Hematopoietic stem cells acquire survival advantage by loss of RUNX1 methylation identified in familial leukemia

RUNX1 is one of the most commonly mutated/altered genes in a variety of hematologic diseases, including acute myeloid leukemia (AML), therapy-related leukemia, myelodysplastic syndrome (MDS), and familial platelet disorder with associated myeloid malignancy.^1,2  Since the first identification of RUNX1 as a gene located on the breakpoint of chromosomal translocation t(8;21), which creates the RUNX1-ETO fusion gene, many chromosomal translocations involving the RUNX1 gene have been reported in leukemia, including RUNX1-MECOM, ETV6-RUNX1, and RUNX1-CBFA2T3 fusion genes.^3,4  Somatic and germline point mutations in the RUNX1 gene are also frequently found.^5-7  Multiple lines of evidence indicate that most RUNX1 missense mutations are clustered in the conserved Runt DNA-binding domain and result in defective DNA binding, and that truncating nonsense and frameshift mutations are distributed along the entire coding sequence. However, it is not well understood how missense mutations outside the Runt domain lead to leukemia.

Transcription factors are regulated not only by transcriptional and translational control but also by alternative splicing and posttranslational modifications (PTMs). PTMs include phosphorylation, acetylation, ubiquitination, SUMOylation, glycosylation, and methylation, and they represent an important layer of regulation for fine-tuning transcriptional activity.⁸  Indeed, numerous biochemical and cellular studies have confirmed that RUNX1 activity is regulated either positively or negatively by PTMs in vitro.^9-11  However, unexpectedly, previous mice models with mutant RUNX1 PTMs were phenotypically normal or showed a minimal phenotype.^12,13  Hence, the physiologic relevance of RUNX1 PTMs is largely unknown.

Here we report a family with 3 occurrences of AML. Targeted deep sequencing discovered a novel mutation in RUNX1 that precluded RUNX1 methylation. By using a corresponding mice model, we show that methylation-deficient mutations of RUNX1 lead to an expanded and less quiescent hematopoietic stem cell (HSC) pool with progenitor cell gene priming. We also uncover that defective RUNX1 methylation in HSCs confers resistance to apoptosis induced by endogenous and genotoxic stress, and a subsequent survival advantage under stress conditions, which can explain a preleukemic state in the affected individuals.

Informed consent was obtained from all participating subjects in accordance with the Declaration of Helsinki, and the studies were approved by the Human Research Ethics Committee of Kumamoto University School of Medicine. Genomic DNA was extracted from peripheral blood.

Runx1^KTAMK/KTAMK mice were previously described¹³  and used after more than 5 backcrosses to C57BL/6N mice. Animal care was in accordance with the guidelines of the National University of Singapore.

Fluorescence-activated cell sorting (FACS) analysis and sorting were performed as previously described.¹⁴  Briefly, bone marrow cells were harvested from femurs, tibias, and the spine of 2- to 4-month-old age- and sex-matched mice. The cells were dissociated to a single-cell suspension by filtering through a 70-μm nylon mesh. Cells were Fc-blocked and stained with anti-mouse primary antibodies for 60 minutes. All antibodies were purchased from Thermo Fisher Scientific, BD Biosciences, or BioLegend. For sorting, cKit⁺ cells were preenriched with the CD117 MicroBeads and the MACS LS columns (Miltenyi Biotec). For intracellular analysis, cells were stained by using Cytofix/Cytoperm Fixation/Permeabilization Solution (BD Biosciences) in accordance with the manufacturer’s instructions. Aldehyde dehydrogenase (ALDH) enzymatic activity was measured by using the Aldefluor kit (Stemcell Technologies). For apoptosis analysis, cells were stained with Annexin V, Alexa Fluor 488, or APC conjugate (Thermo Fisher Scientific), Hoechst 33342, and Annexin V Binding Buffer (BD Biosciences). Cells were analyzed or sorted by using LSRII and FACS Aria II cytometers (BD Biosciences). Subsequent data analyses were performed with the FlowJo analysis software (FlowJo, LLC).

Bone marrow cells were harvested from control and Runx1^KTAMK/KTAMK (Ly5.2) mice, and a mixture of 2 × 10⁵ test cells and 1 × 10⁶ competitor cells from C57BL/6-Ly5.1 mice were transplanted into lethally irradiated (9.5 Gy, total body irradiation) C57BL/6-Ly5.1 congenic mice. Peripheral blood donor chimerism was analyzed monthly. Recipient mice were euthanized for analysis 4 months after bone marrow transplantation (BMT). Secondary transplantations into lethally irradiated C57BL/6-Ly5.1 mice were performed by using 2 × 10⁶ bone marrow cells from primary recipients. To provoke stress conditions, lethally irradiated (9.5 Gy) Ly5.1 recipient mice were coinjected with a mixture of indicated number of test cells (Ly5.2) and competitor whole bone marrow cells (Ly5.1). After BMT, mice were treated with 4 mg/kg of polyinosinic:polycytidylic acid (pIpC; InvivoGen) intraperitoneally twice a week or irradiated (2 Gy) once a week.

All values are presented as the mean ± standard deviation. Statistical analysis was performed by using the Student t test. Values were considered to be significant at P < .05. Other methods are provided in the supplemental Methods (available on the Blood Web site).

We performed targeted deep sequencing on a family with 3 occurrences of AML. The family pedigree is shown in Figure 1A. The proband (II-1) had AML (M1), patient II-2 had refractory anemia with excess blasts/AML, and patient II-8 had AML (M2) (Table 1). Genomic DNA samples were extracted from the peripheral blood of 3 affected and 2 unaffected individuals of this family (II-1, II-2, II-8, III-1, and III-2), and all coding exons of 545 candidate genes that can be involved in malignancy (supplemental Table 1) were sequenced. After removing variations in untranslated regions, variations registered in the dbSNP database¹⁵  or the 1000 Genomes Project,¹⁶  and synonymous variations, only one RUNX1 mutation was detected as common among the 3 affected individuals but not shared with the 2 unaffected individuals. Subsequent Sanger sequencing confirmed a heterozygous mutation that caused amino acid substitution from arginine (AGG) to lysine (AAG) (R237K; p.Arg210Lys and c.629G>A for RUNX1a or RUNX1b, and p.Arg237Lys and c.710G>A for RUNX1c) (Figure 1B). Although the mutated arginine residue was outside the Runt DNA-binding domain where most previously reported mutations were located (Figure 1C),¹  comparative genomics showed that the mutated residue was completely conserved among species and among RUNX family proteins (Figure 1D-E), implicating its functional importance. Indeed, the mutated R237 is one of two arginine residues (R233 and R237) methylated by protein arginine methyltransferase 1 (PRMT1). RUNX1 methylation by PRMT1 dissociates the transcription corepressor, SIN3 transcription regulator family member A (SIN3A), and potentiates the transcriptional activity of RUNX1.¹¹  Thus, loss of RUNX1 methylation by the R237K mutation was considered to be the most likely cause of familial AML in this pedigree.

To explore the biologic significance of RUNX1 methylation in vivo, we used RUNX1 R233K/R237K double-mutant Runx1^KTAMK/KTAMK mice, in which 2 arginine-to-lysine (RTAMR-to-KTAMK) mutations precluded RUNX1 methylation by PRMT1 (Figure 2A). We previously reported that Runx1^KTAMK/KTAMK mice exhibited a slight decrease in lymphocytes and a slight increase in neutrophils and monocytes in peripheral blood, and a reduction of T cells in spleen.¹³  In addition, detailed analysis of Runx1^KTAMK/KTAMK bone marrow showed that the percentage of CD34^–Flt3^– Lin^–Sca1⁺cKit⁺ (LSK) long-term HSCs (LT-HSCs) increased approximately twofold, and that CD34⁺Flt3^– LSK short-term HSCs also increased in Runx1^KTAMK/KTAMK mice (Figure 2B). Phenotypic analysis of HSCs and multipotent progenitors (MPPs) using SLAM markers also showed an increase of HSCs, multipotent MPP2, and myeloid-biased MPP3 in Runx1^KTAMK/KTAMK mice (Figure 2C). Conversely, loss of RUNX1 R233/R237 methylation did not alter more committed progenitors and Lin⁺ cells in bone marrow, except for a slight increase of the Lin^–Sca1^–cKit⁺ progenitor fraction (supplemental Figure 1A-B). Cell cycle analysis revealed that Runx1^KTAMK/KTAMK LT-HSCs were less quiescent, which could explain the increase of LT-HSCs (Figure 2D).

We next assessed how loss of RUNX1 R233/R237 methylation affected hematopoietic stem/progenitor cell (HSPC) potential in vitro and in vivo. Bone marrow cells from control Runx1^+/+, Runx1^+/KTAMK, and Runx1^KTAMK/KTAMK mice were cultured in methylcellulose-based medium for 12 days. Runx1^+/KTAMK and Runx1^KTAMK/KTAMK bone marrow showed increased short-term colony-forming activity in vitro (Figure 2E). In agreement with the in vitro data, competitive BMT assays showed that Runx1^KTAMK/KTAMK bone marrow cells exhibited higher myeloid-lineage reconstitution in peripheral blood in 4 weeks. However, Runx1^KTAMK/KTAMK bone marrow cells did not show a statistically significant increase in reconstruction activity 16 weeks after the BMT in peripheral blood and bone marrow (Figure 2F; supplemental Figure 1C-D), and the higher repopulating capacity of Runx1^KTAMK/KTAMK cells was totally lost in the secondary BMT (supplementary Figure 1E). This outcome may suggest that the higher short-term reconstitution capacity of Runx1^KTAMK/KTAMK cells was canceled by exhaustion due to the loss of quiescence observed in Figure 2D.

Thus, although dispensable for hematopoietic cell differentiation to all lineages, the loss of RUNX1 R233/R237 methylation perturbs HSC quiescence and expands a phenotypic HSC pool with no improvement of long-term reconstitution activity.

To elucidate the nature of the Runx1^KTAMK/KTAMK LT-HSC defects, RNA-sequencing of LT-HSCs from Runx1^+/+ and Runx1^KTAMK/KTAMK mice was performed (Figure 3A). First, we compared our RNA-sequencing data vs published data of RUNX1-depleted HSCs. Gene set enrichment analysis (GSEA) using gene sets of control and Runx1^flox/flox; Vav1-Cre Flt3^– LSK cells¹⁷  revealed that genes upregulated in Runx1^flox/flox; Vav1-Cre Flt3^– LSK cells were also enriched in Runx1^KTAMK/KTAMK LT-HSCs, and that genes downregulated in Runx1^flox/flox; Vav1-Cre Flt3^– LSK cells were enriched in our control Runx1^+/+ LT-HSCs (Figure 3B). This scenario indicates that in vivo loss of RUNX1 R233/R237 methylation in LT-HSCs at least partially phenocopies RUNX1 deficiency and conforms to in vitro observations that defective RUNX1 methylation impairs RUNX1 transcriptional activity.^11,13 

GSEA using gene sets of HSCs and a series of MPPs¹⁸  revealed that genes associated with MPPs were enriched in Runx1^KTAMK/KTAMK LT-HSCs, whereas genes associated with normal HSCs were enriched in control Runx1^+/+ LT-HSCs, indicating Runx1^KTAMK/KTAMK LT-HSCs were primed for progenitor cell–specific gene expression (Figure 3C; supplemental Figure 2A). In agreement with these results, genes related to HSC function, including Alcam,^19,20 Aldh1a1,²¹ Jun, and Junb,²²  were downregulated in Runx1^KTAMK/KTAMK LT-HSCs. Reduced expression of these genes was verified by quantitative polymerase chain reaction (Figure 3D). Activated leukocyte cell adhesion molecule (ALCAM), also known as CD166, mediates homophilic (CD166-CD166) and heterophilic (CD166-CD6) adhesion¹⁹  and is a functional marker of HSCs with high long-term repopulating potential.²⁰  ALDH is also highly expressed in HSPCs and marks cells with high repopulating function.²¹  Consistent with the reduced messenger RNA (mRNA) expression of Alcam and Aldh1a1, cell surface ALCAM protein and intracellular ALDH activity were also decreased in Runx1^KTAMK/KTAMK HSCs (Figure 3E-F). In addition, GSEA using the Gene Ontology biologic process database revealed that genes upregulated in Runx1^KTAMK/KTAMK LT-HSCs were enriched in many cell cycle–related categories, reflecting the loss of quiescence in Runx1^KTAMK/KTAMK LT-HSCs (supplemental Figure 2B-C).

Next, we globally profiled chromatin accessibility in control and Runx1^KTAMK/KTAMK LT-HSCs using assay for transposase-accessible chromatin using sequencing (ATAC-sequencing). Although overall chromatin accessibility was comparable between control and Runx1^KTAMK/KTAMK LT-HSCs (Figure 4A), the Aldh1a1 promoter showed a higher peak in control LT-HSCs, reflecting higher expression of corresponding mRNA as detected by RNA-sequencing (Figure 4B). In contrast, the promoter regions of MPP-related genes (Rag1 and Rag2) and a cell cycle–related gene Ccnd1 had higher peaks in Runx1^KTAMK/KTAMK LT-HSCs. De novo motif discovery revealed that the sequence TGTGGTTT, which completely matched the known RUNX1 motif, was enriched in regions with higher chromatin accessibility in control HSCs; this was not the case when examining the regions with higher peaks in Runx1^KTAMK/KTAMK HSCs (Figure 4C). This scenario suggests that loss of RUNX1 R233/R237 methylation decreases its ability to open the chromatin regions around the RUNX1 binding sites.

We next took advantage of published chromatin immunoprecipitation (ChIP)-sequencing data sets from LT-HSCs and MPPs for 2 histone modifications, histone H3 lysine 27 acetylation (H3K27Ac) for active enhancers and histone H3 lysine 4 monomethylation (H3K4me1) for both poised and active enhancers (Figure 4D). After identifying enhancer regions with a higher histone modification signal in LT-HSCs or MPPs, we compared ATAC peak signal in these regions between control and Runx1^KTAMK/KTAMK LT-HSCs. Although we observed similar ATAC signal intensity in enhancer regions associated with LT-HSCs, MPP-associated enhancer regions exhibited a higher ATAC peak signal in Runx1^KTAMK/KTAMK LT-HSCs. This indicates that Runx1^KTAMK/KTAMK LT-HSCs have an epigenetic signature skewed toward MPPs. Because R233 and R237 are outside the DNA-binding domain, it is unlikely that mutant RUNX1 acquires the ability to bind to a novel motif other than the canonical RUNX1 motif and binds preferentially to the promoters/enhancers of MPP signature genes. Upregulation of MPP signature genes in Runx1^KTAMK/KTAMK LT-HSCs is probably due to the secondary effect of loss of HSC signature genes.

Collectively, these analyses suggest that loss of RUNX1 R233/R237 methylation changes the genomic and epigenomic signature of phenotypic LT-HSCs to a poised progenitor state.

GSEA using the Gene Ontology biologic process database revealed that 3 pathways were significantly enriched in Runx1^+/+ LT-HSCs compared with Runx1^KTAMK/KTAMK LT-HSCs. Notably, all 3 of these pathways were related to unfolded protein response (UPR) and endoplasmic reticulum (ER) stress (Figure 5A; supplemental Figures 2B and 3A). We thus analyzed how loss of RUNX1 R233/R237 methylation affected cellular responses against ER stress. Runx1^+/+ and Runx1^KTAMK/KTAMK LT-HSCs were treated with the ER stress inducer tunicamycin, and the expression of a series of UPR genes was examined. As expected, defects in RUNX1 methylation diminished tunicamycin-stimulated induction of UPR genes, including Atf4, Ddit3, Gadd34, Atf6, Bip, Edem1, and Sec61a1. Subsequent apoptosis triggered by ER stress was also attenuated in Runx1^KTAMK/KTAMK cells (Figure 5C).

Because Runx1^KTAMK/KTAMK LT-HSCs lose quiescence (Figure 2D) and p53 is critical for HSC quiescence,²³  we examined whether p53 pathways were impaired by defective RUNX1 methylation. Runx1^+/+ and Runx1^KTAMK/KTAMK mice were irradiated and Flt3^– LSK HSCs were analyzed. Although radiation resulted in a prominent increase of the p53 target genes Bbc3, Pmaip1, and Cdkn1a, the induction was partially suppressed in Runx1^KTAMK/KTAMK mice (Figure 5D), and apoptosis was also attenuated in Runx1^KTAMK/KTAMK mice (Figure 5E). Apoptosis after DNA damage is one of the critical mechanisms to prevent proliferation of damaged cells and to maintain genome integrity of HSCs. Hence, we evaluated DNA damage accumulation in Runx1^KTAMK/KTAMK Flt3^– LSK HSCs subjected to in vivo radiation. As shown in Figure 5F, γH2AX staining indicative of DNA double-strand breaks was increased in Runx1^KTAMK/KTAMK cells.

These results show that defects in RUNX1 methylation confer resistance to apoptosis induced by endogenous and genotoxic stress despite the presence of DNA damage accumulation, which is a hallmark of a preleukemic clone.^17,24-28 

We next performed competitive BMT experiments to address whether Runx1^KTAMK/KTAMK HSPCs exhibited a survival advantage due to resistance to apoptosis in vivo. Lethally irradiated Ly5.1 recipient mice were coinjected with a mixture of either Runx1^+/+ or Runx1^KTAMK/KTAMK (Ly5.2) LSK cells and Ly5.1 competitor cells. After BMT, recipient mice were treated with pIpC twice a week to induce sustained inflammation by a type I interferon response (Figure 6A). Although the repopulating activity of control Runx1^+/+ HSPCs was impaired by serial pIpC treatment, Runx1^KTAMK/KTAMK HSPCs showed relative resistance to pIpC treatment and higher reconstitution after 4 weeks. We next evaluated the effects of nonlethal repetitive irradiation on Runx1^+/+ and Runx1^KTAMK/KTAMK HSPCs (Figure 6B). Repeated irradiation (2 Gy) once a week after BMT also augmented the repopulation activity of Runx1^KTAMK/KTAMK HSPCs but not Runx1^+/+ HSPCs. These assays have consistently revealed the survival advantage of Runx1^KTAMK/KTAMK HSPCs under stress conditions.

Activating transcription factor 4 (ATF4) protein expression is well known to be regulated translationally by eukaryotic translation initiator factor-2 (eIF2α) phosphorylation through the protein kinase RNA–like ER kinase (PERK) pathway.²⁹  Conversely, transcriptional regulation of Atf4 mRNA is poorly understood. Our finding that the induction of Atf4 mRNA was diminished in Runx1^KTAMK/KTAMK cells (Figure 5B) prompted us to hypothesize that Atf4 might be a direct downstream target of RUNX1. To search for this possibility, publicly available RUNX1 ChIP-sequencing data were analyzed. We found that RUNX1 bound to the promoter region of ATF4 both in human CD34⁺ HSPCs and mouse EML hematopoietic precursor cells (Figure 7A).^30,31  The promoter activity of Atf4 as marked by H3K27Ac in Runx1^KTAMK/KTAMK LSK HSPCs was reduced compared with control Runx1^+/+ cells (Figure 7B). Reporter assays showed that the human ATF4 promoter was transactivated by wild-type RUNX1 and mutant RUNX1 R233K, and as expected, the activation was diminished by the R237K mutation (Figure 7C). Furthermore, reporter assays using a series of deletion/mutant constructs revealed that the transactivation effects were mediated mainly by the RUNX1 binding site between −61 bp and −55 bp of the transcription initiation site both in hematopoietic and non-hematopoietic cells (Figure 7D; supplemental Figure 4A). These data show that RUNX1 transcriptionally regulates ATF4.

Deep targeted sequencing of >500 candidate genes identified a novel germline mutation of RUNX1 in a pedigree of familial AML. This mutated residue is a methylation site by PRMT1, and previous studies confirmed that loss of methylation resulted in impaired transcriptional activity of RUNX1.^11,13  To study the biologic significance of RUNX1 methylation in vivo, we analyzed corresponding knock-in mice as a disease model and found that defective RUNX1 methylation induced HSC expansion, loss of quiescence, and an impaired genomic and epigenomic HSC signature in phenotypic LT-HSCs. HSCs with defects in RUNX1 methylation acquired a survival advantage by resistance to apoptosis, which explained predisposition to malignancy in this pedigree. We also identified ATF4 as a direct transcriptional target of RUNX1.

Here we reported a novel case of the familial germline RUNX1 R237K mutation (Figure 1). RUNX1 R237K mutations have been reported twice as somatic and sporadic mutations (COSM87292) in the Catalogue Of Somatic Mutations In Cancer database⁵ : one among chronic myelomonocytic leukemia³²  and the other among −7/del(7q) AML.³³  These 2 sporadic cases further corroborate the causal relationship of loss of RUNX1 R237 methylation and AML. However, the germline variation of the other methylation site, R233H, is found in general population databases, and recently ClinGen expert panels have classified R233H as likely benign because its allele frequency is greater than expected for disorder.³⁴  This suggests that R233 methylation is not biologically important for the activity of RUNX1, and it is unlikely that loss of R233 methylation causes leukemia. Although our experiments to study the effects of RUNX1 methylation were conducted by using R233K/R237K double-mutant mice, we speculate that the single R237K mutation is sufficient to cause familial AML in human.

Previous in vitro studies have revealed that RUNX1 is regulated by PTMs, including phosphorylation, acetylation, methylation, and ubiquitination.^9-11  However, RUNX1 mutations at the PTM sites are extremely rare even in sporadic cases; the exception is K83, a putative ubiquitination acceptor site, whose mutation is likely to contribute to leukemogenesis not through loss of ubiquitination but by disrupted DNA binding of RUNX1.⁹  Moreover, mice with RUNX1 mutations that prevented phosphorylation at either S249/S266 or S249/S276 were phenotypically normal.¹²  Only in vitro data have confirmed that RUNX1 with 4 phosphorylation-deficient mutations impairs T-cell differentiation activity, and RUNX1 with 5 phosphorylation-deficient mutations completely loses its hematopoietic activity.¹⁰  Acetylation and ubiquitination at K24/K43 also appear to be dispensable for RUNX1 function.³⁵  These observations have thus challenged the predicted biologic importance of RUNX1 PTMs in vivo. RUNX1 and SIN3A bind via the SIN3A interaction domain (172-237 for RUNX1b and 199-264 for RUNX1c).³⁶  Previous in vitro studies have shown that PRMT1 methylates R233 and R237 inside the SIN3A interaction domain and that PRMT1 abrogates the binding of SIN3A to RUNX1 wild type, but not to RUNX1 R233K/R237K, and potentiates its transcriptional activity.¹¹  However, the in vivo consequence of RUNX1 methylation was largely unknown. Our results obtained from human and mice data have unequivocally confirmed the physiologic relevance of RUNX1 PTMs in vivo.

Hematopoietic phenotypes observed in Runx1^flox/flox; Mx1-Cre mice^37-40  or Runx1^flox/flox; Vav1-Cre mice^17,38,41  were partially recapitulated in Runx1^KTAMK/KTAMK mice, including phenotypic HSC expansion with loss of quiescence and antiapoptotic properties. These findings suggest that loss of RUNX1 R233/R237 methylation diminishes its transcriptional activity in vivo (Figure 2). We showed that phenotypic LT-HSCs increased in Runx1^KTAMK/KTAMK mice and that Runx1^KTAMK/KTAMK bone marrow cells exhibited increased short-term in vitro colony-forming activity and short-term myeloid lineage reconstitution in vivo. These findings are in line with previous reports in which RUNX1 deficiency expanded HSPCs, especially myeloid progenitors.^38,41  However, we detected no improved long-term reconstitution activities of Runx1^KTAMK/KTAMK cells. Premature initiation of the progenitor gene expression program in LT-HSCs may impair HSC properties in Runx1^KTAMK/KTAMK HSCs and lead to HSC exhaustion, which is shown in Runx1^flox/flox; Mx1-Cre mice^39,40  (Figures 3 and 4).

UPR is a network of interconnected signaling pathways to restore ER homeostasis initiated by 3 signal transducers: inositol-requiring protein 1 (IRE1), activating transcription factor 6 (ATF6), and PERK.²⁹  Under ER stress conditions, PERK inhibits general protein translation through the phosphorylation of eIF2α to decrease protein overload and also leads to the selective translation of ATF4 mRNA. Upregulation of ATF4 promotes an antioxidant response, the folding capacity of the ER, and macroautophagy. However, if homeostasis cannot be restored, sustained ATF4 expression initiates apoptosis. HSCs have higher basal PERK–eIF2α activation and preferentially activate the pro-apoptotic PERK–eIF2α–ATF4 branch of UPR, sensitizing them to ER stress–induced apoptosis to maintain HSC integrity.⁴²  We have shown that ATF4 is a direct transcriptional target of RUNX1, a previously unrecognized direct link between RUNX1 and UPR genes (Figure 7). Recent studies have reported that 3 branches of UPR cross-react, and the PERK–eIF2α–ATF4 pathway can activate both the IRE1⁴³  and the ATF6^44,45  pathways. Thus, RUNX1 may activates all 3 branches of UPR via ATF4. However, the interrelationship among the 3 UPR branches has only been studied in specific cell types and may not apply ubiquitously to HSCs. It is possible, therefore, that RUNX1 directly activates UPR genes other than Atf4.

We have clarified that defects in RUNX1 methylation provide HSCs with a growth advantage to overcome endogenous and genotoxic stress (Figures 5 and 6). Impaired UPR and p53 pathways and resistance to stress were also reported in Runx1^flox/flox; Vav1-Cre HSCs.¹⁷  Although the similarity between Runx1^KTAMK/KTAMK mice and Runx1^flox/flox; Vav1-Cre mice further fortifies the notion that loss of RUNX1 methylation suppresses its transcriptional activity in vivo, it is somewhat surprising in light of its modest effects in vitro and relatively weak phenotypes of Runx1^KTAMK/KTAMK mice. These findings might suggest that a substantial proportion of RUNX1 activities in HSCs are regulated by RUNX1 methylation in vivo, or that RUNX1 methylation has a specific role in these pathways. Resistance to apoptosis is considered a hallmark of a preleukemic clone as programmed cell death by apoptosis serves as a natural barrier to cancer development.^17,24-27  We speculate that defective RUNX1 methylation can confer resistance to apoptosis even in DNA damage–accumulated HSCs. Attenuated p53 pathways by loss of RUNX1 methylation may explain a complex karyotype in this pedigree, which is a typical feature of TP53 mutated leukemia.⁴⁶ 

Runx1 deletion per se is not sufficient for leukemogenesis in mouse models,^37-39  indicating that cooperating “second-hit” mutations are required. Of note, as mentioned earlier, the identical RUNX1 R237K mutation was reported as a somatic and sporadic mutation in −7/del(7q) AML, and one of our patients (II-1) also had a −7 karyotype (Table 1). Concomitant functional loss of RUNX1 and EZH2, one of the genes responsible for −7/del(7q) MDS/AML, results in a highly penetrant and fatal MDS phenotype,⁴⁷  and an additional activating Flt3 mutation leads to aggressive acute leukemia.⁴⁸  Although Runx1^KTAMK/KTAMK mice do not exhibit any hematologic malignancies, additional loss of genes on chromosome 7q, including Ezh2 and Cux1,⁴⁹  may cause leukemia. Further studies are required to clarify whether and how loss of RUNX1 methylation and additional mutations collaborate to initiate leukemia.

In conclusion, we present a new germline mutation of RUNX1 at the PTM site in a family with 3 occurrences of myeloid malignancy. By combining human and murine data, we have clearly shown that PTM of RUNX1 is associated with leukemogenesis. Our studies will lead to a better understanding of how PTM dysregulation and subsequent dysfunction of a transcription factor can contribute to human disease.

The authors thank all patients and their families for participation in this research study. They also thank Hibiki Mochida, Deng Jianwen, and Chua Lee Hui for technical assistance.

This research was supported by the National Research Foundation Singapore and the Singapore Ministry of Education under its Research Centres of Excellence initiative (NMRC/STaR/MOH-000149, T.S.; NMRC/CNIG/1155/2016, T.M.), and JSPS Grant-in-Aid for Scientific Research (S) (18H05284, T.S.).

Contribution: T.M. and T.S. conceived and designed the experiments; T.M., A.N.-I., S.S.N.A.M., D.Q.T., and C.Q.W. performed the experiments; T.M., K.T., N.A., and M.O. identified study subjects, performed clinical phenotyping, and contributed biologic samples; T.M., R.T.-M., S.S., and T.B. analyzed bioinformatics data; T.O. generated the mice model; T.M. and T.S. wrote the manuscript; and T.B., T.O., N.A., M.M., M.O., and T.S. jointly supervised the research.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Takayoshi Matsumura, Cancer Science Institute of Singapore, National University of Singapore, 14 Medical Dr, Singapore 117599; e-mail: csitkm@nus.edu.sg or Toshio Suda, Cancer Science Institute of Singapore, National University of Singapore, 14 Medical Dr, Singapore 117599; e-mail: csits@nus.edu.sg.