Additional sex combs like 1 (ASXL1) is frequently mutated in myeloid malignancies and clonal hematopoiesis of indeterminate potential (CHIP). Although loss of ASXL1 promotes hematopoietic transformation, there is growing evidence that ASXL1 mutations might confer an alteration of function. In this study, we identify that physiological expression of a C-terminal truncated Asxl1 mutant in vivo using conditional knock-in (KI) results in myeloid skewing, age-dependent anemia, thrombocytosis, and morphological dysplasia. Although expression of mutant Asxl1 altered the functions of hematopoietic stem cells (HSCs), it maintained their survival in competitive transplantation assays and increased susceptibility to leukemic transformation by co-occurring RUNX1 mutation or viral insertional mutagenesis. KI mice displayed substantial reductions in H3K4me3 and H2AK119Ub without significant reductions in H3K27me3, distinct from the effects of Asxl1 loss. Chromatin immunoprecipitation followed by next-generation sequencing analysis demonstrated opposing effects of wild-type and mutant Asxl1 on H3K4me3. These findings reveal that ASXL1 mutations confer HSCs with an altered epigenome and increase susceptibility for leukemic transformation, presenting a novel model for CHIP.
Additional sex combs like 1 (ASXL1) is frequently mutated in patients with all forms of myeloid malignancies, including myelodysplastic syndromes (MDS), chronic myelomonocytic leukemia, myeloproliferative neoplasms, and acute myeloid leukemia (AML) with MDS-related changes. In each of these diseases, mutations in ASXL1 are consistently associated with poor prognosis and advanced age (Bejar et al., 2011; Thol et al., 2011; Haferlach et al., 2014).
Mutations in ASXL1 most commonly occur in the last exon as frameshift and nonsense mutations before the C-terminal plant homeofinger domain (Bejar et al., 2011; Thol et al., 2011; Haferlach et al., 2014). In addition, similar ASXL1 mutations are found in clonal hematopoiesis of indeterminate potential (CHIP; Genovese et al., 2014; Jaiswal et al., 2014; Xie et al., 2014), a condition associated with an increased likelihood of progression to myeloid malignancies. Previous work from us and others identified that Asxl1 loss in vivo results in progressive multilineage cytopenias and dysplasia with increased numbers of hematopoietic stem and progenitor cells (HSPCs) and impaired mature cell differentiation consistent with features of human MDS (Abdel-Wahab et al., 2013; Wang et al., 2014). At the same time, we previously reported that these same features are recapitulated by retroviral overexpression of mutant ASXL1 cDNA, suggesting that mutant forms of ASXL1 may inhibit the function of WT ASXL1 (Inoue et al., 2013). Moreover, recent work has suggested that expression of ASXL1 mutations may also confer a gain of function by promoting the enzymatic activity of BRCA1-associated protein 1 (BAP1), a deubiquitinase known to promote histone H2A lysine 119 (H2AK119) deubiquitination (Balasubramani et al., 2015).
The above findings raise the question of whether the recurrent ASXL1 mutations result in a loss of protein or expression of a truncated form lacking the C-terminal plant homeofinger domain. The fact that mutations in ASXL1 occur in a hotspot region located in the last exon and predicted to result in mutant transcripts that escape nonsense-mediated mRNA decay supports the possibility that most ASXL1 mutations may generate a stable truncated protein (Inoue et al., 2016). Consistent with this, truncated mutant forms of ASXL1 have been detected in leukemic cells with homozygous ASXL1 mutations using mTRAQ-based mass spectrometric analysis (Inoue et al., 2016). These data suggest that expression of mutant ASXL1 may be a more appropriate model to study the effects of ASXL1 mutations rather than models with deletion of ASXL1. However, our prior data regarding the effects of mutant ASXL1 expression to date have used retroviral overexpression, a method susceptible to experimental artifacts from supraphysiologic expression of cDNA, retroviral integration leading to up-regulation of neighboring genes, and forced cell cycle progression of quiescent hematopoietic stem cells (HSCs) when exposed to a high concentration of cytokines before transplantation (Cook and Pardee, 2013).
To overcome these limitations and evaluate the effects of mutant Asxl1 expression on disease development and progression, we generated conditional Asxl1 mutant knock-in (Asxl1-MT KI) mice. Hematopoietic expression of mutant Asxl1 in this manner perturbed hematopoiesis but was not sufficient to transform HSCs. At the same time, mutant ASXL1 expression increased susceptibility of HSCs to transformation from additional mutations provided by insertional mutagenesis or expression of mutant RUNX1 (Runt-related transcription factor 1) or the MLL-AF9 fusion oncogene. In addition, we identified differences in the direct binding targets of WT Asxl1 and mutant and their distinct effects on histone modifications. These data provide a faithful model to study ASXL1 mutations in vivo and explain the enrichment of ASXL1 mutations in CHIP and early forms of MDS.
Generation of conditional Asxl1-MT KI mice
To generate Asxl1-MT KI mice, we mimicked the well-described human ASXL1 mutation, p.E635RfsX15 (Thol et al., 2011; Chen et al., 2014; Paschka et al., 2015). To allow for conditional ASXL1 mutant expression with epitope and fluorescent tagging, a floxed neomycin and stop cassette followed by an N-terminal 3xFLAG-tagged mouse Asxl1 mutant (1–1,890) cDNA followed by internal ribosome entry site (IRES)/enhanced GFP (EGFP) were inserted into the Gt(ROSA)26Sor (Rosa26) locus by homologous recombination (Fig. 1 A). Correctly targeted embryonic stem cells (ESCs) identified using 3′ genomic PCR (Fig. 1 B) and Southern blotting with 5′ genomic probes (Fig. 1 C) were used to create chimeric mice that transmitted the mutated allele through the germline. Mice carrying the floxed allele (3xFLAG-Asxl1 mutant flox/flox-IRES-EGFP) were crossed with Vav-Cre transgenic mice, in which Cre recombinase is activated exclusively in the hematopoietic lineage (de Boer et al., 2003). Vav-Cre Asxl1 mutant heterozygous (fl/wt) and homozygous (fl/fl) mice were born at expected Mendelian ratios (unpublished data) with no apparent developmental abnormality. PCR and Western blot analysis demonstrated Cre recombination and the expression of the mutant proteins exclusively in hematopoietic tissues (Fig. 1, D and E) with Asxl1-MT KI cells clearly identified by GFP positivity (Fig. 1 F). Finally, Sanger sequencing and quantitative RT-PCR of cDNA from hematopoietic tissues confirmed expression of mutant Asxl1 expression at levels comparable to the WT alleles, and we demonstrated no effect of Asxl1 mutant on WT expression by quantitative RT-PCR and Western blotting (Fig. 1, G–I).
Characterization of Asxl1-MT KI mice
Examination of peripheral blood (PB) parameters revealed no significant change in younger mice (24 wk old; not depicted) and a reduction in red blood cell (RBC) counts and increase in mean cellular volume (MCV) and platelets in KI mouse at 70 wk of age (Fig. 2 A). We also observed increased frequency of CD11b+ myeloid cells, in particular, neutrophils (CD11b+ Gr1+), but not monocytes (CD11b+ Gr1−), in the PB and bone marrow (BM) and an increased proportion of neutrophils and CD11b+ c-Kit+ immature myeloid cells in the BM (although the total white blood cell [WBC] count did not significantly increase; Fig. 2, A–C; and Fig. S1, A and B). To further characterize anemia in the KI mice, we examined committed erythroid precursor maturation in the BM based on CD71 and Ter119 expression, which subdivides cells into several stages of erythroblast maturation (Chen et al., 2009; Liu et al., 2013). This revealed a modest block in erythroid development, characterized by an accumulation of cells in stage I and II of erythroblast maturation in the BM of KI mice, similar to previous findings in Asxl1 KO mice (Shi et al., 2016; Fig. 2 D). Consistent with the thrombocytosis seen in KI mice, a connection between ASXL1 mutations and higher platelet counts was seen in a cohort of MDS–refractory anemia patients. In this cohort, platelet counts were significantly higher in patients with ASXL1 mutations than in those WT for ASXL1 (median, 135,000/μl; range, 34,000–607,000 vs. 68,500/μl; range, 3,000–442,000, P = 0.0068). Although no hepatosplenomegaly or hypercellularity in the BM was detected at any age (Fig. 2 E), increased red pulp along with decreased follicles in the spleen were identified, suggesting more activated extramedullary hematopoiesis in the KI mice (Fig. S1 C). Moreover, histological analysis revealed increased megakaryocytes in the BM as well as spleen in the old KI mice (70 wk old; Figs. 2 F and S1 D). Finally, dysplastic features in the BM, including nucleated erythroid precursors, abnormal nuclear maturation, hypersegmentation, and megaloblastic change of erythroid lineage in relatively minor population, were detected (Figs. 2 G and S1 E).
Impaired HSC function in Asxl1-MT KI mice
We next determined whether expression of Asxl1 mutant altered HSPC frequency by multiparameter flow cytometry. Intriguingly, there was a profound decrease in the frequency of BM Lin−Sca-1+c-Kit+ (LSK) cells, CD150+CD48− long-term repopulating HSCs (LT-HSCs), and CD150−CD48+ multipotent progenitors (MPPs; Fig. 3, A and B). Of note, we observed the increase in the number of megakaryocyte/erythroid progenitors (MEPs; Lin−c-Kit+Sca1−CD34−FcγR−) in KI group, whereas other myeloid progenitor populations were not affected (Fig. 3, A and C). In addition, increased MEPs of KI mice were found to differentiate toward megakaryocyte progenitors (MkPs; CD41highCD71lowMEP) rather than erythrocyte progenitors (ErPs; CD41lowCD71highMEP; Fig. 3 D; Murphy et al., 2013), which is consistent with the increased megakaryocytes and platelets shown in Fig. 2 (A and F) and Fig. S1 D. In line with myeloid skewing in the PB and with fewer HSCs in the BM, total colonies, erythroid CFU (CFU-Es), and erythroid burst-forming units (BFU-Es) were profoundly reduced in KI cells, whereas the number of granulocyte-macrophage CFU (CFU-GM) colonies was increased when we started with whole BM cells (Fig. 3, E and G). To exclude the possibility that the colony output might be affected by the reduction of HSPCs shown in Fig. 3 B, we repeated the assay using the same number of LT-HSC and MPP and confirmed the reduction of BFU-Es and the increase of CFU-GMs (Fig. 3 F). In contrast to the result of whole BM cells, the total number of colonies from LT-HSC and MPP was enhanced because of dramatic changes in CFU-GMs (Fig. 3 G). Of note, ectopic expression of ASXL1-WT in Lin− cells completely or partially rescued the reduction of BFU-Es/CFU-Es found in KI cells (Fig. 3, H and I). However, CFU-GMs decreased after WT expression (Fig. 3, H and I), suggesting the opposite effect of WT and mutant protein on erythroid and myeloid colony formation.
We next performed a competitive repopulation assay to determine whether expression of Asxl1 mutant affects HSC self-renewal and reconstitution in vivo (Fig. 4 A). This revealed that Asxl1-MT KI mice had a significant competitive disadvantage in PB and whole BM cells (Fig. 4, B and C). Although engraftment of B220+ cells was severely impaired, the proportion of CD11b+ cells from KI donor cells was substantial (Fig. S2 A). Chimerism in whole BM cells, Lin− BM cells, MPs, common myeloid progenitors (CMPs), granulocyte/macrophage progenitors (GMPs), and short-term repopulating HSC fractions was severely reduced, as observed in PB cells (Fig. 4 C). In contrast, however, the donor-derived chimerism, frequency, and absolute numbers of LSK, LT-HSC, and MPP populations were not impaired in KI mice relative to controls (Fig. 4, C and D; and Fig. S2 B). In another competitive (Fig. S2 C) and a noncompetitive repopulation assay, Asxl1 KI cells had impaired and delayed repopulation, respectively (Figs. 4 E and S2 D). The recipient mice with KI cells showed an increased proportion of myeloid cells relative to control (Figs. 4 F and S2 E). Overall, these data reveal that Asxl1 mutant expression compromised the quality and quantity of stem cell fraction without affecting the survival of HSCs in vivo.
Mutant RUNX1 promoted MDS/AML development in the Asxl1 mutant background
No Vav-cre Asxl1-MT KI mice developed AML after long-term follow up (18 mo), suggesting that Asxl1 mutant clones might require additional mutation for full transformation. Recent work has identified that mutations in ASXL1 and RUNX1 tend to coexist frequently in MDS as well as AML (Papaemmanuil et al., 2013; Chen et al., 2014; Haferlach et al., 2014). Similarly, across a cohort of 368 patients with de novo World Health Organization 2008–defined MDS (Vardiman et al., 2009) where ASXL1 was mutated in 17.4% (64/368), patients with ASXL1 mutation had a significantly higher incidence of concurrent RUNX1 mutations than their ASXL1 WT counterparts (19/64 [29.7%] vs. 20/304 [6.6%]; P < 0.01; Fig. 5 A). None of these RUNX1 mutations were found to be germline line mutations in this cohort. ASXL1/RUNX1 comutation in MDS was associated with poor overall and progression-free survival (Fig. 5 B). Prior work has demonstrated that RUNX1 mutations can be classified into missense mutations in the N-terminal Runt domain and C-terminal truncated type mutations resulting from missense or frameshift mutations (Harada and Harada, 2009). As shown in Fig. 5 A, regardless of ASXL1 mutational status, RUNX1 mutations were more commonly found to be the C-terminal truncation type, such as the RUNX1-S291fsX mutation, rather than missense mutations in the Runt domain. Based on these findings, we next examined the in vivo effect of expression of RUNX1-S291fsX in the Asxl1-MT KI or Asxl1 WT background. Intriguingly, mice transplanted with BM cells expressing both Asxl1 and RUNX1 mutations developed macrocytic anemia over time (Figs. 5 C and S2 F) and died of MDS/AML with remarkable hepatosplenomegaly after a short latency (median survival, 160 d), whereas all of the mice transplanted with BM cells expressing only RUNX1-S291fsX survived for 1 yr (Fig. 5, D and E). Frequent myeloid cells with dysplastic feature were evident in the BM, spleen, and PB of the double-mutant animals with infiltration of these cells into the liver and spleen (Fig. 5, F and G). In addition, leukemic cells from mice with both RUNX1-S291fsX and Asxl1-MT KI were serially transplantable into sublethally irradiated recipients (not depicted). Coexpression of RUNX1-S291fsX increased the LSK fraction (Fig. 5 H). An increase in CMPs (Lin−c-Kit+Sca1−CD34+FcγR−) with decreased MEPs in Asxl1/RUNX1 double mutants was compatible with the AML phenotype observed in this model (Fig. 5 H). Interestingly, we observed the more significant effects of RUNX1 mutant expression on LT-HSC and MPP2 (CD150+CD48+LSK) rather than on LSK (Fig. 5 I; Pietras et al., 2015). Moreover, apoptosis was significantly reduced by RUNX1 mutant expression in KI mice, which presumably contributed to the transformation, although cell cycle arrest was found in LSK and progenitors of the mice with both mutations, as previously described in Runx1 mutant–expressing cord blood cells (Goyama et al., 2013; Fig. S2, G and H). CFU-GM and BFU-E colony output of Asxl1-MT KI cells was enhanced by RUNX1-S291fsX, whereas the CFU-E number was significantly decreased (Fig. 5 J). These data reveal that expression of mutant RUNX1 in Asxl1-MT KI mice increased the stem cell fractions of the KI mice and induced myeloid leukemia at the expense of erythroid lineage maturation.
KI mice are susceptible to leukemia from insertional mutagenesis
To further assess the susceptibility of KI mice to leukemia, we applied retrovirus-mediated insertional mutagenesis by using MOL4070LTR retrovirus (MOL4070A; Wolff et al., 2003; Ikeda et al., 2016; Ueda et al., 2016), which integrates into the mouse genome and up-regulates expression of the neighboring genes. During a 1.5-yr observation period, all of MOL4070A-infected KI mice developed acute leukemia (Fig. 5 K). Macroscopically, most of the diseased Asxl1-MT KI + MOL4070A mice exhibited massive hepatosplenomegaly and immature myeloid blasts in terms of morphology of BM and spleen cells (Fig. 5, L and M). Most of the GFP+ BM cells were positive for CD11b and c-Kit, suggesting that leukemic cells derived from Asxl1 mutant-expressing immature myeloid cells (Fig. S2 I). We detected the genomic location of integration by inverse PCR of six recipients’ BM cells (Table 1). As retroviral common integration sites, we identified Hhex, Rnf220, and H2-Eb2 across recipients. We also observed integration near Setbp1 gene in a single recipient. This matched our previous result showing ASXL1-mutated MDS collaborates with SETBP1 mutation, which impairs ubiquitin-proteasome dependent degradation of SETBP1 protein, during leukemic transformation (Inoue et al., 2015). Moreover, MLL-AF9 fusion oncogene-induced AML was enhanced in the Asxl1-MT KI background over Asxl1 WT background (57 d vs. 51 d, median survival; P = 0.0179, log-rank test; Fig. 5 N). Overall, these in vivo results indicate that KI cells require further oncogenic alterations for leukemic transformation.
Transcriptional and epigenetic alterations in mutant Asxl1 progenitors
To understand the basis for the impaired erythroid differentiation and stem cell dysfunction observed in Asxl1-MT KI mice, we performed RNA-seq of sorted c-Kit+ BM cells from 2.5-mo-old KI mice and age-matched littermate controls. In line with previous studies, Hoxa9 was up-regulated in KI cells and gene set enrichment analysis (GSEA) demonstrated gene sets containing Hoxa9, and Myc target genes were enriched in c-Kit+ BM cells from KI mice compared with controls (Schuhmacher et al., 2001; Hess et al., 2006), whereas GATA-binding protein 1 (GATA1) or RUNX1 targets (Suzuki et al., 2009, 2013), both of which are known to contribute to erythroid differentiation (Yokomizo et al., 2008; Suzuki et al., 2013), were repressed in KI mice (Fig. 6 A). Consistent with the data shown in Fig. 3 C, c-Kit+ KI cells showed significant loss of gene sets enriched in HSC against MEPs (Ng et al., 2009), whereas genes involved in erythroid differentiation and heme metabolism were inhibited (Figs. 6 A and S3 A; Addya et al., 2004; Riz et al., 2007). Of note, the gene sets enriched in HSCs relative to MEPs and the targets of RUNX1 and GATA1 were also significantly down-regulated in LSK fractions of KI mice compared with those of control mice, which support our results regardless of cell population analyzed (Suzuki et al., 2009, 2013; Fig. S3 B). We also compared RNA-sequencing data from KI mice with those of LSK cells from Asxl1 KO mice (Abdel-Wahab et al., 2013). There were 939 and 266 genes commonly up-regulated and down-regulated across both models, respectively (Fig. 6 B). Gene ontology (GO) biological process enrichment analysis identified the common ontologies related to RNA splicing, ribosome, erythroid/myeloid differentiation, immune activation, and kinase activation in gene sets commonly dysregulated or differentially expressed only in KI cells (Figs. 6 C and S3 C).
Previous studies have associated Asxl1 loss or mutant overexpression with loss of histone H3 lysine 27 trimethylation (H3K27me3) as well as decreased H2A lysine 119 monoubiquitination (H2AK119Ub) caused by alterations in PRC2 and BAP1 function, respectively, in ASXL1 mutant cells (Abdel-Wahab et al., 2012; Inoue et al., 2013; Wang et al., 2014; Balasubramani et al., 2015). However, these potential effects of ASXL1 alterations on chromatin state have not been investigated with physiological expression of mutant Asxl1 previously. We therefore performed anti-H3K4me3 (H3 lysine 4 trimethylation), H3K27me3, and H2AK119Ub chromatin immunoprecipitation (ChIP) followed by next-generation sequencing (ChIP-seq) in c-Kit+ BM cells from KI mice and littermate controls. Consistent with the result of Western blot analysis (Fig. 6 D), H3K4me3 and H2AK119Ub were globally decreased in Asxl1-MT KI cells relative to controls, whereas H3K27me3 was not significantly affected (Figs. 6 E and S3 D). However, evaluation of site-specific abundance of histone modifications at the Hoxa loci revealed reductions in H3K27me3 (Fig. 6 F), which were focal and not global, and there was no correlation between EZH2 target genes and genes dysregulated in Asxl1-MT KI mice (Fig. 6 A; Kamminga et al., 2006). Given that reductions in H3K4me3, a mark associated with active gene expression, were the most apparent histone modification change in Asxl1-MT KI cells, we evaluated genes that had particularly strong reductions in H3K4me3 in the KI cells. Such sites were identified at Sox6, inhibitor of DNA binding 3 (Id3), Tjp1, and Hba loci, whose mRNA expression levels were validated in quantitative RT-PCR (Fig. 6, F and G; and Fig. S3 E) and all of which encode proteins involved in erythroid differentiation and/or maturation (Deed et al., 1998; Cantù et al., 2011; Zhao et al., 2016). Changes in H3K4me3 abundance were positively correlated with gene expression changes and negatively correlated with H2AK119Ub changes, whereas changes in H2AK119Ub abundance were negatively correlated with gene expression changes (Fig. S3 F). We therefore next investigated whether H3K4me3 loss occurred in a specific manner. The reduction of H3K4me3 can be observed independently of transcribed levels as shown in Fig. S3 G, where we classified protein-coding genes into four categories based on expression levels. Next, we focused on loci marked by both H3K4me3 and H2AK119Ub in control cells. Within the overlapped peaks, we classified these into four groups; (1) both modifications remained in KI (4,468 peaks), (2) only H3K4me3 remained in KI (1,236 peaks), (3) only H2AK119Ub remained in KI (185 peaks), and (4) both modifications absent in KI (80 peaks; Fig. S4 A). Of note, even in group 2, where H2AK119Ub disappeared in KI cells, H3K4me3 signals were also reduced in KI cells. Similarly, in group 3, where H3K4me3 disappeared in KI cells, H2AK119Ub signals decreased but H3K27me3 signals increased in KI cells (Fig. S4 B). These data suggest profound effects of Asxl1 mutant on the reduction of both H3K4me3 and H2AK119Ub even in genes with both modifications. We used the above genomic data to identify mutant Asxl1 target genes and evaluate whether any of these possible targets rescue the phenotype of Asxl1-MT KI BM cells. ID3 is a previously described target of WT ASXL1 and RUNX1, and its expression or loss promotes or inhibits erythropoiesis, respectively (Deed et al., 1998; Abdel-Wahab et al., 2013; Zhao et al., 2016). Id3 expression was down-regulated in Asxl1-MT KI cells associated with reduced H3K4me3 at the Id3 locus (Fig. 6 F). Remarkably, reexpression of Id3 in Asxl1-MT KI BM cells rescued CFU-E and BFU-E numbers (Fig. 6 H). These results suggest that H3K4me3 reduction around Id3 resulted in transcriptional repression and, at least in part, is responsible for the maturation block in erythroid lineage.
Effects of genome-wide binding of WT versus mutant ASXL1 on histone modifications
Next, to examine the potential effects of WT and mutant Asxl1 binding on H3K4me3, we performed ChIP-seq with N-terminal Asxl1 and FLAG antibody using c-Kit+ control (WT) and KI BM cells. Nearly all mutant (FLAG) binding sites were shared with Asxl1-WT binding sites (2,365 of 2,520 genes, 93.8%), compatible with a study showing that the predicted DNA-binding region of ASXL1 is located in its N terminus (Sanchez-Pulido et al., 2012; Fig. 7 A). Although mutant binding sites were more enriched at promoter regions than WT binding sites (Fig. 7 B), binding motif enrichment analysis of WT and FLAG (mutant) ChIP-seq revealed similar results, including a variety of E–twenty six (ETS) transcriptional factors (Table S1), In fact, Id3 gene was identified as the target of both Asxl1-WT and mutant (Fig. 7 C), suggesting Asxl1 is directly involved in Id3 expression. Next, we investigated how WT Asxl1 affects global H3K4me3, H3K27me3, and H2AK119Ub levels. Intriguingly, the intensity of Asxl1 binding in control cells was well correlated with H3K4me3 (R2 = 0.75) and H2AK119Ub (R2 = 0.4) abundance, but not that of H3K27me3 (R2 = 1.1e-05), indicating that Asxl1 itself supports H3K4me3 and H2AK119Ub modification (Fig. 7 D). Asxl1 WT binding signals tend to correlate with the ratio of H3K4me3 to H3K27me3 (Fig. 7 E), although H3K27me3 was decreased at specific loci, such as the posterior Hox A locus, without global reduction (Fig. 6 F). Combined with ChIP-seq analyses for H3K4me3, we demonstrated that the intensity of Asxl1 mutant (FLAG) binding tends to be correlated with the reduction of H3K4me3 in KI mice (Fig. 7 F). Given that WT and mutant share the same targets, these data strongly suggest that WT supports H3K4me3 modification, but mutant does not. Moreover, motif enrichment analysis of differential H3K4me3 peaks in KI versus WT revealed enrichment of motifs characteristic of ETS family of transcription factors (TFs; Fig. 7 G), similar to the binding motifs of both WT and mutant (Table S1; Abdel-Wahab et al., 2013). In addition, the motif of IRF8, whose targets were significantly down-regulated in KI cells (Fig. 7 G), was detected, consistent with the observation that silencing of IRF8 promotes development of “preleukemic” clones and that MDS patients with a higher risk of progression to AML had lower IRF8 levels (Will et al., 2015). Finally, we compared H3K27me3 peaks in c-Kit+ BM cells from KI and KO mice, which were previously reported (Fig. S4, C and D; Abdel-Wahab et al., 2013). As expected, we demonstrated that the number of genes whose promoters are modified with H3K27me3 was much larger in KI cells than in KO cells, indicating the different roles of KI and KO in H3K27me3 (Fig. S4, C and D).
Here, we show that expression of mutant Asxl1 in hematopoietic tissues results in myeloid skewing and age-dependent mild anemia with erythroid differentiation block, modest dysplasia, and thrombocytosis. These results are consistent with the enrichment of ASXL1 mutations in myeloid neoplasms and the worsened anemia encountered in MDS and myeloproliferative neoplasm patients with ASXL1 mutations (Quesada et al., 2011; Zhang et al., 2012; Shi et al., 2016). Prior models of constitutive as well as conditional Asxl1 deletion in vivo revealed that Asxl1 loss results in impaired HSC self-renewal (Abdel-Wahab et al., 2013; Wang et al., 2014). However, the impaired HSC self-renewal with Asxl1 loss appears to conflict with the fact that ASXL1 mutations are enriched in settings of clonal advantage seen in CHIP and myeloid neoplasms (Genovese et al., 2014; Jaiswal et al., 2014; Xie et al., 2014). This discrepancy, in part, has led to the hypothesis that ASXL1 mutations may alter ASXL1 function rather simply conferring a loss of function. Here, we have directly addressed this hypothesis by generating and characterizing mice engineered to express one of the common mutations in ASXL1 in a conditional, hematopoietic-specific manner, although KI at Rosa26 instead of the endogenous Asxl1 locus cannot retain all the regulatory elements that control mutant expression. In this model, expression of mutant Asxl1 results in an impaired quantity and quality of HSPCs, similar to what has been encountered in Asxl1 KO mice (Abdel-Wahab et al., 2013; Wang et al., 2014). However, the chimerism of LSK and LT-HSC did not decrease in the competitive transplantation assay, whereas that of more mature progenitors as well as whole BM cells exhibited competitive disadvantage. These data indicate that ASXL1 mutations maintain survival at the HSC level and promote transformation into MDS/AML after acquiring additional mutations, such as in RUNX1. One possible hypothesis for this is that the epigenetic alterations associated with mutations in ASXL1 might have context-specific effects depending on the cell type, resulting in distinct phenotypic effects in stem cells versus more differentiated progeny. In fact, we identified that expression of mutant Asxl1 increases susceptibility of HSPCs to transformation by several oncogenic stimuli. For example, expression of mutant RUNX1 rescued the disadvantage of HSPCs derived from KI mice (Fig. 5) and increased LT-HSC/MPP and reduced apoptosis in KI cells (Figs. 5 I and S2 H), suggesting that the double-mutant model can be considered a slowly progressive AML model with expanded LT-HSC/LSK/MPP populations (Fig. 5, H and I). These data suggest that ASXL1-mutated cells may be founder clones for human MDS/AML and that ASXL1 mutations render HSPCs susceptible to transformation with the acquisition of additional mutations. The RUNX1/Asxl1 double-mutant model generated here also provides a model for the one of the most adverse genotypes of AML, as recently described in the newest clinical guidelines for AML management (Döhner et al., 2017). Furthermore, we identified some common integration sites shown in Table 1. Hhex (hematopoietically expressed homeobox) is a direct target of Runx1 and supports PRC2-mediated repression of Cdkn2a in AML cells (Shields et al., 2016; Behrens et al., 2017).
Asx, the Drosophila melanogaster homologue of mammalian ASXL1-3, is required to both maintain repression and activate expression of Hox genes, and ASXL genes are thought to mediate the balance between polycomb and trithorax functions (Milne et al., 1999). More recent work has identified that Asx and ASXL1/2 serve as cofactors for the H2AK119Ub deubiquitinase BAP1 (Sowa et al., 2009; Scheuermann et al., 2010; Dey et al., 2012), and mutant ASXL1 rather than WT enhances deubiquitinase activity. Simultaneous overexpression of BAP1 and mutant ASXL1 promotes the deubiquitinase activity of BAP1, leading to the loss of H2AK119Ub followed by H3K27me3 reduction (Balasubramani et al., 2015). At the same time, ASXL1 loss or mutation in hematopoietic cells is associated with down-regulation of H3K27me3 (Abdel-Wahab et al., 2012; Inoue et al., 2013; Wang et al., 2014). However, the consequences of physiological Asxl1 mutant expression have not been fully explored. Here, we first demonstrated that ASXL1 WT binding was associated with the abundance of H3K4me3 and H2AK119Ub and that physiological expression of mutant in vivo was involved in global reductions in these modifications (Fig. 6, D–F; and Fig. 7, D–F). Balasubramani et al. (2015) used retroviral overexpression of both ASXL1 mutant and BAP1 to show dual reduction in H2AK119Ub and H3K27me3. In our model, BAP1 mRNA did not change in KI mice, and physiological expression of Asxl1 mutant seemed to mildly support Bap1 function, which is not sufficient for global H3K27me3 reduction. In fact, overexpression of either ASXL1 mutant or BAP1 has not been found to profoundly reduce global H3K27me3 levels.
Regarding the molecular mechanism by which ASXL1 WT supports H3K4me3 levels, we recently identified OGT (O-linked N-acetylglucosamine [GlcNAc] transferase) and HCFC1 (host cell factor C1) as interacting partners of both ASXL1 WT and mutant proteins by mass spectrometry and immunoprecipitation (Inoue et al., 2018). Previous studies identified that OGT/HCFC1 recruit trithorax homologues, including MLL, SET1/COMPASS, and MLL5 (Yokoyama et al., 2004; Tyagi et al., 2007; Deplus et al., 2013; Zhou et al., 2013). We also clarified that knockdown of ASXL1, OGT, HCFC1, or MLL5 reduces global H3K4me3 and impairs hematopoietic differentiation with robust overlap between H3K4me3–down-regulated loci with knockdown of each factor (Inoue et al., 2018). These data suggest that ASXL1 can support the methyltransferase activity of trithorax homologues through OGT/HCFC1. Considering that ASXL1 WT and mutant share the same putative DNA binding domain whereas mutant binding signals negatively correlate with H3K4me3 changes in KI mice with differential H3K4me3 peaks containing binding sites of ETS family TFs (Sanchez-Pulido et al., 2012; Abdel-Wahab et al., 2013; Fig. 7, B–G), it is tempting to speculate that Asxl1-truncated mutants may compete with WT Asxl1 to access DNA around the transcription start site (TSS) and/or to recruit H3K4 methyltransferases and ETS family TFs.
Down-regulation of gene expression and loss of H3K4me3 were apparent at genes involved in erythroid differentiation as well as targets of GATA1 and RUNX1, both of which are known regulators of erythroid maturation (Yokomizo et al., 2008; Suzuki et al., 2009, 2013). For example, reduced expression and lower levels of H3K4me3 were evident at Id3 and Sox6, which play important roles in erythroid differentiation (Cantù et al., 2011; Zhao et al., 2016) and are known Asxl1 and Runx1 targets (Abdel-Wahab et al., 2013). Reexpression of Id3 cDNA into Asxl1-MT KI BM cells could rescue the colony-forming ability of Asxl1 mutant cells. These data suggest that Id3 may be a key target in mediating impaired erythropoiesis in ASXL1 mutant cells. Given that ID3 is also a target of the TGF-β pathway, evaluation of drugs targeting this pathway, which have shown promise in improving anemia in ongoing clinical trials of MDS patients based on the ASXL1 genotype of MDS patients, may be of therapeutic importance (Paulson, 2014; Blank and Karlsson, 2015).
Collectively, our experiments revealed that stable, physiological expression of mutant Asxl1 results in dysfunction of HSPCs and perturbed erythroid-lineage differentiation. Given the paucity of mouse models for clonal hematopoiesis, we believe our model will contribute to future studies that aim to elucidate the molecular basis for MDS/AML development from precancer clones. These results will be critically important in identifying potential novel therapies targeting cells bearing expression of mutant Asxl1.
Materials and methods
Adult patients (n = 368) with de novo MDS diagnosed according to the 2008 World Health Organization classification (Vardiman et al., 2009) at the National Taiwan University Hospital (NTUH) who had cryopreserved BM cells for study were recruited for gene mutation analyses. All patients signed informed consents for sample collection in accordance with the Declaration of Helsinki. The clinical portion of this study was approved by the institutional review board of NTUH (approval no. 201507084RINA). The coding region of ASXL1 and RUNX1 was amplified and sequenced as described previously (Chen et al., 2014).
A mutant mouse Asxl1 cDNA was subcloned between a floxed neomycin resistance-stop/pA cassette and an Frt-flanked IRES-EGFP/pA cassette of STOP-eGFP-ROSA26TV, which was obtained from Y. Sasaki (Kyoto University, Kyoto, Japan) via H. Koseki (RIKEN, Yokohama, Kanagawa, Japan). Electroporation and screening of ESCs were performed as previously described (Ueda et al., 2016). To verify homologous recombination in the mouse Rosa26 locus, individual clones were screened by 3′ genomic PCR using P1 (5′-CTCTATGGCTTCTGAGGCGGAAAGAACCAG-3′) and P2 (5′-CTTTAAGAGCCATGGCAATGTTCAAGCAGG-3′) primers, followed by a 5′ Southern blot using EcoRI-digested DNA and a 5′ external probe (Fig. 1 A). Correctly targeted ESCs were microinjected into blastocysts derived from C57BL/6 × BDF1 mice, and chimeric male mice were crossed with C57BL/6 female mice to transmit the mutant allele to progeny. After four or five generations of backcross with C57BL/6, mice harboring the KI allele were crossed with Vav-Cre transgenic mice (de Boer et al., 2003). Recombination was confirmed by PCR with 5′-CATCTGTAGGGCGCAGTAGTC-3′ (Fig. 1 A, P3) and 5′-CCGTCGTGGTCCTTGTAGTC-3′ primers (Fig. 1 A, P4). We performed quantitative RT-PCR in Fig. 1 H using the following primers: N-terminal Asxl1, 5′-GGACTGGGGTTATGCTACCTC-3′ and 5′-GCCTCGAATGGCACTGTTG-3′; C-terminal Asxl1, 5′-CAGACTCTCTGTTGCTGTCC-3′ and 5′-CTGTGACCTGGAGGAAGAACC-3′; and Actb (internal control), 5′-TGTTACCAACTGGGACGACA-3′ and 5′-GGGGTGTTGAAGGTCTCAAA-3′. The experiments were approved by the Committee on the Ethics of Animal Experiments, University of Tokyo, and all mice were maintained according to the guidelines of the Institute of Laboratory Animal Science (PA13-19, PA16-31, and PA17-26).
Primary BM cells, sorted LT-HSCs and MPPs, and Lin− cells were plated in MethoCult (M3234; StemCell Technologies). Methocult for BFU-Es (day 7), CFU-Es (day 3), CFU-GMs (day 7), and the M3434 condition (CFU-Gs, CFU-Ms, CFU-GMs, BFU-Es, CFU-GEMMs, day 10) were scored, and morphological changes of colony-forming cells were examined according to the manufacturer’s protocol. Cytokine conditions were as follows: BFU-E, 50 ng/ml recombinant mouse stem cell factor (rmSCF), 10 ng/ml rmIL-3, and 4 U/ml recombinant human erythropoietin (rhEPO); CFU-E, 3 U/ml rhEPO; CFU-GM, 50 ng/ml rmSCF, 10 ng/ml rmIL-3, 10 ng/ml rmIL-6, 10 ng/ml rmGM-CSF; M3434 condition, 50 ng/ml rmSCF, 10 ng/ml rmIL-3, 10 ng/ml rhIL-6, and 3 U/ml rhEPO. All cytokines were purchased from R&D Systems.
Flow cytometry and antibodies
mAbs recognizing the following antigens were used in flow cytometry and cell sorting: CD45.2 (109813 and 109824; BioLegend), CD45.1 (110716; BioLegend), Gr-1 (108408; BioLegend), CD11b/Mac-1 (101216; BioLegend; and 12–0112-85; eBioscience), Ter-119 (116222; BioLegend; and 12–5921-83; eBioscience), CD71 (113815; BioLegend; and 553267; BD PharMingen), CD41 (133914; BioLegend), B220 (103222; BioLegend; and 12–0452-85; eBioscience), CD3 (100236; BioLegend; and 12–0031-85; eBioscience), CD117/c-Kit (105808 and 105814; BioLegend), Sca-1 (108112 and 108139; BioLegend), CD150 (115903 and 115910; BioLegend), CD48 (103427 and 103431; BioLegend), CD34 (128615; BioLegend; and 48–0341-82; eBioscience), FcγRII/III (101307 and 101327; BioLegend), human CD271/NGFR (345106; BioLegend), Ki67 (50–5698-82; eBioscience), and Annexin V (640920; BioLegend). Dead cells were eliminated by staining with DAPI (422801; BioLegend) or propidium iodide solution (421301; BioLegend). All flow cytometric analyses and cell sorting were performed on a FACSAria or FACSCalibur (BD Biosciences), and all flow cytometry data were analyzed with FlowJo software.
Purification of hematopoietic cells
Hematopoietic cells were harvested from long bones that were triturated and passed through 40-µm nylon mesh to obtain a single-cell suspension. Isolation of lineage-depleted (Lin−) BM cells was performed with a Lineage Cell Depletion kit (130–090-858; Miltenyi Biotec) according to the manufacturer’s protocol. BM cells were incubated with a cocktail of biotinylated monoclonal antibodies against a panel of lineage antigens (CD5, B220, CD11b, Gr1, 7/4, and Ter119) and anti–Biotin MicroBeads and separated on a MACS Column in the magnetic field of a MACS Separator (Miltenyi Biotec). Lin− cells were further stained with a cocktail of biotinylated lineage-specific antibodies. Then, Brilliant violet 605–conjugated streptavidin (405229; BioLegend) and a combination of mAbs, including PE-Dazzle 594– or eFluor450-conjugated anti-CD34, allophycocyanin (APC) or Brilliant violet 785 Sca-1, PE or APC-Cy7-FcγR, PE-Cy7-c-Kit, PE or APC-CD150, and APC-Cy7 or Brilliant violet 421 CD48, APC or APC-Cy7-CD45.2 mAb were used as an additional marker for donor-derived cells in the BM of B6-CD45.1 recipient mice. LSK, LT-HSC (LSK CD150+CD48−), short-term HSC (LSK CD150−CD48−), MPP (LSK CD150−CD48+), CMP (Lin−c-Kit+Sca-1−FcγR−CD34+), GMP (Lin−c-Kit+Sca-1−FcγR+CD34+), and MEP (Lin−c-Kit+Sca-1−FcγR−CD34−) fractions were defined as previously described (Mayle et al., 2013). eFluor660-Ki67, APC-Annexin V, and DAPI were used as additional markers for cell cycle/apoptosis analysis as previously described (Cai et al., 2016). MkPs (CD41highCD71lowMEP) and ErPs (CD41lowCD71highMEP) were defined as previously described (Murphy et al., 2013).
Cell lysates were subjected to immunoblotting using the following antibodies: FLAG (F1804; Sigma-Aldrich), anti-Asxl1 (sc-85283; Santa Cruz Biotechnology), tubulin (B-5-1-2; Santa Cruz Biotechnology), actin (A5541; Sigma-Aldrich), and H2AK119Ub (D27C4; Cell Signaling Technology). H3K4me3 (CMA304), H3K27me3 (CMA323), and H3 were obtained from H. Kimura (Tokyo Institute of Technology, Tokyo, Japan; Kimura et al., 2008).
MOL4070A infection and inverse PCR
Newborn mice were inoculated intraperitoneally with a MOL4070LTR (MOL4070A) retrovirus solution containing ∼1 × 105 virus particles, which was provided by L. Wolff (National Cancer Institute, Bethesda, MD), as previously described (Wolff et al., 2003; Ikeda et al., 2016; Ueda et al., 2016). Retroviral integration sites were identified using inverse PCR as previously described (Yamasaki et al., 2010).
PCR was performed to amplify cDNA derived from the BM cells of Vav-cre–positive Asxl1-MTfl/fl mice using one forward (5′-TTTCACGTATCAAACCACCCTG-3′) and two reverse primes (5′-AGTAGTTGTGTTCGCTGTAGATC-3′ and 5′-GGTACCCTCGAGGAAGTTCC-3′). Sequencing primer (5′-ATCGTCCCCATCACGGAGTC-3′) was used for the Sanger sequencing.
Transfection and retrovirus production
Retroviral production of RUNX1-S291fsX was performed as described previously (Watanabe-Okochi et al., 2008). Id3 cDNA was provided by T. Ikawa (RIKEN, Yokohama, Japan). In brief, retroviruses were generated by transient transfection of RUNX1-S291fsX, MLL-AF9, ASXL1, or Id3 cDNA into Plat-E packaging cells using the calcium phosphate coprecipitation method. Primary BM cells were infected with retroviruses as previously described (Watanabe-Okochi et al., 2008; Inoue et al., 2013).
Mouse BM transplantation assays
Freshly dissected femora and tibiae were isolated from Vav-Cre, Vav-Cre–negative Asxl1-MTfl/wt, Vav-Cre–positive Asxl1-MTfl/wt, Vav-Cre–negative Asxl1-MTfl/fl, Vav-Cre–positive Asxl1-MTfl/fl, and CD45.1 WT mice. BM was flushed with a 3-cc insulin syringe into cold PBS (without Ca2+ and Mg2+) supplemented with 2% BSA to generate single-cell suspensions. BM cells were spun at 1,500 rpm for 5 min by centrifugation, and RBCs were lysed in ammonium–chloride–potassium bicarbonate lysis (ACK) buffer for 5 min on ice. After centrifugation, cells were resuspended in PBS/2% BSA filtered through a 40-µM cell strainer. For competitive transplantation experiments in Fig. 4 A, 0.5 × 106 total BM cells from Vav-Cre or Vav-Cre–positive Asxl1-MTfl/fl (CD45.2 positive) mice were mixed with WT control (CD45.1 positive) mice and transplanted via tail vein injection into 6-wk-old lethally irradiated (2 × 4.75 Gy) CD45.1+ recipient mice. Similarly, for another competitive assay (Fig. S2 C), we mixed BM cells from Vav-Cre–positive Asxl1-MTfl/wt (GFP positive) and Vav-Cre–negative Asxl1-MTfl/wt (GFP negative) mice and transplanted the cells into CD45.1 WT recipient mice. PB chimerism was assessed by flow cytometry. For noncompetitive transplantation experiments, 106 total BM cells from Vav-Cre–negative Asxl1-MTfl/wt control (CD45.2+) and Vav-Cre–positive Asxl1-MTfl/wt mice (CD45.2+) were injected into lethally irradiated (2 × 4.75 Gy) CD45.1+ recipient mice.
For RUNX1-S291fsX retroviral primary BM transplantation experiments, donor cells from Vav-Cre–negative Asxl1-MTfl/WT, Vav-Cre–positive Asxl1-MTfl/WT, Vav-Cre–negative Asxl1-MTfl/fl, and Vav-Cre–positive Asxl1-MTfl/fl mice were treated with a single dose of 150 mg/kg 5-fluoruracil followed by BM harvest from the femora and tibiae 3 d later. RBCs were removed using ACK lysis buffer, and nucleated BM cells were transduced with viral supernatants containing RUNX1-S291fsX for 2 d followed by tail vein injection into sublethally irradiated CD45.1 recipient mice.
Library preparation for whole-transcriptome analysis was performed according to the manufacturer’s instructions (SureSelect Strand Specific RNA Preparation kit; Agilent Technologies). In brief, poly(A) RNA purified from 2 µg total RNA was chemically fragmented to appropriate sizes. First-strand cDNA was synthesized from the fragmented poly(A)-selected mRNA, followed by the synthesis of second-strand cDNA. Then, adapter oligo-DNA was ligated to both ends of the double-stranded cDNA. Libraries were prepared with 12 cycles of PCR amplification of adapter-ligated cDNA using primers that are complimentary to the adapters. The libraries were sequenced using a GAIIx next-generation sequencer (Illumina) using the single-end 36-bp sequencing protocol. The generated sequence tags were mapped onto the mouse genomic sequence (mm10; UCSC Genome Browser) and mRNA expression levels were normalized as reads per kilobase per million. We validated RNA-sequencing data by quantitative RT-PCR using the following primers: Id3, 5′-CTGTCGGAACGTAGCCTGG-3′ and 5′-GTGGTTCATGTCGTCCAAGAG-3′; Gata1, 5′-TATGGCAAGACGGCACTCTAC-3′ and 5′-GGTGTCCAAGAACGTGTTGTT-3′; Runx1, 5′-GATGGCACTCTGGTCACCG-3′ and 5′-GCCGCTCGGAAAAGGACAA-3′; Sox6, 5′-ACCAGTGACTTCTGGGTGCT-3′ and 5′-CAAGTCCTGGGTCATTGCTT-3′; Tjp1, 5′-GCTTTAGCGAACAGAAGGAGC-3′ and 5′-TTCATTTTTCCGAGACTTCACCA-3′; Hba-a1, 5′-CACCACCAAGACCTACTTCCC-3′ and 5′-CAGTGGCTCAGGAGCTTGA-3′; and Actb, 5′-TGTTACCAACTGGGACGACA-3′ and 5′-GGGGTGTTGAAGGTCTCAAA-3′.
For ChIP-seq in mouse cells, 10 million BM c-Kit+ cells isolated with anti–mouse c-Kit Biotin (13–1171-85; eBioscience) and anti–Biotin Microbeads (130–090-485; Miltenyi Biotec) were used. In brief, cells were fixed in a 1% methanol-free formaldehyde solution and then resuspended in SDS lysis buffer. Lysates were sonicated in an E220 focused ultrasonicator (Covaris) to a desired fragment size distribution of 100–500 bp. Immunoprecipitation reactions were performed with the following antibodies, each on ∼500,000 cells as previously described (Krivtsov et al., 2008). Antibodies used for ChIP include anti-H3K4me3 (9751S; Cell Signaling), anti-H3K27me3 (07-449; Millipore), anti-H2AK119Ub (8240; Cell Signaling), anti-Asxl1 (sc-85283; Santa Cruz Biotechnology), and anti-FLAG (F1804; Sigma-Aldrich). ChIP assays were processed and barcoded library and raw data analyses were performed as previously described (Micol et al., 2017).
Statistical significance was determined by unpaired Student’s t test or ANOVA after testing for normal distribution and equal variance, unless indicated otherwise. For Kaplan–Meier survival analysis, the Mantel–Cox log-rank test was used to determine statistical significance. Data were plotted using GraphPad Prism 8 software as mean values; error bars represent SD.
Accession numbers for RNA-seq and ChIP-seq
RNA-sequencing and ChIP-seq data have been deposited into the NCBI Gene Expression Omnibus portal under the accession no. GSE111067.
Online supplemental material
Fig. S1 shows additional characterization of Asxl1-MT KI mice. Fig. S2 shows additional characterization of HSC and progenitor cells of KI mice. Fig. S3 shows transcriptional and epigenetic effects in KI mice. Fig. S4 shows the effects of Asxl1 mutant expression on histone modification. Table S1 contains the motif enrichment analysis of Asxl1 mutant and Asxl1-WT binding sites.
We acknowledge Dr. Linda Wolff, Dr. Yoshiteru Sasaki, Dr. Haruhiko Koseki, Dr. Tomokatsu Ikawa, and Dr. Hiroshi Kimura for providing materials described above.
D. Inoue is supported by the Friends of Leukemia Research Fund and the Japan Society for the Promotion of Science KAKENHI Grant-in-Aid for Young Scientists (26893051) and Overseas Research Fellowships. D. Inoue and O. Abdel-Wahab are supported by the Leukemia and Lymphoma Society. T. Kitamura is supported by the Japan Society for the Promotion of Science KAKENHI Grant-in-Aid for Scientific Research (B 15H04855), a research grant from the Tokyo Biochemical Research Foundation, and a research grant from the Uehara Memorial Foundation. O. Abdel-Wahab is supported by the US Department of Defense Bone Marrow Failure Research Program (grant W81XWH-12-1-0041), the Edward P. Evans Foundation, the Taub Foundation, and the Starr Foundation (grants I8-A8-075 and I9-A9-059).
The authors declare no competing financial interests.
Author contributions: D. Inoue, R. Nagase, H. Honda, T. Fujino, O. Abdel-Wahab, and T. Kitamura designed the study. R. Nagase, D. Inoue, T. Fujino, A. Pastore, N. Yamasaki, H. Honda, H.-A. Hou, Y. Sera, and H.-F. Tien performed experiments and analyzed data. R. Nagase, D. Inoue, A. Pastore, and A. Kanai performed RNA-seq and ChIP-seq analyses. D. Inoue, R. Nagase, O. Abdel-Wahab, and T. Kitamura prepared the manuscript with help from all coauthors.
R. Nagase and D. Inoue contributed equally to this paper.
O. Abdel-Wahab and T. Kitamura contributed equally to this paper.