), consistent with the transcriptomic GO analysis (Fig. S2A, B).

Fig. 2: Nuclear retention of NPM1 enhanced NPM1 and CTCF interaction and alters leukemic TAD formation and gene expression.
figure 2

A Immunofluorescence staining of NPM1 and CTCF localization and association upon XPO1i treatment or WT NPM1 expression. Note: The NPM1 antibody used recognized both WT and NPM1C+ mutant. B Heat map of RNA-seq analysis shows upregulated and downregulated genes upon WT NPM1 overexpression or XPO1i treatment in OCI-AML3 cells. C Overlapping of TADs identified from WT OCI-AML3 cells and NPM1OE OCI-AML3 cells by Hi-C assay. D Overlapping of enhanced or reduced TAD encompassed gene and upregulated or downregulated genes by RNA-seq upon NPM1OE in OCI-AML3 cells. E Hi-C interacting maps in part of human chromosome 6 regions containing CDKN1A gene comparing WT, XPO1i treated, or NPM1OE OCI-AML3 cells. Publicly available CTCF ChIP-seq data in human K562 cells was integrated on top with Hi-C data (GSM1010820). Note: CTCF binding at the boundaries of altered TADs or sub-TADs (red arrow heads). F Hi-C interacting maps in part of human chromosome 9 regions containing CDKN2A/CDNK2B genes comparing WT, XPO1i treated, or NPM1OE OCI-AML3 cells. G Hi-C interaction map at the HOXB loci comparing WT, XPO1i treated, or NPM1OE OCI-AML3 cells. Publicly available CTCF ChIP-seq data in human K562 cells was integrated on top with Hi-C data (GSM1010820). Note: CTCF binding at the boundaries of altered TADs or sub-TADs (red arrow heads). H Survival of NSG mice (n = 5) transplanted with WT or NPM1OE OCI-AML3 cells. Data are presented as mean ± SD. *p ≤ 0.05; **p ≤ 0.01; ***p ≤ 0.001.

Full size image

Next, we carried out Hi-C and RNA-seq experiments in WT, XPO1i-treated or NPM1OE OCI-AML3 cells, and integrated our Hi-C data with available CTCF ChIP-seq profiles generated in K562 cells (GSM1010820). NPM1OE altered 96 TADs defined by CTCF boundaries. Among those, 55 TADs were decreased, while 41 TADs were increased by NPM1OE (Fig. 2C). Similarly, XPO1i treatment altered 97 CTCF-defined TADs (59 decreased TADs and 38 increased TADs, data not shown). We further integrated the transcriptomic data with Hi-C datasets from XPO1i-treated or NPM1OE OCI-AML3 cells. Upon NPM1OE, 24% of genes residing within up-regulated TADs were transcriptionally up-regulated determined by RNA-seq (Fig. 2D, top), and they were predominantly involved in cell cycle, cell differentiation, apoptosis, DNA damage and repair, and transcription regulation (Fig. S2C, left). Only 2.4% of genes residing within up-regulated TADs were transcriptionally down-regulated (Fig. S2D, Left). On the other hand, 16% of genes encompassed within reduced TADs were significantly down-regulated according to their expression levels (Fig. 2D, bottom), and were enriched in pathways involved in HOX, pathways in cancer, PI3K-AKT signaling, Wnt signaling, positive regulation of ERK1/ERK2, transcriptional regulation, cell cycle, and positive regulation of cell proliferation (Fig. S2C, right). Only 4.5% of genes residing within down-regulated TADs were transcriptionally up-regulated (Fig. S2D, Right). XPO1i treatment reveal very similar TADs and transcriptomic changes (Fig. S2E, F). Thus, NPM1C+-driven TAD alteration may be critical for the NPM1C+ signature gene expression.

In particular, Hi-C data revealed that top ten upregulated TADs encompassed genes encoding CDKIs, p21, p15, p16, and p27, while top ten reduced TADs included HOXB in both XPO1i or NPM1OE (Fig. 2E–G). These altered TADs or sub-TADs are defined by CTCF binding when we integrated with CTCF ChIP-seq data from K562 cells (GSM1010820), whereby CTCF bound to boundaries of the altered TADs or sub-TADs (Fig. 2E–G; Red arrowheads). Interestingly, CDKI genes are regulated by MYC-MIZ-1 transcription axis that is involved in control of LSC proliferation and cell cycle progression [9].

To further support that NPM1 nuclear relocalization is in part inducing genome TAD reorganization, we performed time course experiments by treated OCI-AML3 cells with XPO1i for 0, 1, 3, 6, 24, and 48 h. We observed a significantly increase in nuclear localization of NPM1C+ after 24 h treatment by WB (Fig. S3A) and immunofluorescent-staining (Fig. S3B, C). Time course 3 C assay revealed that CTCF-loops encompassed anterior HOXB genes (HOXB1-HOXB5) were significantly impaired as early as 3 hr upon blocking NPM1C+ cytoplasmic localization, while the CTCF mediated looped interactions encompassing CEBPA/G genes were increased immediately upon blockage of NPM1C+ cytoplasmic translocation (Fig. S3D, Table S2 listed all primers/probes). Coordinately, expression levels of CDKIs and CEBPA genes were only upregulated upon 24 hr and 3 days XPO1i treatment, respectively (Fig. S3E), suggesting that TAD reorganization precede gene expression changes.

It has been proposed that MIZ-1 activates its target genes including CDKN1A and CDKN2B by associating with NPM1 and p300, while in MYC transformed cancer cells, association of MYC and MIZ-1 repels NPM1 and p300 leading to aberrant proliferation [10, 11]. We further tested whether restoration of nuclear NPM1 switches the MIZ-1 transcription circuit from repression to activation of CDKIs by disassociating from MYC in favor of interacting with NPM1. Although MIZ-1 level remains relatively consistent upon XPO1i treatment or NPM1OE, these interventions promoted MIZ-1 interacting with NPM1 in nucleus compared to untreated OCI-AML3 cells (Fig. S4A, bottom). As a consequence, the amount of MIZ-1-associated p300 was increased, while MIZ-1 binding to MYC and G9a, a MYC-associated H3K9-metyltransferase required for repression, was markedly decreased upon nuclear retention of NPM1 (Fig. S4A). RT-qPCR and RNA-seq revealed that expression of several MIZ-1 targets, CDKI and CEBP genes, was significantly increased (Fig. 2B, S3E, S4B). ChIP-qPCR further revealed that MYC and G9a recruitment to these loci was significantly reduced comparing to the constant MIZ-1 chromatin binding (Fig. S4C–E), while recruitment of NPM1 and p300 was significantly increased upon XPO1i treatment or NPM1OE (Fig. S4F, G). As a consequence, active H3K27ac chromatin marks were increased (Fig. S4H) while H3K9me2 repressive chromatin marks were significantly reduced at these MIZ-1 target genes (Fig. S4I) comparing to control IgG (Fig. S4J), However, we cannot distinguish whether WT-NPM1 or NPM1C+ regulates CDKI and CEBP genes upon nuclear retention of NPM1 because the antibody used recognized both forms.

Although NPM1 antibody that we used cannot distinguish WT from mutant protein and their role in CDKI and CEBP gene regulation, analyzing Bru-seq data of WT and XPO1i treated OCI-AML3 cells from recent published data [2] also revealed that the nascent transcripts of CDKN1A and CEBPA genes are consistently enhanced by XPO1i treatment (Fig. S4K), supporting that nuclear restoration of NPM1 reverses aberrant TAD topology and switches the MYC-MIZ-1 axis to restore cell cycle control and myeloid differentiation. However, it remains to be determined how NPM1c+/NPM1 were targeted to distinct chromatin sites to differentially activate oncogenic vs myeloid differentiation pathway. Given that NPM1 interacts with MIZ-1 and chromatin bound CRM1/XPO1 recruits NPM1C+ onto HOX cluster [2, 3, 10, 12]. It is reasonable to suspect that MIZ-1 or CRM1/XPO1 directs NPM1 or NPM1C+ to the intended chromatin targets, respectively. In addition, HOX associated lncRNAs, HOTTIP and HOXBLINC, are aberrantly activated in NPM1C+-AML cells that may also play a role in NPM1C+-driven oncogenic HOX gene expression [8, 13, 14].

Consistent with upregulation of CDKIs along with myeloid TFs, CEBPA and CEBPD, nuclear retention of NPM1 promoted OCI-AML3 cell differentiation seen by increased expression of myeloid-specific surface markers CD11b and CD14 (Fig. S5A–C). Furthermore, nuclear retention of NPM1 consistently increased the proportion of cells in G1 phase at the expense of cell in the S phase and G2/M phase transition (Fig. S5D), which coincided with enhanced apoptosis (Fig. S5E) and decreased cellular proliferation (Fig. S5F). Of note, XPO1i impairs OCI-AML3 cell growth and differentiation in a p53 independent manner [15]. We further test whether restored NPM1 in nuclei of NPM1C+ AML cells affects leukemogenesis and survival of AML mouse models in vivo, we transplanted 1 × 106 WT, NPM1OE OCI-AML3 cells into NSG mice. All animals transplanted with WT OCI-AML3 cells died around 38 days after transplantation, while mice that received OCI-AML3 cells with NPM1OE survived between 54–58 days (Fig. 2H). Thus, nuclear NPM1 localization mitigates leukemogenesis in a xenograft mouse model of AML with NPM1C+, perhaps blocking NPM1C+-driven TAD topology and signature gene expression and restoring cell cycle control and myeloid differentiation.