Vitamin C–dependent lysine demethylase 6 (KDM6)-mediated demethylation promotes a chromatin state that supports the endothelial-to-hematopoietic transition

Hematopoietic stem cells (HSCs)/progenitor cells (HPCs) are generated from hemogenic endothelial cells (HECs) during the endothelial-to-hematopoietic transition (EHT); however, the underlying mechanism remains poorly understood. Here, using an array of approaches, including CRSPR/Cas9 gene knockouts, RNA-Seq, ChIP-Seq, ATAC-Seq etc., we report that vitamin C (Vc) is essential in HPC generation during human pluripotent stem cell (hPSC) differentiation in defined culture conditions. Mechanistically, we found that the endothelial cells generated in the absence of Vc fail to undergo the EHT because of an apparent failure in opening up genomic loci essential for hematopoiesis. Under Vc deficiency, these loci exhibited abnormal accumulation of histone H3 trimethylation at Lys-27 (H3K27me3), a repressive histone modification that arose because of lower activities of demethylases that target H3K27me3. Consistently, deletion of the two H3K27me3 demethylases, Jumonji domain-containing 3 (JMJD3 or KDM6B) and histone demethylase UTX (UTX or KDM6A), impaired HPC generation even in the presence of Vc. Furthermore, we noted that Vc and jmjd3 are also important for HSC generation during zebrafish development. Together, our findings reveal an essential role for Vc in the EHT for hematopoiesis, and identify KDM6-mediated chromatin demethylation as an important regulatory mechanism in hematopoietic cell differentiation.

terms of efficient generation from human pluripotent stem cells (hPSCs) such as hESCs (human embryonic stem cells) or hiPSCs (human pluripotent stem cells) (1)(2)(3)(4). One possible reason is that in vitro differentiation failed to fully recapitulate the developmental principles of hematopoiesis in vivo. Therefore, understanding the molecular mechanisms underlying hematopoiesis both in vitro and in vivo would be important to promote HSPC generation in vitro.
During mouse hematopoiesis, the first hematopoietic program, primitive hematopoiesis emerges in the yolk sac at embryonic (E) day 7.5, whereas the definitive program occurs later at around E10.5 at the aorta-gonad-mesonephros (AGM) region (5). HSPCs derived at the yolk sac or AGM display distinct functional characteristics in terms of engraftment in irradiated neonatal and adult mouse recipients (6). At the cellular level, hematopoiesis involves sequential lineage specifications from mesoderm, hematopoietic, and vascular fate to HSPCs, and then different subtypes of blood cells. Insights from different model organisms demonstrate that HSCs arise from a special type of endothelial cells, termed hemogenic endothelial cells (HECs) through endothelial-to-hematopoietic transition (EHT) (7)(8)(9). Several signaling pathways are known to regulate EHT in the mouse model, such as retinoic acid, Notch, and transforming growth factor-␤ signaling (10 -12). The HEC population is a very special subset of vascular endothelium that holds hematopoietic potential to give rise to multiple lineage HSPCs within a short developmental window. HECs isolated from mouse embryos could produce functional HSPCs in vitro (13), highlighting its essential function in HSPC generation during development. However, despite its essential role in hematopoiesis, the molecular events that specify functional HECs and the subsequent EHT remains largely unknown, particular in human background.
Here in this study, we discovered that vitamin C (Vc) is required for the generation of HPCs from hPSCs through regulating EHT. Mechanistically, Vc plays an essential role to specify a permissive chromatin state that allow endothelial cells to give rise to HPCs. Moreover, Vc is also important for HPC generation during zebrafish development. These findings reveal a previously unidentified but essential role of Vc dependent epigenetic mechanism underlying EHT during hematopoietic development.

Vitamin C is required for generation of HPCs from hPSCs in a defined condition
We sought to develop an efficient approach to differentiate blood cells from hPSCs in a chemically defined, serum-free and monolayer condition. At the early stage of embryogenesis, blood lineages were originally developed from primitive streak (PS) and the downstream lateral mesoderm (LM) (14). Loh et al. (14) reported that Wnt inhibition and BMP activation promote LM specification in the hPSC-derived PS population. Based on this report and other literatures (4,14,15), we developed a stepwise strategy to differentiate hPSCs in a defined, monolayer condition that recapitulates main stages of early hematopoiesis, including the PS, LM, HECs, and then HPCs (Fig. 1A). Through combined Activin/GSK3b inhibition/PI3K inhibition/FGF treatment for 1 day, nearly 100% of H1 hESCs differentiated into the Brachyury (T) positive PS mesoderm in monolayer conditions (Fig. 1B). Then, Wnt inhibition and BMP activation treated the following day promoted LM specification, whereas suppressing the paraxial mesoderm cell fate, as demonstrated by the activation of HAND1, the LM marker and suppression of MSGN1, the paraxial mesoderm marker (Fig. 1B) (15). Later at day 4, the cells were switched into the typical EHT medium containing hematopoietic cytokines to allow further hematopoiesis. A significant number of CD43 ϩ HPCs(1) could be detected after 2 days of culture in EHT medium (Fig. 1C). Morphologically, a typical EHT process could be observed at this stage, i.e. the endothelial cells acquired the hematopoietic morphology and became floated during culture from day 4 to 8 ( Fig.  1D). At day 8, over 90% of floating cells were CD43 ϩ (Fig. 1E), the typical phenotype of hPSC-derived HPCs(1) and they could form various blood CFUs (cfu) (Fig. 1F and Fig. S1A). Later at day 10, a certain percentage of CD43 ϩ cells became CD45 ϩ , a marker indicating mature HPCs (4, 16) (Fig. 1E). Furthermore, both adult and embryonic globin HBB, HBE, and HBG1 could be detected in the erythroid CFUs (CFU-E) (Fig. 1, H and I), demonstrating both the definitive and primitive hematopoiesis occurred during differentiation. Together, we developed a monolayer approach to differentiate hPSCs into hematopoietic cells in a chemically defined condition.
To further characterize the role of the individual factor in the basal medium in HPC generation, we surprisingly found that vitamin C was essential for HPC generation. In defined conditions with no Vc, the generation of CD43 ϩ HPCs, but not the pan-endothelial cells were significantly reduced ( Fig. 1, J and K, Fig. S1B). Consistently, the endothelial cells express both CD31 and CD43, the typical phenotype indicating EHT were significantly reduced in conditions without Vc (Fig. 1K). Quantitatively, the percentage of CD43 ϩ HPCs as well as their hematopoietic potential to form CFUs were significantly reduced in the Vc Ϫ condition (Fig. 1, L and N). The role of Vc in HPC generation was also verified on different hPSC lines, including multiple iPSC lines that we generated previously from different backgrounds, such as UH10, UC1, UC5, UH1, and HU-AL hiPSC (17) (Fig. S1, C-F). Notably, we found that increasing the concentration of Vc showed minor further enhancement on HPC generation, suggesting the function of Vc is not dose dependent in this case (Fig. S1G). Also, we found that only Vc, but not the other antioxidants could promote the HPC generation (Fig. 1N, Fig. S1H). Nevertheless, these data above suggest that Vc is an essential factor for hematopoietic differentiation in hPSCs in a defined condition.

Vitamin C is required to specify functional HECs
We then examined the role of Vc in more detail in the time course of blood differentiation of hPSCs. At the early stage, the presence of Vc; Vc Ϫ _G2ECs, the G2ECs generated in the absence of Vc; ATAC-seq, transposase-accessible chromatin with sequencing; CHIP-seq, chromatin immunoprecipitation followed by sequencing; TSS, transcriptional start sites; H3, histone 3; CFU-E, erythroid CFUs; BM, basic medium; MO, morpholinos; bFGF, basic fibroblast growth factor; sgRNA, single guide RNA; qPCR, quantitative PCR; DAPI, 4Ј,6-diamidino-2-phenylindole; BMP, basic metabolic panel.

Vc promotes a chromatin state that supports EHT
T-positive PS population showed little difference between Vc ϩ and Vc Ϫ conditions ( Fig. 2A). To analyze the role of Vc at the HEC stage, we employed the previously generated GATA2/GFP reporter hESC line (18). Consistent to our previous findings, GATA2 expression discriminated hemogenic potential endothelial cells from the nonhematopoietic endothelial cells in hESC differentiation (Fig. S2, A and B). In addition, the GATA2 ϩ endothelial cells (G2ECs) almost do not express the nonhematopoietic endothelial cell marker, CD73 (16) (Fig. S2C). At day 4 of differentiation, the generation of CD31 ϩ CD43 Ϫ GATA2/eGFP ϩ endothelial cells (G2ECs) that contain HECs did not show much difference between Vc ϩ and Vc Ϫ conditions (Fig. 2, A and B, Fig. S2D). However, following 2 days of further differentiation, the CD43 ϩ HPCs were significantly reduced at day 6 in the Vc Ϫ condition ( Fig. 2A). Notably, some key hematopoietic genes such as RUNX1, MYB etc. started to show differences in expression at day 4 between two conditions (Fig. 2C). To precisely examine the role of Vc in HPC generation at the EHT stage, we sorted the G2ECs generated in the Vc ϩ or Vc Ϫ conditions at day 4 and re-plated them again in EHT medium either containing Vc or no Vc (Fig. 2D). Interestingly, the G2ECs generated in the presence of Vc (Vc ϩ _G2ECs) successfully produced HPCs with efficient CFU potential ( Fig.  S2E) in EHT conditions no matter with or without Vc (Fig. 2D). In contrast, the G2ECs generated in the absence of Vc (Vc Ϫ _G2ECs) failed to efficiently produce HPCs even re-plated in EHT medium containing Vc (Fig. 2D). Together, these data suggest that Vc is mainly required for the specification of endothelial cells that have hemogenic potential for HPC generation.

Endothelial cells generated in the absence of Vc are incompetent to undergo EHT
To investigate the molecular mechanism underlying the functional defect in Vc Ϫ _G2ECs, we generated RNA-seq data in the time course of the subsequent EHT process. Surprisingly, Vc Ϫ _G2ECs and Vc ϩ _G2ECs sorted at day 4 of hPSC differentiation before EHT showed a much more similar profile in the whole genome expression based on RNA-seq data (Fig. 3A). Selected endothelial or blood-related genes showed much more similar expressions between two populations (Fig. 3B). However, when re-plated in the EHT condition, they showed distinct differentiation pathways based on principal component analysis (Fig. 3C). Although the EHT of Vc ϩ _G2ECs progressed well toward HPCs, Vc Ϫ _G2ECs did not progress well toward HPCs (Fig. 3C). The up-or down-regulated genes during EHT clearly showed distinct patterns between Vc Ϫ _G2ECs and Vc ϩ _G2ECs (Fig. 3, D and E). The endothelial related genes, such as CDH5, JAG1, and TEK were significantly down-regulated in Vc ϩ _G2ECs, but remained highly expressed in Vc Ϫ _G2ECs (Fig. 3, E and F). The hematopoietic genes, like RUNX1, MYB, and IKZF1 were significantly up-regulated in Vc ϩ _G2ECs, but repressed in Vc Ϫ _G2ECs during the EHT process (Fig. 3, E and G). Together, these data demonstrate that Vc Ϫ _G2ECs are incompetent to undergo EHT, despite showing similar gene expression profiles as Vc ϩ _G2ECs.

Vc promotes chromatin accessibility on hematopoietic genes
Ours and other reports (19 -21) showed that Vc regulates somatic cell reprogramming through histone and DNA demethylases. To see if similar mechanisms also regulate the generation of HECs, we examined the chromatin accessibility landscapes and epigenetic modifications in Vc Ϫ _G2ECs and Vc ϩ _G2ECs by transposase-accessible chromatin with sequencing (ATAC-seq) and ChIP followed by sequencing (CHIP-seq) on active (H3K4me3) and repressive (H3K27me3) histone modifications. Broadly, all ATAC-seq samples exhibited the expected periodicity of insert length (Fig. S3A) and the ATAC-seq peaks were enriched in transcriptional start sites (TSS) (Fig. S3B). Despite similar distributions of ATAC-seq peaks in genomic regions (Fig. 4A), a significant number of differential accessible chromatin regions were detected between Vc Ϫ _G2ECs and Vc ϩ _G2ECs (Fig. 4, B and C). Regions with increased accessibility in Vc ϩ _G2ECs were enriched for genes associated with hematopoietic function (Fig.  4D). Conversely, chromatin regions enriched in motifs of known hematopoietic regulators were more assessable in Vc ϩ _G2ECs than that in Vc Ϫ _G2ECs (Fig. 4E, Fig. S3C). These data indicate that Vc promotes chromatin accessibility on hematopoietic genes. Consistently, the global level of the repressive histone marker, H3K27me3, was significantly reduced in Vc ϩ _G2ECs (Fig. 4F). Although the active genes and neither of the genes were similar, the poised bivalent genes (H3K4me3/K3K27me3) (22) showed a big difference between Vc Ϫ _G2ECs and Vc ϩ _G2ECs (Fig. 4G, Fig. S4, A and B). A significant number of poised genes in Vc ϩ _G2ECs became more repressed in Vc Ϫ _G2ECs due to failure to remove H3K27me3 at those loci. These genes include many known important genes in hematopoiesis, for example, MYB and IKZF1 (Fig. 4, H-J). Conversely, some bivalent genes in Vc ϩ _G2ECs showed reduced H3K27me3 in Vc Ϫ _G2ECs, but they are not reported to involve in hematopoiesis (Fig. 4, H-J). Together, these data demonstrate that Vc promotes the hema-

Vc promotes a chromatin state that supports EHT
topoietic potential in endothelial cells by opening genomic loci of hematopoietic regulators.

Vc regulates hematopoietic differentiation of hPSC through KDM6s and TET1
The data shown above indicate that H3K27me3 might be a downstream repressor regulated by Vc in HPC generation. H3K27me3 modification on chromatin could be removed by its demethylase, KDM6s, which contain two members, UTX (KDM6A) and JMJD3 (KDM6B) (23). Notably, Vc has been reported to regulate KDM6 demethylase activity in the dopamine neuron (24). We then examined the global level of H3K27me3 in ECs generated with or without Vc. Consistent to the CHIP-seq data in Fig. 4F, Vc ϩ _G2ECs showed a reduced level of global H3K27me3 compared with Vc Ϫ _G2ECs (Fig.  5A). Moreover, based on a previously reported assay for KDM6 demethylase activity (24), Vc ϩ _G2ECs also showed a much higher activity of KDM6 demethylase (Fig. 5B). These data indicate that KDM6s might be a critical downstream effector of Vc during hematopoietic differentiation of hPSCs. To further examine the direct role of KDM6s, we deleted the JMJC domain, the critical domain for demethylase activity in UTX and/or JMJD3 in hESCs (Fig. 5C, Fig. S5, A-D) (25). Upon differentiation, hESCs with deletion of either UTX (UTX Y/Ϫ ) or JMJD3 (JMJD3 Ϫ/Ϫ ), or both (dKO) showed significant defects in HPC generation, but not endothelial cells in the presence of Vc (Fig. 5, D-F, Fig. S5E). These data indicate that JMJD3 and/or UTX are required for normal EHT and not functionally redundant in this process. We then analyzed the differentiation of these KDM6 mutant cells in more detail. First, a pluripotency marker, such as OCT4, was down-regulated along the differentiation in all the examined cell lines (Fig. 5G), indicating the exit of pluripotency was generally not impeded in KDM6-deficient cells. The PS marker T, the LM marker HAND1, and GATA2 were all successfully up-regulated at the early stage of differentiation in all KDM6-deficient hESC lines (Fig. 5G). In contrast, the hematopoietic genes, such as MYB, IKZF1, LMO2, and TAL1 failed to up-regulate at later stage (Fig. 5G). These data suggest that the KDM6s mainly act at a later stage of blood differentiation to ensure a normal EHT. Accordingly, H3K27me3 were more enriched on the selected hematopoietic genes in KDM6-deficient cells compared with WT hESCs at the later stage of blood differentiation (Fig. 5H). In contrast, CD31 ϩ CD43 Ϫ endothelial cells showed no big difference between all cell lines examined (Fig. 5F). Also, deletion of JMJD3 did not affect differentiation of the GATA2 ϩ ECs (Fig.  S5, F and G). Together, these data demonstrate that KDM6mediated H3K27me3 demethylation is required for HPC generation at the later stage of differentiation in hPSCs.
It is noteworthy that Vc was also reported to suppress leukemogenesis through TET family enzymes, the dioxygenases that catalyze 5mC to 5hmC on DNA (20,27,28). To determine whether TET-dependent DNA methylation is also involved in the regulation of Vc on HPC generation, we performed deletions of TET1 in human ESCs and examined their blood differentiation potential (Fig. 5, I and J). Indeed, HPC, but not the EC generation was reduced in TET1-deficient hESCs in the presence of Vc (Fig. 5, I and J). On the other hand, Vc showed a mild enhancement on HPC generation in the absence of TET1. These data indicate TET-mediated DNA methylation might be another downstream mechanism of Vc in HPC generation, apart from the KDM6-mediated H3K27me3 demethylation.

Vitamin C and jmjd3 are important for HSC emergence in zebrafish
To examine the role of Vc to promote hematopoiesis in vivo, we employed zebrafish as a development model because zebrafish are one of those species that cannot synthesize Vc by themselves (29). Based on a previous report (30), we fed the flk1:mCherry/cmyb:GFP (31) transgenic zebrafish a defined diet with or without Vc (Fig. 6A). After continuous feeding with Vc Ϫ diets for over 2 months, the total Vc concentration in the D5.5 zebrafish embryos was significantly reduced (Fig. 6B). We then analyzed blood development in embryos developed in Vc ϩ or Vc Ϫ condition. The zebrafish embryos developed in the Vc Ϫ condition showed a significantly reduced generation of flk1 ϩ cmyb ϩ cells (emerging HSCs (32)) in the AGM region compared with the Vc ϩ condition and controls (Fig.  6C), indicating that Vc plays an important role in blood development in zebrafish. Last, we also examined the role of jmjd3 in HSC commitment during zebrafish development with normal Vc. We injected morpholinos (MO) to knock down jmjd3 in flk1:mCherry/cmyb:GFP transgenic zebrafish (31). The flk1 ϩ cmyb ϩ emerging HSCs were remarkably reduced in jmjd3 morphants (Fig. 6D) in the AGM region compared with controls, suggesting jmjd3 is also important for blood development in zebrafish.

Discussion
The hematopoietic differentiation of hPSCs provides a good model to investigate human hematopoiesis, but is largely inefficient and variable (33,34). Based on previously published literature (4, 14, 15), we developed a stepwise blood differentiation strategy in hPSCs, which recapitulates the main stages of early hematopoiesis, including the PS, LM, HECs, and then HPCs in a defined, monolayer condition. This monolayerbased approach exhibits reduced batch to batch viabilities and much higher efficiencies compared to previously reported protocols based on co-culture with OP9 stromal cells (35) (Fig.  S6A). Even though HPCs generated in this condition still showed limited engraftment in vivo (data not shown), it pro-

Vc promotes a chromatin state that supports EHT
vides a simple and efficient model to study human hematopoiesis in a totally defined and monolayer condition.
Based on this approach, we discovered that Vc is an essential factor in generation of HPCs, particularly at later stages of differentiation in specification of functional HECs (Fig. 1). Interestingly, Vc would no longer be required for HPC generation after the HEC specification stage. We further revealed that endothelial cells generated in the presence or absence of Vc harbor a distinct epigenetic state. Specifically, endothelial cells generated in the Vc Ϫ condition exhibited a high level of H3K27me3 and less DNA accessibility on hematopoietic genes. In previous reports, the methyl groups on H3K27 was shown to be removed by the JMJC domain-containing protein KDM6s (KDM6A/UTX and KDM6B/JMJD3) (36). Interestingly, Vc was shown to be crucial for the optimized demethylation activity of the JMJC domain (25). Consistent with these findings, KDM6s show much reduced demethylase activity in the Vc Ϫ condition. Further impairment of KDM6s in hESCs impeded EHT in the presence of Vc. More importantly, Vc and jmjd3 are also important to regulate the emergence of hematopoietic cells in zebrafish development. Therefore, our data reveal a previously unknown, but essential role of the Vc-dependent epigenetic mechanism in hematopoiesis, both in the human model and zebrafish development.
Vc is also a well-known antioxidant, however, other antioxidants did not rescue the HPC defect in hPSCs in the Vc Ϫ condition (Fig. 1N, Fig. S1H). Also, the presence or absence of Vc does not have a significant impact on expression of the Vc transporter genes, SVCT1/2 in the G2ECs (Fig. S6B). These data indicate that the major role of Vc to promote HPC generation is mainly through epigenetic regulations. The epigenetic mechanisms, such as active or repressive histone modifications of H3K4me3/H3K27me3 have been shown to be important in different types of stem cell models, including the hematopoietic cells (37,38). Recently, EZH1, a component of PRC2 that catalyze H3K27me3 was reported to play a critical role in restricting multipotency of primitive hematopoiesis (39). However, at the earlier stage of hematopoietic development, little is known about how the functional HECs are specified and what molecular events drive them into HSPCs. Our findings here reveal

Vc promotes a chromatin state that supports EHT
that Vc-dependent KDM6 specifies a competent epigenetic state in endothelial cells with EHT potential, further highlighting the critical role of Vc-dependent epigenetics at the early stage of hematopoiesis.
Recently, Agathocleous et al. (20) reported that Slc23a2 (Svct2, the key Vc transporter) knockout mice showed mild effects on embryonic HSCs frequency in the fetal liver at E17.5. However, ascorbate-depleted mice showed a higher level of HSC frequency in spleen and bone marrow, indicating that depletion of Vc might result in HSC expansion at a later stage. Consistently, Vc was reported to suppress leukemogenesis through TET family enzymes, the dioxygenases that catalyze 5mC to 5hmC on DNA (20,27,28). Apparently, DNA methylation change mediated by TETs is also an important epigenetic mechanism to regulate DNA accessibility. We also proved that TET-mediated DNA methylation might be another downstream mechanism of Vc in HPC generation (Fig. 5, I and J). Besides, Vc was also reported to be an important factor to regulate other stem cells and in reprograming (19,21), highlighting its essential role in cell fate decision. Thus, our findings extended the current understanding of the role of Vc to regulate cell fate decisions during both development and reprogramming.
To delete the JMJC domain of the KDM6 gene, JMJD3 for instance, we further constructed a targeting vector, which contains two homology arms cloned from the genomic DNA of the H1 and H1-GATA2 w/eGFP cell line about 1 kb in the upstream and downstream of the JMJC domain, respectively. A loxP-flanked PGKpuromycin cassette was further cloned into the two homology arms in the Puc57 vector. For gene targeting, we electroporated 4 g of the targeting vector linearized by EcoRI and 2 g of CRISPR/ cas9-sgRNA into H1 hESCs. Then the cells were cultured in medium with the addition of Y-27632 (10 M, Sigma). After about 4 days, the hESC clones were selected by puromycin (0.5 g/ml, Sigma). The positive clones were further analyzed by PCR to select the homozygous mutation. The same strategy was used to delete the JMJC domain of the UTX in H1 to generate the H1-UTX Ϫ/Ϫ . For H1-JMJD3 Ϫ/Ϫ 2# cell lines, we electroporated 400 ng of Cre recombinase into the H1-JMJD3 Ϫ/Ϫ cells to remove the loxpflanked PGK-puromycin cassette, followed by seeding in a single cell state in the presence of Y-27632. About 7 days later, the clones were picked and further analyzed by PCR to select the homozygous mutation. For H1-UTX Y/Ϫ /JMJD3 Ϫ/Ϫ (dKO) cell lines, the same strategy was used to delete the JMJC domain of the UTX in H1-JMJD3 Ϫ/Ϫ 2#. The homozygous mutated clones failed to express JMJD3 and/or UTX in the mRNA level (Fig. S5, C and F).

Maintenance and differentiation of hPSCs
All hPSC cell lines were maintained in mTeSR1 medium (Stem Cell Technologies) on Matrigel-coated plates (1:100 dilution; BD Bioscience). Prior to differentiation, the ϳ80% confluent hPSCs were dissociated by Accutase (Sigma) and plated on growth factor reduced Matrigel-coated 12-well plates (1:100 dilution; BD) at the initial density of 0.7 to 1.5 ϫ 10 5 /well. Particularly, thiazovivin (0.1 M, Selleck) was added in the culture medium to inhibit hPSCs apoptosis. After overnight culture (designated as day 0), the confluence of the plated hPSCs should be around 10%. Then, the hPSCs were induced for stepwise differentiation as described in Fig. 1A. First, at day 0 -1 of differentiation, 40 ng/ml of BMP4 (Peprotech), 30 ng/ml of ACTIVIN A (Sino Biological Inc.), 20 ng/ml of bFGF (Sino Biological Inc.), 6 M CHIR99021 (Selleck), and 10 M LY294002 (Selleck) were added to the basic medium (BM, mimics of the Custom mTeSR1 (40)) of Dulbecco's modified Eagle's medium/ F-12 (GIBCO) supplemented with 1% insulin-transferrin-selenium (GIBCO), 70 mg/ml of vitamin C (Vc, 2-phospho-L-ascorbic acid trisodium salt solution, Sigma). In particular, the osmotic pressure of the BM was about 340, adjusted by 9% NaCl. Second, 30 ng/ml of BMP, 1 M A8301 (Selleck), and 2 M IWR-1-endo (Selleck) were added in the BM at days 1-2 of differentiation. Then, 40 ng/ml of vascular endothelial growth factor (Sino Biological Inc.) and 50 ng/ml of bFGF were in the BM at day 2-4 of differentiation. Finally, 40 ng/ml of vascular endothelial growth factor, 50 ng/ml of bFGF, 10 M SB431542 (Selleck), 10 ng/ml of stem cell factor (Peprotech), 50 ng/ml of thrombopoietin (Sino Biological Inc.), 10 ng/ml of interleukin 3 (Sino Biological Inc.), and 50 ng/ml of interleukin 6 (Sino Biological Inc.) were added in the BM at days 4 -6 of differentiation and further hematopoietic commitment and maturation. The hematopoietic differentiation medium in each step should be changed every day, as the cells expanded very quickly in the system. The other OP9 co-culture and embryoid body methods for hematopoietic differentiation of hPSCs were performed according to a previous publication (35). The hPSCs and differentiating cells were maintained and differentiated in standard conditions (37°C, 5% CO 2 , over 95% humidity).

Real-time quantitative PCR
The total RNA were extracted by the RaPure Total RNA Micro Kit (Magen), and 2 g of RNA were reverse transcribed into cDNA according to the manufacturer's instructions (Takara). RT-qPCR were performed with SYBR Green Master Mix (Bio-Rad). Glyceraldehyde-3-phosphate dehydrogenase were used for normalization. All primers used in this study were listed in the Table S1.

Immunofluorescence
The cells were fixed by 4% paraformaldehyde for 30 min and incubated with the CD31 nonconjugated antibody and then stained with Alexa Fluor 568. After that, the cells were further stained with CD43-FITC antibody and followed by DAPI staining. The antibodies used in this study were listed in the Table S2.

Western blotting
Western blotting was performed as previously described (41). Briefly, the H3 (histone 3) was used as the control and the reference for quantification of the H3K27me3 modification.

Measurement of the KDM6 demethylase activity
The demethylase activity of KDM6s were measured by the JMJD3/UTX demethylase activity assay colorimetric kit (Epigentek). Briefly, nuclear factions were obtained using the NE-PER nuclear and cytoplasmic extraction kit (Thermo), and 5 g of nuclear extracts were used for the demethylase activity detection.

Flow cytometry and FACS
For cell surface staining, cells were dissociated by Accutase (Sigma) and prepared in PBS supplemented with 2% FBS and labeled with multicolor antibody combinations, incubated at 4°C for 15 to 30 min. For intracellular staining, cells were fixed by fixation buffer (BD Biosciences) and then permeabilized using permeabilization solution (BD Biosciences) before staining. For cell sorting, DAPI were used to exclude the dead cells. For flow cytometry, samples were analyzed by C6 or Fortessa (BD Biosciences); for cell sorting, cells were sorted by the Aria (BD Bioscience) or Moflo (Beckman). All the antibodies used in this study were listed in the Table S2.

CFU assay
The CFU assay was conducted using the manufacturer's instructions for Methocult H4435 (Stem Cell Technologies). Briefly, single cells of the indicated number were suspended in the 100 l of Iscove's modified Dulbecco's medium supplemented with 2% FBS (Biological Industries), and then mixed with 1 ml of Methocult H4435. The mixture was transferred into 35-mm ultra-low attachment plates (Stem Cell Technologies). After 12 to 16 days, the CFUs were classified and counted according to the morphology. The CFU assay was performed in standard conditions (37°C, 5% CO 2 , over 95% humidity). In particular, the peripheral blood CD34 ϩ cells were isolated from the mobilized peripheral blood CD34 ϩ cells by MACS from the volunteer, which was approved by the IRB of the Third Affiliated Hospital, Sun Yat-sen University.

RNA-Seq
In brief, total RNA were isolated by Directzol RNA MiniPrep kit (Zymo Research), and sequencing libraries were prepared with a TruSeq RNA Sample Prep Kit (Illumina) under the manufacturer's instructions. The samples were run on a MiSeq system with MiSeq Reagent Kits version 2 (50 cycles) (Illumina).
All RNA-Seq data were processed as previously described (42,43). In brief, reads were aligned to an index generated from the Ensembl transcriptome version 74 (hg38), using RSEM (version 1.2.19), Bowtie2 (version 2.2.5), and normalized with EDASeq (version 2.2.0). Particularly, gene expression is expressed as "normalized tag count." A threshold of at least 20 normalized tags was used to filter lowly expressed transcripts. Differential expression was performed using DESeq2 (version 1.8.1) and genes were considered significant if they had a Benjamini-Hochberg corrected p value Ͻ 0.05. Gene ontology was performed using clusterProfiler.

ATAC-Seq
ATAC-Seq and data processing were performed according to previous reports (44 -46). Briefly, 50,000 cells of each sample were used to generate DNA libraries for sequencing using NextSeq 500. All sequencing data were mapped on to hg38 using bowtie2. Peaks were called using MACS2 and differential accessibility was assessed using DESeq2 as previously described (46). Regions were called differentially accessible if the absolute value of the log 2 (fold-change) was Ͼ0.5 at an FDR Ͻ0.1. All genome views are to the same vertical scale (0 -20).

ChIP-Seq
ChIP-Seq and data processing were performed as previously described (47,48). In brief, 5,000,000 cells of each sample were used to generate DNA libraries for sequencing by NextSeq 500. All sequencing data were mapped onto hg38 using bowtie2, and peaks were called using SICER. To identify the H3K4me3-and H3K27me3-associated regions and map the genes, we choose the gene region containing the peaks extending Ϯ2.5 kb from the TSS. In addition, intersectBED was used to identify the active (H3K4me3 only), repressed H3K27me3 only (K27), bivalent or poised (H3K4/27me3), and neither (H3K4/27me3 null) gene regions. Gene ontology was performed using DAVID.

ChIP-qPCR
ChIP-Seq and data processing were performed as previously described (13,26). DNA samples including the Input positive control and the IgG negative control were used for SYBR Green-base quantitative PCR analysis.

Zebrafish and MOs
The transgenic zebrafish lines flk1:mCherry and cmyb:eGFP (generously provided by Dr. Wenqing Zhang) were raised and maintained at 28.5°C. The zebrafish embryos were acquired by natural spawning, and further reared in an incubator at 28.5°C. This study was approved by the Ethical Review Committee of the Guangzhou Institutes of Biomedicine and Health, Chinese Academy of Sciences.
For jmjd3 knockdown in zebrafish, the zebrafish were fed with conventional diets, and the embryos were acquired by natural spawning. The jmjd3 translation blocking MO (5Ј-CCCA-TCTCGCTGTTACTGTGTTTTC-3Ј) was a gift from Dr. Yu Shanhe (Liu Tingxi's lab). The control MO (5Ј-CCTCTTACC-TCAGTTACAATTTATA-3Ј) was purchased from Gene Tools. All MOs were microinjected into the embryos at the 1-cell stage. Also, at 36 hpf the flk1 ϩ cmyb ϩ zebrafishes were harvested and the emerging HPCs were analyzed in the AGM region by confocal microscopy.

Measurement of Vc in diets and zebrafish
To measure the Vc concentration in the diets, the diets were dissolved in H 2 O, centrifuged, and the supernatants were collected for analysis. To detect the Vc concentration in the zebrafish, the fish about 5 days post-fertilization were euthanized by an overdose of tricaine, weighed, and homogenized by ultrasonication. The homogenates were then centrifuged, and the supernatants were collected for analysis. The Vc concentration was determined by the HPLC with electrochemical detection as previously described (29).

Statement
The CFU assay involved human peripheral blood were approved by IRB of the Third Affiliated Hospital, Sun Yat-sen University. Also this study abides by the Declaration of Helsinki principles.