Haematopoietic malignancies caused by dysregulation of a chromatin-binding PHD finger.

Histone H3 lysine 4 methylation (H3K4me) has been proposed as a critical component in regulating gene expression, epigenetic states, and cellular identities1. The biological meaning of H3K4me is interpreted by conserved modules including plant homeodomain (PHD) fingers that recognize varied H3K4me states. The dysregulation of PHD fingers has been implicated in several human diseases, including cancers and immune or neurological disorders. Here we report that fusing an H3K4-trimethylation (H3K4me3)-binding PHD finger, such as the carboxy-terminal PHD finger of PHF23 or JARID1A (also known as KDM5A or RBBP2), to a common fusion partner nucleoporin-98 (NUP98) as identified in human leukaemias, generated potent oncoproteins that arrested haematopoietic differentiation and induced acute myeloid leukaemia in murine models. In these processes, a PHD finger that specifically recognizes H3K4me3/2 marks was essential for leukaemogenesis. Mutations in PHD fingers that abrogated H3K4me3 binding also abolished leukaemic transformation. NUP98-PHD fusion prevented the differentiation-associated removal of H3K4me3 at many loci encoding lineage-specific transcription factors (Hox(s), Gata3, Meis1, Eya1 and Pbx1), and enforced their active gene transcription in murine haematopoietic stem/progenitor cells. Mechanistically, NUP98-PHD fusions act as 'chromatin boundary factors', dominating over polycomb-mediated gene silencing to 'lock' developmentally critical loci into an active chromatin state (H3K4me3 with induced histone acetylation), a state that defined leukaemia stem cells. Collectively, our studies represent, to our knowledge, the first report that deregulation of the PHD finger, an 'effector' of specific histone modification, perturbs the epigenetic dynamics on developmentally critical loci, catastrophizes cellular fate decision-making, and even causes oncogenesis during mammalian development.

Recent studies have showed that an H3K4me3-binding PHD finger in the NURF, ING2 or TFIID complex helps to recruit and/or stabilize these effectors and associated factors onto appropriate target promoters during transcriptional regulation1,6-10; An unmodified H3K4 (H3K4me0)-engaging PHD finger in the DNMT3L or LSD1-complex connects activities of DNA methylation or H3K4 demethylation to repressive chromatin11,12. Interestingly, germ-line mutation in the PHD finger of RAG2 abrogates its recognition of H3K4me3 and causes immunodeficiency13; Mutations in the PHD finger of ING1 have been implicated in cancers3,8,14. However, evidence supporting a causal role for PHD finger mutation and inappropriate interpretation of histone modification in oncogenesis is still elusive.

In clinically reported AML patients4,5, chromosomal translocation fuses the C-terminal PHD finger of JARID1A (also known as RBP2/KDM5A) or PHF23, together with nuclear localization signals, to NUP98, a common leukemia fusion partner that harbors transactivation activities15-17 (Supplementary Fig.1). Notably, the JARID1APHD3 motif is excluded from an alternatively spliced isoform of JARID1A and the corresponding NUP98-JARID1A fusion (hereafter referred to as NJS), while it is retained in the longer fusion isoform (hereafter referred to as NJL; Fig.1a). We asked whether JARID1APHD3 as a putative chromatin-‘reading’ module is involved in hematopoietic malignancies. To test this, we examined leukemogenic potential of both fusion isoforms using hematopoeitic progenitor transformation assay18 (Supplementary Fig.2a). While bone marrow-derived hematopoietic stem/progenitor cells transduced with empty retrovirus or retrovirus encoding NJS proliferated transiently and differentiated into mature cells, those transduced with NJL proliferated indefinitely as undifferentiated progenitors (Fig.1b-c). NJL-transduced marrow cells proliferated in a cell-autonomous manner, exhibited typical myeloblast morphology (Fig.1d) and expressed early myeloid progenitor antigens (c-Kit+/Cd11b+/Cd34+/Gr-1-/Cd19-/B220-/low; Fig.1e and Supplementary Fig. 2b). The arres of myeloid differentiation by NJL indicated that it would induce leukemia in vivo. Indeed, all of 12 mice transplanted with bone marrow progenitors transduced with NJL died of AML in an average of 69 days, whereas those reconstituted with vector- or NJS-transduced progenitors remained healthy after one year (Fig.1f). NJL-induced leukemia exhibited a myeloid phenotype (Supplementary Fig.2c-d), and typically presented with an enlarged spleen, packed progenitors in bone marrow, and massive increase in peripheral white blood cells (Supplementary Table 1; Fig.1g-h). Taken together, NJL represents a potent leukemia oncogene in both cellular and animal models.

Figure 1

The PHD finger-containing NUP98-JARID1A fusion isoform (NJL), but not that lacking the PHD finger (NJS), confers leukomogenic potentials to hematopoietic stem/progenitor cells

a, NUP98-JARID1A and NUP98-PHF23 structure (see Supplementary Fig.1 for details). b, Immunoblot of hematopoietic cells transduced with empty vector (lanes 1-2) or that encoding FLAG-tagged NJS (lanes 3-4) or NJL (lanes 5-6). c, Proliferation kinetics of lineage-negative hematopoietic cells after transduction of empty vector, NJL or NJS. Data are presented as mean ±s.d. of 6 experiments. d, Wright-Giemsa staining (insert, microscopy image) and e, FACS of NJL-transformed cells. f, Leukemia kinetics in mice (12 each group) after transplantation of bone marrow transduced with vector, NJL or NLS. g, Hematoxylin-Eosin staining of spleen section and h, Wright-Giemsa staining of bone marrow from NJL-induced AML mice. Scale bar, 20μM.

The fact that NJS failed to induce leukemia indicated that the PHD finger is required for leukemogenesis. Indeed, deletion of JARID1APHD3, but not JARID1A sequences prior to or following it, abolished NJL-mediated transformation of hematopoietic cells (Supplementary Fig.2f-h). We next asked whether JARID1APHD3 recognizes histone methylation. First, only histone H3 associated with recombinant JARID1APHD3 using total histone extracts (Supplementary Fig.3a). When a mini-library of H3 peptides harboring either unmodified, mono-, di- or tri-methylated K4, K9, K27, K36 or K79 were screened in biotinylated peptide pull-down, JARID1APHD3 only interacted with those containing H3K4me3/2 (Fig.2a; Supplementary Fig.3b). Such specificity was further confirmed by immunostaining and co-immunoprecipitation using Flag-NJL stable expression cells— NJL exhibited a speckled nuclear staining pattern and significantly co-localized with H3K4me3, but not H3K9me3 (Supplementary Fig.4); The vast majority of NJL were bound to mononuclesomes containing H3K4me3, but not H3K27me3 (Supplementary Fig.3c). Calorimetry-based measurements revealed a dissociation constant (Kd) of ∼0.75 μM for JARID1APHD3 binding to H3K4me3, with reduced affinities to H3K4me2/1/0 (Supplementary Fig.3d).

Figure 2

JARID1APHD3, an essential motif for NJL-mediated leukemia, specifically recognizes H3K4me3/2 marks

a, Capability of JARID1APHD3, PHF23PHD and JARID1APHD1 (the first PHD finger of JARID1A, Supplementary Fig.1) to interact with H3 peptides harboring different Kme in peptide pull-down assay. JARID1APHD1 interacted with H3K4me0 as BHC80PHD11. b, Crystal structure of JARID1APHD3 (cyan) complexed with H3K4me3 peptide (yellow) and a close-up view of the H3K4me3-binding channel (inset) formed by two orthogonally aligned Trp residues. The residue of JARID1APHD3 and H3 is shown in red and black, respectively. c, Capability of wildtype or mutant JARID1APHD3 to bind to H3K4me3/2. d, CoIP showing that NJL containing the wildtype, but not mutant (W1625A) PHD finger, associated with H3K4me3 or H3 in transiently trasfected 293 cells.

We determined the structure of JARID1APHD3:H3K4me3 complexes using X-ray crystallographic and NMR spectroscopic techniques. Both analyses revealed that the JARID1APHD3-H3K4me3 interaction was established via (i) anti-parallel β-sheet pairing between the H3 backbone and a β-sheet of JARID1APHD3, (ii) a hydrophobic cleft formed by two Trp residues (W1625, W1635) that anchor the H3K4me3 side chain, and (iii) positioning of H3R2 in an acidic pocket (Q1627/D1629/D1633) (Fig.2b; Supplementary Fig. 5b,6c). H3K4me3 is stacked between the indole rings of two orthogonally aligned Trp residues with intermolecular contacts showed in Fig.2b and Supplementary Fig.5b,6d. The X-ray (a domain-swapped dimer of one molecule and a crystallographic symmetry-related molecule) and solution NMR (monomer) analyses are summarized in Supplementary Fig.5 (statistics in Supplementary Table 2) and Supplementary Fig.6-7 (statistics in Supplementary Table 3), respectively. Comparison between JARID1APHD3 structures in the free and H3K4me3-bound state (Supplementary Fig.6a-b) revealed no overall conformational changes. Residues W1625 and W1635 are evolutionarily conserved among JARID1 homologues (Supplementary Fig.8a). Mutations targeting these Trp residues disrupted the H3K4me3-binding in vitro (Fig.2c) and in cells (Fig.2d). Such a two-sided H3K4me3-binding tryptophan channel is a varied form of the H3K4me3-engaging pocket involving 3-4 hydrophobic residues found in the PHD finger of BPTF7, ING28, Yng119 or RAG213 (Supplementary Fig.8b-d). Yet, it exhibited a stronger H3K4me3-binding affinity (Kd=0.75μM). Collectively, the PHD finger, an essential motif of NUP98-JARID1A, uniquely recognizes H3K4me3/2 using an aromatic engaging channel.

To gain insight into mechanisms of NJL-induced AML, we used microarray analyses to compare the transcriptome of NJL-transformed progenitors and control cells— committed myeloid progenitors generated as described before18. Strikingly, a significant portion of genes upregulated in NJL-transformed progenitors were those either targeted by polycomb proteins20,21 or exhibiting ‘bivalent domain pattern’22 in stem cells, many of which encode developmentally critical transcription factors (Hoxa5/a7/a9/a10, Gata3, Meis1, Eya1, Pbx1; Supplementary Table 4). Such upregulation was further confirmed by RT-PCR using vector- versus NJL-transduced marrow cells (Supplementary Fig. 9a-c). Other Hox-A genes (a1, a2, a11, a13) were not expressed in NJL-transformed progenitors. We detected a similar target specificity for Hox-A genes using chromatin immunoprecipitation (ChIP)— NJL directly bound to the promoters of Hoxa6-a10, but not distal Hoxa1-a3 or Hoxa11-a13 (Fig.3a and Supplementary Fig.9d; green); NJL-binding specificity among Hox clusters was correlated to H3K4me3— H3K4me3 was abundant in Hoxa6-a10, while low/absent in Hoxa1-a4 or Hoxa11-a13 (Fig.3b). Enforced expression of Hox and Meis1 has been shown sufficient to induce AML23. This indicated that NJL blocks hematopoietic differentiation and induces AML by enforcing the transcription of these genes.

Figure 3

NUP98-JARID1A enforced high H3K4me3 and active transcription associated with developmentally critical loci such as Hox

a, ChIP for NJL- or Ezh2-binding to A-cluster Hox promoters in committed myeloid progenitor line18 (cell 1) or in hematopoietic stem/progenitor cells three weeks after transduction of control vector (cell 2) or 3xFlag-tagged NJL (cell 3-4). b, ChIP of H3K4me3, H3K27me3 and general H3 among Hox-A gene cluster in hematopoietic progenitors three weeks after transduction of vector or NJL. c, Hoxa9/a10 expression in hematopoietic stem/progenitor cells after days of in vitro cultivation (day 0, 4, 8, 12), macrophages (mϕ), NIH-3T3 fibroblasts or NJL-transformed progenitors. d, α-Hoxa9 blot in marrow progenitors 10 days after transduction of MLL-ENL, empty vector, NJS or NJL. e, ChIP for Hoxa9/a10 promoter-associated H3K4me3 in hematopoietic stem/progenitor cells after days of in vitro culture, mϕ and marrow progenitors 20 days post transduction of vector or NJL. n=3, error bar indicates s.d; *, P<0.01; **, P<0.001; ***, P <10-4.

It has been reported that the A-cluster Hox gene expression is high in hematopoietic stem cells (HSC) and early progenitors, followed by down-regulation and shut-off during terminal differentiation24. Our ex vivo hematopoietic stem/progenitor cell system recapitulated such dynamics— coincident to the silencing of HSC marker and activation of differentiation marker (Supplementary Fig.9f), Hoxa9/a10 were down regulated >10- or 60-fold respectively in 8 days of culture (Fig.3c); Concurrent loss of Hoxa9/a10-associated H3K4me3 was observed in these cells (Fig.3e). Strikingly, NJL persistently enforced high levels of Hoxa9/a10 expression and Hoxa9/a10-associated H3K4me3 in marrow cells, whereas Hoxa9/a10 was silenced ten days after transduction of vector or NJS in similarly maintained cells (Fig.3c-e). To rigorously test the role of H3K4me3 recognition during leukemogenesis, we mutated the H3K4me3-engaging residues. NJL harboring mutation on the residueW1625 or W1635 failed to bind to H3K4me3 or H3 (Fig. 2d), failed to bind to the Hoxa9 promoter that exhibited high H3K4me3 in 293 cells (Fig.4a; Supplementary Fig.9i), failed to enforce the Hoxa9 expression (Fig.4b) or Hoxa9-associated H3K4me3 in hematopoietic progenitors (Fig.4c), and failed to transform the hematopoietic cells (Fig.4d), whereas the irrelevant mutation (V1609G) did not affect these activities (Supplementary Fig.10e). To assess whether NJL-induced phenotype was unique to JARID1APHD3, we investigated another similar de novo translocation, NUP98-PHF23 (Fig.1a)5, and also swapped JARID1APHD3 with other PHD fingers reported before. PHF23PHD specifically engaged H3K4me3/2 as predicted1 (Fig.2a); NUP98-PHF23 robustly enforced Hoxa9-associated H3K4me3 and transformed hematopoietic progenitors (Fig.4c,e; Supplementary Fig.10). Strikingly, swapping JARID1APHD3 with another H3K4me3/2-binding PHD finger from ING28 or even S. cerevisiae Yng119 also succeeded in the transformation, whereas replacing it with an H3K4me0-binding PHD finger, either BHC80PHD11 or JARID1APHD1 (Fig.2a), abolished the transformation (Fig.4c,e). Therefore, engaging H3K4me3/2 by NUP98-PHD fusion causes leukemia by enforcing an active state on developmentally critical loci.

Figure 4

The H3K4me3/2 engagement by NUP98-JARID1A perturbs the epigenetic state of developmentally critical loci during hematopoiesis

a, Impact of mutations on the Flag-NJL binding to HOXA9 in 293 cells. b, Immunoblot of hematopoietic progenitors ten days post transduction of vector, wildtype or mutant NJL. Phospho-c-Kit, marker of mast cells. c, ChIP for Hoxa9 promoter-associated NUP98-fusion proteins (3xFlag-tagged) and H3K4me3 in marrow progenitors 10 days after transduction. d, Transforming capacities after introducing mutation to NJL or e, those by NUP98-PHF23 or after replacing JARID1APHD3 with another PHD finger that engages either H3K4me3/2 or H3K4me0. Total progenitor number was counted at day 1, 10, 25 and 40. f, ChIP for SUZ12 and g, MLL2 binding to Hoxa9/a11 and h, Hoxa9-associated H3 acetylation in marrow progenitors 15 days after transduction of vector or NJL. Error bar indicates s.d (n=3); *, P<0.05; **, P<0.005; ***, P<10-4; *****, P<10-6. i, A scheme that NUP98-PHD fusion acts as “boundary factor” and prevents the spreading of polycomb factors from Hoxa13/a11 to Hoxa9, thus inhibiting H3K4me3 removal and H3K27me3 addition during hematopoiesis.

Because the H3K4me3 recognition cannot provide DNA sequence specificity and yet NJL-upregulated genes were enriched with polycomb-targeted 20,21 or ‘bivalent domain’ genes22 in stem cells (e.g., Hox(s), Gata3, Meis1; Supplementary Table 4), we asked whether such specificity is due to their dynamically regulated characteristics. Towards this end, we examined the effect of NJL on two distinct gene classes— developmentally critical genes, and housekeeping genes that exhibit constitutive H3K4me3 (Supplementary Fig.11a, top panel). Interestingly, although NJL bound to housekeeping genes, it had little affect on their expression during cell differentiation (Supplementary Fig.11a, middle and bottom panels). Thus, NJL tends to affect the developmentally critical loci specifically during hematopoeisis. We next pursued the possibility that NJL interferes with activities of polycomb proteins at these developmentally critical loci. Using ChIP, we found that, while Ezh2 or Suz12 was spread throughout Hox-A clusters in vehicle-infected marrow progenitors that underwent differentiation, these polycomb factors were restricted within Hoxa11-a13 in NJL-infected progenitors (Fig.3a,4f, red). In the NJL-transduced cells, H3K27me3 was also only detected at Hoxa13-a11— the differentiation-associated spreading of H3K27me3 was inhibited at a region from Hoxa10 to Hoxa1 (Fig.3b). The spreading of polycomb factors from distal Hox loci (a13-a11) seemed to be blocked at Hoxa10-a9 by NJL that were bound there (Fig.3a; Supplementary Fig.9d). Similar result was also found at Meis1 (Supplementary Fig.9e). Consistent to previous reports15,16, the recruitment of p300 and dramatic elevation of H3 acetylation (H3K27ac by >2,000 fold) were observed on Hoxa9 in NJL-transduced cells (Fig.4h; Supplementary Fig.11b). Collectively, NUP98-PHD fusion dominated over the spreading of polycomb and enforced an H3K4me3/acetylated histone state at developmentally critical loci, an epigenetic state that defines leukemia stem cells.

In summary, we have demonstrated for the first time that fusing an H3K4me3-engaging PHD finger (plus nuclear localization signal) to a common partner NUP98 is sufficient to induce leukemia. We showed that NUP98-PHD fusion prevented the silencing of critical loci encoding master transcription factors (Hox(s), Gata3, Mesi1, Pbx1) during hematopoietic differentiation. NUP98 fusion partners can be grouped into two major groups, DNA-binding homoedomain and chromatin-associated factors including PHD fingers (JARID1A, PHF23)17. Although the existence of additional unknown ligand is possible for PHD fingers in the latter group (as H3K4 site cannot be mutated in mammals), the most straightforward interpretation of our findings is that binding H3K4me3/2 marks is responsible for leukemia described here. In support, a genetic interaction was demonstrated in yeast between H3K4 and the Yng1 PHD finger25, a module that imparted similar oncogenic properties when swapping into our assays (Fig.4e). Several PHD fingers exist in NSD1, another NUP98-fusion partner16, however, none contains critical H3K4me3-engaging residues1. Thus, our report represents the first example wherein inappropriate interpretation of histone modification can actively induce a deregulation of developmentally critical loci, perturb cellular/epigenetic identities, and even induce oncogenesis. NUP98-PHD fusion coordinates acts of H3K4me3/2 and histone acetylation, mimicking mechanisms utilized by evolutionarily conserved ING(s)-complexes for robust gene activation19,26 (Supplementary Fig.12). H3K4me3 bound by NUP98-PHD may serve as ‘seed’ of propagation mediated by WDR5-MLL2/3 complexes1,27 that is also coupled with UTX/Jmjd3-mediated H3K27 demethylation28,29, as we detected high levels of WDR5, RBBP5, and MLL2 on Hoxa9 in NJL-transduced marrow cells (Fig.4g; Supplementary Fig.11c-d). We suggest that NUP98-PHD acts as ‘boundary factors’, using the PHD finger to protect H3K4me3 from JARID1(s)-mediated demethylation29 and also inducing H3K27ac to block H3K27me addition (Fig.4i). In support, we observed a ‘bivalent domain’ feature22 at Hoxa11-a10, the junction region of two antagonizing mechanisms (Fig.3b). Loss-of-function mutation of RAG2PHD in immunodeficiency and gain-of-function mutation involving PHD fingers in malignancies described here indicate a new type of diseases that arise from ‘misinterpreting’ the ‘histone code’3,30. With ∼200 PHD fingers in human genome and some intimately associated to diseases3, we expect similar ‘mis-reading’ mechanisms responsible for some unstudied diseases. These pathologies together with those caused by ‘mis-writing’ or ‘mis-erasing’29 histone modification, underscore the significance in investigating the biological readout of histone marks.

METHODS SUMMARY

Hematopoietic cell transformation assays

Protocols for the culture of primary hematopoietic stem/progenitor cells were previously described18. Briefly, 100,000 lineage-negative bone marrow stem/progenitor cells were subjected to retroviral infection, followed by kinetics analyses of proliferation versus differentiation in ex vivo culture system as described before18.

Peptide pull-down assay

Pull-down using biotinylated histone peptide and recombinant protein was performed as described6,11. After binding, peptide-Avidin beads were washed extensively in solution containing 50mM Tris pH 7.5, 150mM NaCl (250mM as stringent washing), 0.05% NP-40, 0.3mg/ml BSA and 1mM DTT.