Establishing human iPSC lines with FPDMM-mimicking RUNX1 heterozygous mutations
To establish human iPSC lines with FPDMM-mimicking RUNX1+/− mutations, we performed CRISPR–Cas9-mediated genome editing of RUNX1, followed by single-cell cloning of the resulting genome-edited cells (Fig. 1a) [39–41]. We introduced two distinct RUNX1 mutations: R201Q and Y287X [3]. R201Q (also known as R174Q depending on the RUNX1 isoform) is characterized by a C-to-T transversion in exon 5 that introduces a single missense codon in the RUNT homology domain (RHD) that affects the decreased functions of RUNX1, such as dimerization with core-binding factor CBFβ, protein stabilization, and DNA-binding, in all three main RUNX1 isoforms, RUNX1a, b, and c [42–46]. Y287X (also reported as Y260X) is characterized by a G-to-T transversion in exon 7b, which introduces a stop codon in the TAD and affects the decreased function of RUNX1b and RUNX1c isoforms, but not that of RUNX1a (Fig. 1b) [43, 44].
We confirmed the successful introduction of the RUNX1 mutations using a T7E1 assay (Supplementary Fig. 1a). We next isolated 37 and 24 single-cell clones from the R201Q-targeted and Y287X-targeted heterogeneous cell populations, respectively. Among these, two of the 37 and two of the 24 clones showed the RUNX1WT/R201Q (RUNX1 wild-type and R201Q) and RUNX1WT/Y287X (RUNX1 wild-type and Y287X) heterozygous point mutations, respectively (Fig. 1c).
To confirm whether these mutations were monoallelic, we performed TA cloning [47] of the mutated regions followed by capillary sequencing. Of the 10 Escherichia coli clones tested for each mutation, five had the C-to-T transversion representing the R201Q mutation, whereas the other five did not. Additionally, five had the G-to-T transversion representing the Y287X mutation, whereas the other five did not. Furthermore, WES analysis confirmed the absence of meaningful off-target mutations in iPSCs with FPDMM-mimicking mutations (Supplementary Fig. 1b and 1c). Although we identified total 8 putative SNVs at protein-coding regions in two cell lines, of which the expression of genes associated with three of these SNVs showed an upregulation tendency in HPCs compared with iPSCs, none of these genes appeared to be related to FPDMM symptoms or megakaryopoiesis [48–52], suggesting that the effect of these variants on cellular function, particularly megakaryopoiesis, was insignificant (Supplementary Fig. 1d–1f). These results indicated that homologous recombination-mediated heterozygous knockin was accurately introduced at the targeted RUNX1 locus without significant off-target mutations.
Human iPSCs with FPDMM-mimicking RUNX1 heterozygous mutations showed impaired hematopoietic and megakaryocytic differentiation in vitro
To evaluate the differentiation efficiencies of the human iPSCs with FPD-mimicking RUNX1+/− mutations into hematopoietic lineage cells, we first induced the differentiation of the RUNX1+/− iPSCs into HPCs (Fig. 2a and Supplementary Fig. 2a). Suspended cells were harvested on day 12 and sorted according to the expression of CD34 and CD45 (a hematopoietic stem/progenitor cell marker and a pan-leukocyte marker, respectively) using flow cytometer (Fig. 2b). We found that the number of HPCs that emerged from RUNX1WT/R201Q and RUNX1WT/Y287X iPSCs was significantly decreased to 24% (P = 0.008, Student’s t-test) and 34% (P = 0.042, Welch’s t-test), respectively, compared with that emerging from wild-type iPSCs (Fig. 2c and Supplementary Fig. 2c).
We next induced the Mk differentiation of RUNX1+/− HPCs (Fig. 2d and Supplementary Fig. 2b). We found that the frequencies of CD41+CD42b+ Mks in RUNX1WT/R201Q and RUNX1WT/Y287X HPCs were 60% (P = 0.004, Student’s t-test) and 78% (P = 0.007, Student’s t-test), respectively, of that in wild-type HPCs (Fig. 2e).
Thus, these results suggested that RUNX1 heterozygous mutation impairs the hematopoietic and megakaryocytic differentiation potential of iPSCs.
Differentially methylated CpGs in FPDMM-mimicking HPCs
We performed single-base resolution methylome analyses to characterize the global DNA methylation status of RUNX1+/− HPCs. By filtering out read depths of less than 10×, we detected 3 844 825, 5 610 028, and 6 329 803 CpGs in RUNX1WT/R201Q, RUNX1WT/Y287X, and wild-type HPCs, respectively. Of these, 781 298 CpGs were present in the intersection among all samples. We identified 1231 hypermethylated and 166 hypomethylated CpGs in RUNX1WT/R201Q HPCs and 1347 hypermethylated and 274 hypomethylated CpGs in RUNX1WT/Y287X HPCs compared with wild-type HPCs using a logistic regression test, indicating a clear bias towards hypermethylation. Of these, 514 hypermethylated and 11 hypomethylated CpGs were altered in common between RUNX1WT/R201Q and RUNX1WT/Y287X HPCs, with a greater overlap observed among the hypermethylated CpGs (Fig. 3a).
We previously reported that TF-binding motif-enrichment analysis of differentially methylated regions predicted DNA methylation-regulating TFs [17–20]. To identify the putative TFs responsible for differential DNA methylation, we performed a TF-binding motif-enrichment analysis using CentriMo [37]. Interestingly, we noticed that binding motifs for ETS family TFs, such as GABPA, FEV, and ELF1, were significantly enriched in the vicinity of hypermethylated DMCs in both RUNX1WT/R201Q and RUNX1WT/Y287X HPCs (Fisher E-value = 1.3e-29 [GABPA]; Fig. 3b, left; Fisher E-value = 6.6e-34 and 7.4e-22 [FEV and ELF1, respectively]; Fig. 3b, right). ETS family TF motifs were also enriched in the vicinity of the commonly hypermethylated DMCs between RUNX1WT/R201Q and RUNX1WT/Y287X HPCs and RUNX1WT/Y287X HPC-specific hypermethylated DMCs (Fisher E-value = 1.4e-18, 8.5e-17, 5.3e-16, 3.9e-13, and 4.1e-13 [FEV, ELF1, GABPA, ELF2, and ELK1, respectively]; Supplementary Fig. 3a and Fisher E-value = 1.6e-16 [FEV]; Supplementary Fig. 3b) but not RUNX1WT/R201Q HPC-specific hypermethylated DMCs.
We did not detect any significantly enriched TF-binding motifs in the vicinity of hypomethylated DMCs in either RUNX1WT/R201Q or RUNX1WT/Y287X HPCs. Notably, the RUNX1-binding motif was not significantly enriched in the vicinity of hypo- or hypermethylated DMCs (Supplementary Fig. 3c).
These results suggested that, despite the slight differences between RUNX1+/− mutations, ETS family TFs are significantly involved in establishing the differential hypermethylation profiles in FPDMM-mimicking HPCs.
Identification of ETS family TFs associated with differential hypermethylation in FPDMM-mimicking HPCs
Because the ETS family consists of 28 protein-coding genes in humans [53] that share similar binding motifs, the TF-binding motif-enrichment analysis did not specifically identify corresponding ETS family TFs. Therefore, to identify putative ETS family TFs associated with differential hypermethylation in FPDMM-mimicking HPCs, we examined the expression of ETS family TF genes using qRT-PCR. We detected a statistically significant decrease in the expression of ELF1 and FLI1 in RUNX1WT/Y287X HPCs (P = 0.003 and P = 0.032, respectively; Welch’s t-test; Fig. 3c and Supplementary Fig. 3d) compared with that in wild-type cells. In addition, we observed a downregulation tendency in RUNX1WT/R201Q HPCs, although without statistical significance (P = 0.10 and P = 0.43, respectively; Welch’s t-test). We further confirmed that FLI1 was consistently downregulated in cells hematopoietically differentiated from iPSCs derived from patients with FPDMM bearing the RUNX1WT/Y287X mutation (GSE54295) (Supplementary Fig. 3e). Similarly, ELF1 tended to be downregulated, albeit with some variability. Additionally, chromatin immunoprecipitation sequencing (ChIP-seq) for human bone marrow CD34+ cells (SRR772107) revealed that FLI1-binding was enriched in commonly hypermethylated DMCs compared with that in randomly selected CpGs (Fig. 3d). These results suggested that FLI1 was associated with differential hypermethylation in RUNX1WT/Y287X HPCs.
Ectopic overexpression of FLI1 induced binding site-directed DNA demethylation across the genome
To validate whether FLI1 induces DNA demethylation at its binding sites in vitro, we overexpressed FLI1 and performed methylome analysis. We established an iPSC line harboring DOX-inducible FLI1 using a lentivirus vector in which FLI1 was ectopically overexpressed upon DOX treatment (272.4-fold increase by highly concentrated DOX in this case; Fig. 4a). Comparing the methylomes of FLI1-overexpressing and non-overexpressing iPSCs, we identified 1620 demethylated and 2919 methylated CpGs at a differential methylation level of > 80%. TF-binding motif-enrichment analysis showed that the FLI1-binding motif was significantly enriched in the immediate vicinity of demethylated CpGs (Fisher E-value = 5.4e-3; Fig. 4b), indicating that FLI1 induced DNA demethylation in a binding site-directed manner.
FLI1 downregulation repressed megakaryocytic differentiation
To determine the impact of FLI1 on megakaryopoiesis, we knocked-down FLI1 during HPC and Mk differentiation (Fig. 5a). We found that the average numbers of CD34+CD45+ HPCs on day 12 were not significantly decreased in FLI1-knockdown cells relative to that in negative-control cells (Supplementary Fig. 4a and 4b). However, the ratios of CD41+CD42b+ Mks were significantly decreased in FLI1-knockdown HPCs by 3 d after the Mk differentiation onset (day 15) (P = 0.036, Student’s t-test) compared with that in the negative-control cells, suggesting the contribution of FLI1 to Mk differentiation (Fig. 5b and 5c).
FLI1 overexpression rescued megakaryocytic differentiation efficiency
To evaluate whether FLI1 overexpression improved deficient Mk differentiation in RUNX1WT/Y287X cells, we established a RUNX1WT/Y287X iPSC line expressing FLI1 upon DOX treatment and mock control cell line transfected with an empty lentivirus vector (Y287X-FLI1 and Y287X-mock, respectively; Fig. 6a). After adding low-concentrated DOX from day 7 of hematopoietic differentiation, we evaluated the expression of FLI1 in HPCs, HPC emergence on day 12, and Mk differentiation efficiency on day 15. qRT-PCR results revealed an average 1.3-fold increase in the expression of FLI1 (P = 0.043, paired t-test) in Y287X-FLI1 HPCs compared with Y287X-mock HPCs (Fig. 6b).
Although HPC emergence was comparable in Y287X-FLI1 and Y287X-mock cells (Supplementary Fig. 5a and 5c), Mk differentiation efficiency was significantly improved by FLI1 overexpression in Y287X-FLI1 HPCs compared with that in Y287X-mock cells (P = 0.010, paired t-test; Fig. 6c; Supplementary Fig. 5b). Collectively, these results suggested that FLI1 rescues deficient Mk differentiation caused by heterozygous mutations in the TAD of RUNX1.
FLI1 overexpression improved differential hypermethylation in RUNX1WT/Y287X HPCs
We performed single-base resolution methylome analyses to explore effect of FLI1 overexpression at the hypermethylated DMCs in RUNX1WT/Y287X HPCs. Comparing the percent methylation scores of these CpGs between Y287X-FLI1 and Y287X-mock HPCs, we identified an overall bias towards hypomethylation in Y287X-FLI1 HPCs (Fig. 7a and Supplementary Table 6). For example, the score of chr1: 111204637–111204638 was decreased by approximately 73% (Supplementary Fig. 6a). The mean percentage point of these CpGs between Y287X-FLI1 and Y287X-mock HPCs (percent methylation score of Y287X-FLI1 HPCs minus that of Y287X-mock HPCs) was significantly lower than that of the same number of randomly selected CpG sites (mean = -2.03 and − 0.228 [FLI1-mock and random, respectively]; P = 0.0006, Welch’s t-test; Fig. 7b; Supplementary Fig. 6b). Thus, FLI1 overexpression improved not only the Mk differentiation efficiency, but also the differential hypermethylation in RUNX1WT/Y287X HPCs.