Abstract

   Hematopoietic differentiation of human pluripotent stem cells (hPSCs)
   requires orchestration of dynamic cell and gene regulatory networks but
   often generates blood cells that lack natural function. Here, we
   performed extensive single-cell transcriptomic analyses to map fate
   choices and gene expression patterns during hematopoietic
   differentiation of hPSCs and showed that oxidative metabolism was
   dysregulated during in vitro directed differentiation. Applying hypoxic
   conditions at the stage of endothelial-to-hematopoietic transition in
   vitro effectively promoted the development of arterial specification
   programs that governed the generation of hematopoietic progenitor cells
   (HPCs) with functional T cell potential. Following engineered
   expression of the anti-CD19 chimeric antigen receptor, the T cells
   generated from arterial endothelium-primed HPCs inhibited tumor growth
   both in vitro and in vivo. Collectively, our study provides benchmark
   datasets as a resource to further understand the origins of human
   hematopoiesis and represents an advance in guiding in vitro generation
   of functional T cells for clinical applications.

INTRODUCTION

   Human pluripotent stem cells (hPSCs) offer a powerful resource to
   create in vitro models of human hematopoietic development and blood
   disease and have great potential for clinical cell replacement
   therapies. Although there are several reports of hematopoietic cells
   with phenotypes of hematopoietic stem cells (HSCs) or terminally
   differentiated hematopoietic cells generated from hPSCs ([52]1, [53]2),
   a robust method for the generation of these cells with consistent
   function has not yet been established. Thus, identifying key cellular
   and molecular programs required for hematopoietic specification in
   vitro is essential to overcome the current roadblocks.

   During embryonic development, hematopoietic progenitor cells (HPCs)
   arise from transient endothelial-like precursors known as hemogenic
   endothelium (HE), which are found initially in the yolk sac vascular
   plexus and, later, in various intraembryonic and extraembryonic vessels
   ([54]3, [55]4). The HE represents a distinct RUNX1-expressing subset of
   vascular endothelial cells (ECs) with the ability to undergo
   endothelial-to-hematopoietic transition (EHT) ([56]4). RUNX1 has been
   shown to activate the hematopoietic specification program and mediate
   the up-regulation of transcription factors (TFs) (e.g., GFI1 and
   GFI1B), which, in turn, repress endothelial gene expression ([57]5). In
   addition to RUNX1, many other factors have been found to regulate EHT,
   including TFs (e.g., GATA2 and MEIS2) and cell signaling pathways
   [e.g., the transforming growth factor–β (TGF-β) and Notch pathways]
   ([58]6–[59]9). More recently, the N-glycome and cell cycle regulation
   have been identified as drivers of EHT ([60]10, [61]11). Despite these
   discoveries, the mechanisms controlling EHT remain to be fully
   elucidated. A comprehensive understanding of these mechanisms is of
   critical importance in establishing strategies for guiding functional
   hematopoietic cell generation in vitro, especially in humans where the
   feasibility of EHT studies is limited by obvious ethical issues.

   Hematopoietic development is induced by signals from the
   microenvironment provided by anatomically distinct niches and
   tissue-specific vascular niches located in several hematopoietic sites,
   including the aorta-gonad-mesonephros (AGM), the fetal liver, and the
   placenta ([62]3, [63]12). In the AGM region of the mouse embryo, a
   variety of cellular niches—including ECs, mesenchymal cells (Mes), and
   mesonephros cells—have been shown to play distinct roles in HSC
   generation ([64]13–[65]15). In a recent study, single-cell RNA
   sequencing (scRNA-seq) technology was used to identify the cellular and
   molecular programs that underlie the generation of the first HSC from
   HE in human embryos ([66]16). However, the functional contribution of
   these niche cells to human hematopoiesis remains poorly defined.

   Single-cell transcriptional profiling has unique advantages in the
   identification of heterogeneous cell populations and has been
   successfully used to decipher the events in the development of
   hPSC-derived ECs, cardiomyocytes, and kidney organoid cells
   ([67]17–[68]19). In hematopoietic differentiation from hPSCs, HPCs and
   the EHT have also been analyzed at the single-cell level ([69]11,
   [70]20–[71]22). However, strategies for the generation of functional
   hematopoietic cells from hPSCs have not been identified. Here, by
   investigation of the entire process of pluripotent-to-hematopoietic
   transition, we have provided a complete transcriptome landscape of
   hPSC-derived hematopoiesis at the single-cell level. Through comparison
   of in vivo and in vitro data, we found that aerobic metabolism was
   increased during EHT in hPSCs, and exposure to hypoxia enhanced HPC
   generation with functional T cell potential from hPSCs via the
   induction of arterial niche signals.

RESULTS

scRNA-seq analysis of hematopoietic directed differentiation

   To gain insights into the genetic regulation of hematopoietic
   development, we performed single-cell transcriptional profiling of
   human embryonic stem cells (H1 hESCs) navigating from pluripotency
   through the stage-specific transitions of hematopoietic differentiation
   using 10× Genomics Chromium platform. A monolayer-based, chemically
   defined culture was modified as an efficient method to direct
   pluripotent cell differentiation toward the endothelial and
   hematopoietic (EC-HC) lineages as previously described ([72]23–[73]25).
   Mesoderm progenitors (T-GFP^+), CD34^+KDR^+CD144^+ ECs, and CD43^+ HPCs
   were identified at days 2, 4, and 6 of differentiation, respectively
   (fig. S1A), indicating that the period between days 4 and 6 is a key
   stage in EHT. For scRNA-seq analysis, we analyzed cells at time points
   corresponding to stage-specific transitions in the cell state including
   pluripotency (day 0), mesoderm specification (day 2), and progression
   through EC (day 4) and HPC (day 6) ([74]Fig. 1A). A total of 64,021
   cells were captured of which 37,666 cells were retained after quality
   control analysis.

Fig. 1. High-resolution dissection of hematopoietic differentiation of hPSCs
using scRNA-seq.

   [75]Fig. 1.
   [76]Open in a new tab

   (A) Schematic diagram of hematopoietic differentiation from H1 hESCs.
   Cells were collected on days 0, 2, 4, and 6 for scRNA-seq. Meso,
   mesoderm. (B) Identification of cell populations in the differentiation
   culture at days 0, 2, 4, and 6 visualized by uniform manifold
   approximation and projection (UMAP). Each dot represents one cell, and
   colors represent cell clusters as indicated. (C) UMAP visualization of
   the expression of typical feature genes for the identification of cell
   clusters. (D) Expression of genes in seven different categories of the
   indicated cell clusters. (E) Trajectory reconstruction of all single
   cells throughout hematopoietic differentiation (excluding day 0)
   reveals three branches: prebranch (before bifurcation), Mes lineage
   branch, and EC-HC lineage branch. (F) Gene expression heatmap of the
   top 3000 highly variable genes among cells used in the trajectory
   inference analysis presented in a pseudo-temporal order. (G) Expression
   dynamics of the top 3000 highly variable genes cataloged into four
   major clusters in a pseudo-temporal order shown as blue lines (Mes
   lineage) and red lines (EC-HC lineage). Thick lines indicate the
   average gene expression patterns in each cluster. Gene signatures and
   expression dynamics of representative genes in each gene cluster are
   shown. (H) Gene ontology (GO) analyses of each gene cluster identified
   in (F) and (G). ER, endoplasmic reticulum.

   Unsupervised clustering of the entire pooled dataset revealed seven
   transcriptionally distinct populations consisting of hPSC, mesodermal
   progenitor cell (MPC), day 4 Mes (D4-Mes), day 6 Mes (D6-Mes), day 4 EC
   (D4-EC), day 6 EC (D6-EC), and HPC clusters ([77]Fig. 1B and fig. S1B).
   The clusters of hPSC and MPC were identified by feature gene expression
   (NANOG and SOX2 for hPSCs and MIXL1 and MESP1 for MPCs) ([78]Fig. 1, C
   and D). Compared with D4-Mes, D6-Mes were more mature with higher
   expression of PDGFRA and COL1A1 ([79]Fig. 1, C and D). Compared with
   the D6-EC cluster, the D4-EC cluster not only showed endothelial
   features of CDH5 (CD144), PECAM1 (CD31), and FLT1 expression but also
   exhibited unique hematopoietic features of RUNX1 and GATA2 expression
   ([80]Fig. 1, C and D). The HPC cluster was characterized by the
   expression of SPN (CD43), SPI1 (PU.1), and GFI1B ([81]Fig. 1, C and D).
   SPI1 and GFI1B are transcriptional regulators that play crucial roles
   in the generation of functional HPCs from hPSCs ([82]2, [83]26).

   Next, we established the developmental trajectories by ordering cells
   in a pseudo-temporal manner for the lineage reconstruction of
   biological processes based on transcriptional similarity. First,
   analysis of cells from all stages revealed that the cells progressed
   sequentially from the hPSC to the MPC and Mes and, lastly, to the EC
   and HPC lineages (fig. S1C). Then, to improve the resolution and focus
   on the EHT process, we constructed a differentiation trajectory
   following removal of the hPSC and found that MPC-derived D4-Mes
   (prebranch) bifurcated into two diverse branches: one branch
   representing EC-HC lineages and the other branch representing the
   mesenchymal lineage (Mes lineage) ([84]Fig. 1E and fig. S1D).
   Furthermore, we compared the developmental trajectories in vitro and in
   vivo by reanalyzing the published scRNA-seq data from human embryos
   ([85]16) to construct the trajectory of in vivo developmental pathway.
   We found that the developmental trajectory in vivo was similar to that
   in vitro (fig. S1E). Overall, these findings suggest that a
   mesenchymal-like intermediate state is involved in the development of
   EC-HC lineage.

   To gain further insights into the gene expression dynamics of the Mes
   and EC-HC lineages, we analyzed the expression changes of the top 3000
   highly variable genes among cells collected at days 2, 4, and 6. We
   identified four major categories of transcriptional gene clusters in
   characteristic patterns ([86]Fig. 1, F to H, and fig. S1F). The genes
   in cluster 1, largely involved in the regulation of mesoderm (e.g.,
   CDX1 and MESP1), were gradually down-regulated during the
   differentiation ([87]Fig. 1, F to H). The genes in cluster 2, enriched
   in cell division and cell cycle (e.g., RPS28 and CCNB1), were
   transiently up-regulated and then down-regulated along both mesenchymal
   and EC-HC lineages, representing progenitor cells with active
   proliferative potential ([88]Fig. 1, F to H, and fig. S1F). The genes
   in cluster 3, which increased their expression along mesenchymal
   differentiation (e.g., IGF2 and COL1A1), were transiently up-regulated
   and then down-regulated during EC-HC development, suggesting a biphasic
   regulation of mesenchymal genes during hematopoiesis as previously
   reported ([89]Fig. 1, F to H, and fig. S1F) ([90]27). Last, the genes
   in cluster 4 were activated at the later stage of differentiation with
   predominant involved in EC-HC development (e.g., ESM1 and SPN)
   ([91]Fig. 1, F to H, and fig. S1F). These data depict the trajectory of
   hPSC-derived hematopoietic differentiation and reveal the ordered
   activation of transcriptional waves throughout this process.

Dissecting the heterologous cellular components involved in EHT

   To capture the precise cellular components involved in EHT, we
   extracted cell populations involved in the EC-HC lineage (including
   D4-EC, D6-EC, and HPC) for further analysis. By unsupervised
   clustering, the D4-EC population was further divided into three
   subclusters ([92]Fig. 2A). The smallest cluster was Mes-like with high
   expression of collagen genes and typical mesenchymal TFs (e.g., HAND1
   and SNAI2) ([93]Fig. 2, A, B, and E) and predominant enrichment in the
   term of mesenchyme development ([94]Fig. 2C). The mesenchymal
   subpopulation in D4-EC cluster also expressed low levels of
   endothelial- and hematopoiesis-related genes, such as CD34 and RUNX1
   ([95]Fig. 2E), with involvement in the gene ontology (GO) term “blood
   vessel morphogenesis” ([96]Fig. 2C). Therefore, this mesenchymal
   subpopulation represented a unique progenitor population with EC-HC
   potential, thus was named as endothelial and hematopoietic–biased Mes
   (EH-Mes). The largest subcluster in D4-EC cluster was endothelial
   progenitor cells (EPCs) given the main expression of endothelia genes
   (CD34 and KDR) and endothelial-regulated TFs [e.g., ID1 ([97]28)] and
   enrichment in blood vessel morphogenesis ([98]Fig. 2, A to C and E).
   The third subcluster showed high expression of HE-related TFs (e.g.,
   RUNX1 and GATA2) ([99]Fig. 2, B and E) together with endothelial
   features (CD34 and KDR) ([100]Fig. 2E). This subcluster was mainly
   involved in the GO term “myeloid cell differentiation” ([101]Fig. 2C)
   but with very low expression of hematopoietic genes (SPN) ([102]Fig.
   2E); therefore, this subpopulation was annotated as D4-HE.

Fig. 2. Heterologous cellular components and their interactions involved in
EHT.

   [103]Fig. 2.
   [104]Open in a new tab

   (A) UMAP visualization of EH-Mes, EPC, and D4-HE clusters based on
   subdivision of cells in the D4-EC cluster described in [105]Fig. 1B.
   (B) Heatmap showing the average expression of the top seven
   cluster-specific TFs among the EH-Mes, EPC, and D4-HE clusters. (C) The
   major GO Biological Process (GO BP) terms in which cluster-specific
   genes are enriched for EH-Mes, EPC, and D4-HE clusters. (D) UMAP
   visualization of AE, VE, D6-HE, and D6-HPC clusters based on
   subdivision of cells in the D6-EC and HPC clusters described in
   [106]Fig. 1B. (E) Violin plots showing the expression of feature genes
   in each cell cluster. (F) Heatmap showing the average expression of the
   top seven cluster-specific TFs among the AE, VE, D6-HE, and D6-HPC
   clusters. (G) The major GO:BP terms in which cluster-specific genes are
   enriched for D6-HE and D6-HPC clusters. (H) The major GO:BP terms in
   which cluster-specific genes are enriched for D4-HE and D6-HE clusters.
   ATP, adenosine triphosphate; SRP, signal recognition particle. (I)
   Schematic diagram showing the ligand-receptor interactions between HE
   or HPC and other niche clusters (AE, VE, and Mes). (J) Ligand-receptor
   interaction network showing the potential communications among niche
   cells (AE, VE, and Mes) and hematopoiesis-related cells (D6-HE and
   D6-HPC). (K) Heatmap showing the ligand-receptor pairs that exhibit
   different expression patterns among distinct niche clusters when
   coupled with D6-HE or D6-HPC.

   The previously identified HPC cluster shown in [107]Fig. 1B was further
   divided into two subclusters ([108]Fig. 2D). The smaller subcluster
   showed high expression of RUNX1 together with endothelial features, but
   with low expression of SPN, and was, therefore, annotated as D6-HE
   ([109]Fig. 2, D to G). The enrichment of GO terms in D6-HE, including
   translational initiation and mRNA processing ([110]Fig. 2H), was
   transcriptionally similar to that in the HE cluster identified in human
   embryos ([111]16). Compared with D4-HE, D6-HE was in a state of notable
   down-regulation of oxidative metabolism ([112]Fig. 2H). The other
   subcluster, annotated as D6-HPC, was high expression of hematopoietic
   genes (such as SPN and RUNX1) with low expression of endothelial genes
   and mainly enrichment in myeloid activation ([113]Fig. 2, D to G).

   The D6-EC cluster was also further divided into two subclusters
   ([114]Fig. 2D). Although both subclusters showed enrichment in
   angiogenesis and endothelial development (fig. S2A), one of the
   subclusters showed high expression of genes related to arterial
   endothelium (AE) (e.g., SOX17, GJA5, EFNB2, DLL4, HEY1, HEY2, and
   HES4), while the other subcluster showed high expression of genes
   related to venous endothelium (VE), such as NR2F2, EPHB4, NRP2, and
   APLNR ([115]Fig. 2, D to F, and fig. S2B). Thus, the two subclusters
   were identified as AE and VE, respectively.

Heterologous cellular interactions of hemogenic and hematopoietic cells with
niche cells

   To predict the cellular interactions involved in human hematopoietic
   development, we assessed the potential cell-cell interactions of D6-HE
   and D6-HPC with surrounding cellular niches (AE, VE, and D6-Mes) by
   ligand-receptor interaction analysis ([116]Fig. 2I). The interactions
   between hematopoietic-related cells and niche cells were shown to be
   complex and diverse ([117]Fig. 2J). When coupled with D6-HE or D6-HPC,
   the receptor-ligand gene pairs exhibited both similar and different
   expression patterns among the distinct niche populations (fig. S2C). To
   investigate the distinct roles of niche cells in hematopoietic
   development, we took the unique interactions of D6-HE and D6-HPC with
   niche cells into consideration ([118]Fig. 2K). Compared with VE or
   D6-Mes, the receptor-ligand interactions of either D6-HE or D6-HPC with
   AE were involved in NOTCH-related (e.g., JAG-NOTCH and DLL-NOTCH),
   cytokine-related (e.g., PDGFB-PDGFRB), cell adhesion–related (e.g.,
   JAM3-JAM3), and cytotoxin-related (e.g., TNFSF10-TNFRSF10B)
   receptor-ligand interactions ([119]Fig. 2K and fig. S2D). Kyoto
   Encyclopedia of Genes and Genomes pathway enrichment analysis showed
   that the interaction between AE and D6-HE was specifically enriched in
   the Notch signaling pathway (fig. S2E). The interactions of either
   D6-HE or D6-HPC with D6-Mes were mainly enriched in integrin-related,
   cytokine-related, TGF-β–related, and bone morphogenetic protein
   (BMP)–related receptor-ligand interactions ([120]Fig. 2K).
   Collectively, our data establish the critical interactions of HE and
   HPCs with cellular niches involved in human hematopoietic development.
   These interactions should be further investigated to identify potential
   molecular mechanism of niche cells in regulating hematopoietic
   differentiation.

Oxidative metabolism is dysregulated during in vitro directed differentiation
from hPSCs

   We next aimed to identify differences in the activation of signals that
   direct hematopoiesis in vivo and in vitro, which may facilitate the
   development of a robust method for the generation of functional
   hematopoietic cells from hPSCs. By comparing the subclusters identified
   in our study against scRNA-seq data from human embryos ([121]16), we
   found the overall gene expression patterns of cells differentiated from
   hPSCs in vitro were similar to those in vivo ([122]Fig. 3A and fig.
   S3A). To establish a strategy for the efficient production of
   functional hematopoietic cells from hPSCs, we further compared the HE
   and HPC generated in vitro and in vivo by GO enrichment analysis. Our
   GO enrichment analysis showed that hPSC-derived HE and HPC were more
   active in aerobic metabolism than those during embryonic development in
   vivo ([123]Fig. 3, B and C, and fig. S3B). Furthermore, we compared our
   in vitro HPC data with the published data of hematopoietic stem and
   progenitor cells (HSPCs) from the human fetal liver ([124]29) and bone
   marrow ([125]30) and hematopoietic cells from human AGM ([126]16). In
   vitro HPCs showed a predominant enrichment in oxidative and
   mitochondrial metabolism, compared with the fetal liver, bone marrow,
   and AGM (fig. S3, C to E). Collectively, these data demonstrate that
   the early hematopoietic development in vivo occurs under hypoxia.

Fig. 3. Down-regulation of oxygen metabolism at the stage of EHT.

   [127]Fig. 3.
   [128]Open in a new tab

   (A) Classification heatmap showing that the identified subclusters
   differentiated in vitro are similar to those in human embryos. The data
   of Carnegie stage (CS) 10/11 and CS13 cells from human embryos were
   reanalyzed. Hema, hematopoietic cells; Epi, epithelial cells. (B)
   Enriched GO terms in D6-HE and HE in vivo, respectively. (C) Enriched
   GO terms in D6-HPC and hematopoietic cells in vivo, respectively. (D)
   Volcano plot displaying the differentially expressed genes (DEGs)
   between cells of the EC-HC lineage on days 4 and 6. Representative
   respiratory chain (red) and endothelial-hematopoietic (blue) genes are
   indicated. Gray dots represent non-DEGs [approximately <1.6 fold change
   (FC)]. (E) GO analysis of down-regulated and up-regulated genes in the
   EC-HC lineage comparing day 6 with day 4. ncRNA, noncoding RNA; rRNA,
   ribosomal RNA. (F) The raw data of oxygen consumption rate (OCR) on day
   4 and day 6 H1-derived CD34^+ cells determined using a Seahorse XF
   analyzer in the presence of the mitochondrial inhibitors (1 μM
   oligomycin, 5 μM carbonyl cyanide p-trifluoromethoxyphenylhydrazone, or
   1.5 μM rotenone plus antimycin A). (G) Quantitation of basal,
   ATP-linked, and maximal OCR evaluated by two-way analysis of variance
   (ANOVA) (n = 5). (H) The raw data of glycolytic proton efflux on day 4
   and day 6 H1-derived CD34^+ cells following injection of rotenone plus
   antimycin A (0.5 μM) and 2-deoxy-d-glucose (50 mM); data are obtained
   using a Seahorse XF analyzer. (I) Quantitation of basal and
   compensatory glycolysis evaluated by two-way ANOVA (n = 5). Data are
   presented as means ± SEM. **P < 0.01 and ***P < 0.001.

   Considering that D6-HE are in a relatively hypoxic metabolic state
   compared with D4-HE ([129]Fig. 2H), we speculated that the oxygen level
   may be lowered during EHT. To further investigate the molecular changes
   during EHT, we analyzed the differentially expressed genes (DEGs)
   between cells in the EC-HC lineage on days 4 and 6. In comparison with
   the day 4 EC-HC lineage, the up-regulated genes in the day 6 EC-HC
   lineage were strongly enriched for EC-HC development, while the
   down-regulated genes involved in the day 6 EC-HC lineage were enriched
   for GO terms related to oxygen consumption, including typical
   respiratory chain genes, such as UQCRQ, NDUFC1, COX17, MRPL52, and
   TIMM8B ([130]Fig. 3, D and E). To confirm the metabolic status during
   EHT, we used a Seahorse XF analyzer to measure the oxygen consumption
   rate (OCR) of CD34^+ EC-HC cells on days 4 and 6. Compared with day 6
   CD34^+ cells, a significant increase in mitochondrial respiration was
   observed in day 4 CD34^+ cells ([131]Fig. 3, F and G). Using glycolytic
   rate assay to determine the glycolytic proton efflux rate of CD34^+
   cells on days 4 and 6, we observed a significant increase in glycolysis
   in day 6 CD34^+ cells ([132]Fig. 3, H and I). Collectively, these
   results suggest that, although hematopoietic differentiation in vitro
   occurs in a relatively aerobic metabolic state, down-regulation of
   oxygen metabolism is still required to drive EHT and generate HPCs from
   HE cells generated in vitro.

Hypoxia improves hematopoietic differentiation of hPSCs by activation of the
arterial specification program

   To investigate the role of hypoxia in hematopoietic differentiation
   from hPSCs, we cultured cells under conditions of hypoxia with 5% O[2]
   for 6 days of differentiation. Treatment of the differentiated H1 hESCs
   with hypoxia enhanced the formation of semiadhesive round cell clusters
   and increased the number of CD43^+ HPCs on day 6 ([133]Fig. 4, A and B,
   and fig. S4, A and B). The enhanced effect of hypoxia on hematopoietic
   differentiation was also confirmed in human induced pluripotent stem
   cells (iPSCs) (JHUi181 and BC1), which showed similar results to the H1
   hESCs ([134]Fig. 4B). Quantitative real-time polymerase chain reaction
   (qRT-PCR) analyses of the differentiated H1 hESCs showed that the
   expression of key genes associated with early hematopoiesis—including
   MYB, RUNX1, TAL1, and GATA2—was significantly increased under hypoxic
   conditions than that under normoxic conditions (fig. S4C). The
   colony-forming cell (CFC) assay of the differentiated H1 hESCs showed
   that hypoxia-induced hematopoietic differentiation gave rise to
   significantly higher numbers of hematopoietic colonies of myeloid and
   erythroid lineages than those generated under normoxia (fig. S4D),
   indicating that hypoxia enhanced the development of multipotent HPCs.
   To examine the hypoxia-dependent stage more precisely, we applied
   hypoxic conditions for defined intervals during the 6 days of
   hematopoietic differentiation of H1 hESCs ([135]Fig. 4C). As expected,
   applying hypoxia with 1% O[2] consistently for 6 days significantly
   increased the frequency of CD43^+ hematopoietic cells ([136]Fig. 4D).
   When hypoxia was applied between days 4 and 6, the stage coinciding
   with onset of EHT, the frequency of CD43^+ hematopoietic cells was
   significantly increased, compared with other 2-day interval groups
   ([137]Fig. 4D). Collectively, these findings demonstrate that hypoxia
   enhances hematopoietic differentiation of hPSCs at the stage of EHT.

Fig. 4. Hypoxia enhances hematopoietic differentiation by promoting the
arterial specification program.

   [138]Fig. 4.
   [139]Open in a new tab

   (A) Confocal microscopy visualization of emerging round hematopoietic
   cells on day 6 under the hypoxic (5% O[2]) or normoxic conditions.
   Scale bars, 40 μm. (B) The percentage of CD43^+ hematopoietic cells on
   day 6 under the hypoxic (5% O[2]) or normoxic conditions. (n = 3;
   two-way ANOVA). (C and D) Diagram and representative flow cytometric
   analysis demonstrating the five conditions used to test the importance
   of hypoxia exposure (1% O[2]) in promoting hematopoiesis. (n = 3). (E)
   Mean gene expression of all cells at each time point. (F) HIF-1a
   expression measured by Western blot at each time point under the
   normoxic or hypoxic conditions (1% O[2]). (G and H) Representative flow
   cytometric and statistical analyses of the dynamic changes in the
   arterial cell population between days 4 and 6 under hypoxic (1% O[2])
   or normoxic conditions. (n = 3; two-way ANOVA). (I) qRT-PCR analysis of
   the expression of DLL4 and NR2F2 from days 4 to 6 under the hypoxic (1%
   O[2]) or normoxic conditions. (n = 3; two-way ANOVA). (J)
   Representative flow cytometric analysis of the frequency of arterial
   cells in dimethyl sulfoxide (DMSO)– or U0126-treated day 5 cells. (n =
   3). (K and L) Flow cytometric analysis of the frequency and number of
   CD43^+ cells in DMSO- or U0126-treated day 6 cells. (n = 3; two-tailed
   Student’s t test). SSC, side scatter. For (J) to (L), U0126 or DMSO was
   added to the cultures from days 4 to 6 of differentiation. Experiments
   were performed on H1 unless otherwise indicated. Data are presented as
   means ± SEM. *P < 0.05 and ***P < 0.001.

   The vascular niche plays a key role in the specification and
   amplification of HSCs ([140]13, [141]26, [142]31). Further analysis of
   gene expression during EHT revealed that the expression of hypoxia gene
   HIF-1a and its related activated genes increased on day 4, peaked on
   day 5, and then decreased ([143]Fig. 4E and fig. S4E). The high HIF-1a
   expression in mRNA on day 0 may be due to the anaerobic metabolism of
   hPSCs ([144]32). The increased expression of HIF-1a and its related
   activated genes under normoxic and hypoxic conditions during EHT was
   further confirmed by Western blot and qRT-PCR, respectively ([145]Fig.
   4F and fig. S4F). Compared with HIF-1a expression, the
   endothelial-related genes showed similar expression patterns ([146]Fig.
   4E). Compared with venous genes (NR2F2 and EPHB4), arterial genes
   (SOX17 and EFNB2) were notably up-regulated on day 4 and peaked on day
   5, accompanied by a notable increase in NOTCH-related genes ([147]Fig.
   4E and fig. S4G). Therefore, we speculated that the effect of hypoxia
   on hematopoiesis is related to the activation of arterial program. To
   dynamically track the effect of hypoxia on the activation of arterial
   program during EHT, we assessed cell populations with the expression of
   CD73 and CD184, which have been used to distinguish arterial cells
   (CD34^+CD43^−CD73^+CD184^+ AE and CD34^+CD43^−CD73^−CD184^+ HE with
   arterial features) from CD34^+CD43^−CD73^+CD184^− VE ([148]33–[149]35).
   Compared with days 4 and 6, it was interesting to see that VE were
   decreased under hypoxic condition on day 5, whereas arterial cells,
   including AE and HE, were increased significantly on day 5 ([150]Fig.
   4, G and H, and fig. S4, H to J). Using qRT-PCR, we found that hypoxia
   effectively up-regulated the gene expression of DLL4, a key marker of
   arterial cells, and down-regulated the gene expression of NR2F2, a key
   marker of VE ([151]Fig. 4I). Compared to the CD34^+CD43^−CD73^−CD184^+
   HE cell population, the CD34^+CD43^−CD73^−CD184^− cell population has
   also been shown to have the ability to undergo EHT to generate HPCs but
   with limited T cell differentiation potential ([152]9, [153]33,
   [154]35). The modest decrease in the CD34^+CD43^−CD73^−CD184^− cell
   population and significant increase in the CD34^+CD43^−CD73^−CD184^+ HE
   cell population on day 5 under hypoxic conditions ([155]Fig. 4G and
   fig. S4K) suggested that hypoxia enhanced multipotent HPC generation.

   To further determine whether the effect of hypoxia on hematopoiesis is
   arterial program dependent, we applied a mitogen-activated protein
   kinase (MAPK) kinase 1 (MEK1) and MEK2 inhibitor, U0126, to inhibit
   MAPK/extracellular signal–regulated kinase (ERK) signaling pathway that
   was involved in the development of arterial program ([156]33, [157]35,
   [158]36). We found that inhibition of MEK/ERK between days 4 and 6
   impaired the development of arterial cells under hypoxia with 1% O[2]
   and led to a significant decrease in the proportion of CD43^+ HPCs
   ([159]Fig. 4, J to L, and fig. S4L). Overall, these observations
   demonstrate that hypoxia enhances hematopoietic differentiation of
   hPSCs by promoting the arterial specification program.

AE promote the generation of hematopoietic progenitors with T cell potential

   Since arterial program includes AE and arterial-type HE ([160]9,
   [161]35), it is still unclear whether hypoxia effect on hematopoietic
   differentiation is through cell-autonomous niches, cellular niches, or
   both. To further determine the roles of arterial cells and other
   cellular niches in hematopoiesis, AE (CD34^+CD43^−CD73^+CD184^+), VE
   (CD34^+CD43^−CD73^+CD184^−), HE (CD34^+CD43^−CD73^−CD184^+), or Mes
   (CD34^−CD31^−CD90^+CD105^+) were isolated from day 5 differentiated H1
   hESCs and induced EHT for 7 days (D5 + D7) with or without coculture
   ([162]Fig. 5A and fig. S5A). During EHT, AE, VE, or Mes alone was
   unable to generate CD43^+ and CD45^+ hematopoietic cells, indicating
   that the absence of hematopoietic potential in these populations (fig.
   S5B). Compared with the coculture of HE cells with VE (HE + VE) or Mes
   (HE + Mes), HE cells alone were unable to initiate EHT efficiently,
   whereas the coculture of HE cells with AE (HE + AE) for 7 days promoted
   EHT, significantly increasing the frequency and total numbers of CD43^+
   and CD45^+ HPCs ([163]Fig. 5, B and C). The AE niche effect on HPC
   generation from HE cells was also confirmed in BC1 human iPSCs, which
   showed similar results to the H1 hESCs (fig. S5C). Overall, these data
   demonstrate that hypoxia enhances hematopoiesis in two ways: (i)
   increasing arterial-type HE and (ii) promoting EHT by AE cellular
   niche.

Fig. 5. AE promote hematopoietic progenitor generation with T cell potential.

   [164]Fig. 5.
   [165]Open in a new tab

   (A) Schematic of protocol for investigating the niche effect on
   hematopoiesis. (B) Representative flow cytometric analysis of the
   hematopoietic progenitors obtained from H1-derived HE, or HE cocultured
   with Mes (HE + Mes), VE (HE + VE), or AE (HE + AE) following 7 days of
   EHT culture (n = 3). (C) The fold change in CD43^+CD45^+ cell number
   compared to HE. The cells were generated under the conditions of HE, HE
   + Mes, HE + VE, or HE + AE. (n = 3; one-way ANOVA). (D) CFC assay of
   HPCs generated with or without coculture with AE, VE, or Mes.
   Colony-forming units (CFUs) per 2000 cells plated. (n = 3; two-way
   ANOVA). (E) Schematic diagram of the protocol for T cell
   differentiation. MACS, magnetic-activated cell sorting. (F and G) T
   cell potential of hematopoietic progenitors generated from D5 HE
   following 7 days of culture in the presence or absence of VE, AE, or
   Mes. (n = 3; one-way ANOVA). (H) Representative flow cytometric
   analysis of the expression of CD4 and CD8 by CD3^+ cells before and
   after anti-CD3/CD28 stimulation (n = 3). (I) The proportion of CD107a
   in hPSC-derived T cells (hPSC-T) and peripheral blood T cells (PB-T)
   after treatment with phorbol 12-myristate 13-acetate (PMA)–ionomycin.
   (n = 3; two-way ANOVA). (J) Polyfunctional cytokine production by
   hPSC-T and PB-T after treatment with PMA-ionomycin. (n = 3; two-way
   ANOVA). Experiments were performed on H1 hESCs unless otherwise
   indicated. Data are presented as means ± SEM. *P < 0.05 and ***P <
   0.001.

   To determine the multilineage potential of HPCs, we harvested
   hematopoietic cells in suspension after EHT induction with or without
   coculture with niche cells (HE, HE + VE, HE + Mes, or HE + AE) and
   measured their colony forming potential. The total number of colonies
   was significantly increased in the HE + AE cultures, while there were
   no significant differences in the total number of colonies between the
   HE cultures and the HE + VE or HE + Mes cultures ([166]Fig. 5D). These
   results suggest that AE niche not only promotes EHT but also increases
   multilineage potential of emerged HPCs. To determine whether the effect
   of AE niche on hematopoiesis is Notch dependent, we applied the Notch
   inhibitor, DAPT, to inhibit the Notch signaling in AE. We found that
   inhibition of Notch in HE + AE cocultures led to a significant decrease
   in the proportion of CD45^+ cells (fig. S5D).

   T cell potential has been identified as a hallmark of multipotent HPCs
   ([167]9, [168]33, [169]35, [170]37). To further investigate the effect
   of cellular niches on the T cell potential of HPCs, we induced T cell
   differentiation from H1 hESCs using a previously reported artificial
   thymic organoid (ATO) model involving coculture of CD45^+ HPCs with
   OP9–human delta-like 1 (hDLL1) cells in three-dimensional (3D)
   aggregates for 8 weeks ([171]Fig. 5E) ([172]38, [173]39). We found that
   the isolated CD45^+ HPCs derived from AE cocultures (HE + AE) gave rise
   to increased CD3^+ T cell numbers, whereas the T cell potential of
   CD45^+ HPCs from VE or Mes cocultures, or without niche cells, was
   limited ([174]Fig. 5, F and G). The CD3^+ T cells exhibited a naïve
   phenotype with the expression of αβ T cell receptor (TCR-αβ), CD45RA,
   CD62L, CD28, and CD27 (fig. S5E). Before anti-CD3/CD28 treatment,
   CD8^+CD4^− cells and CD8^+CD4^+ cells were included among the CD3^+
   cells ([175]Fig. 5H). After anti-CD3/CD28 treatment, CD8^+CD4^− cells
   accounted for more than 90% of the CD3^+ cells ([176]Fig. 5H). To
   examine whether the T cells derived from AE-primed HPCs in vitro are
   functional, we stimulated the CD3^+ cells in the presence of the T cell
   activator phorbol 12-myristate 13-acetate (PMA)–ionomycin ([177]Fig.
   5E) and quantitated surface marker expression and cytokine production.
   After stimulation, CD25^+CD69^+ T cells significantly increased (fig.
   S5, F and G). Similar to peripheral blood (PB) T cells (PB-T), the
   hPSC-derived T cells (hPSC-T) generated from AE-primed HPCs were highly
   functional, releasing cytotoxic granules (CD107a expression) ([178]Fig.
   5I and fig. S5H) with polyfunctional production of interferon-γ
   (IFN-γ), tumor necrosis factor–α (TNF-α), and interleukin-2 (IL-2)
   ([179]Fig. 5J). Overall, these data demonstrate that AE promote the
   generation of HPC with functional T cell potential from HE cells.

AE-primed hPSC-T showed therapeutic potential following engineered expression
of anti–CD19 chimeric antigen receptor

   To further evaluate the function and the potential application of T
   cells derived from AE-primed HPCs, the T cells were engineered to
   express anti-CD19 chimeric antigen receptor (hPSC-CAR-T) for
   cytotoxicity assays both in vitro and in vivo ([180]Fig. 6A). The
   conventional CAR-T constructed from PB-isolated T cells (PB-CAR-T) were
   used as a positive control; hPSC-T and PB-T transfected with empty
   vectors (hPSC-VEC-T and PB-VEC-T) were used as negative controls. The
   CD19 CAR infection efficiency for hPSC-T was between 60 and 80%, which
   was similar to that of PB-T (fig. S6A). CD19^+ cell lines (Nalm-6 and
   Raji) and a CD19^− cell line (Molm13) were used as target cells (fig.
   S6B). hPSC-CAR-T and PB-CAR-T showed significant cytotoxic granule
   release (CD107a expression) when cocultured with Nalm-6 and Raji cells,
   while little CD107a up-regulation was observed following coculture with
   Molm13 cells ([181]Fig. 6B). After 24 hours of coculture,
   proinflammatory cytokines were measured by enzyme-linked immunosorbent
   assay (ELISA). The results showed that the secretion of IL-2, TNF-α,
   and IFN-γ was significantly increased in both hPSC-CAR-T and PB-CAR-T
   when cocultured with CD19^+ target cells ([182]Fig. 6C). By performing
   true lytic capability assays, we further confirmed in vitro that the
   hPSC-CAR-T killed CD19^+ leukemia cells as effectively as PB-CAR-T
   ([183]Fig. 6D).

Fig. 6. AE-primed hPSC-T showed therapeutic potential toward B-ALL following
engineered expression of anti–CD19-CAR.

   [184]Fig. 6.
   [185]Open in a new tab

   (A) Schematic diagram showing the protocol for generating hPSC-CAR-T
   for cytotoxicity assays both in vitro and in vivo. The AE used to prime
   T cells are generated under the hypoxic condition (1% O[2]). (B) The
   proportion of CD107a^+ cells in effector cells after coincubation with
   target cells (E:T = 1:1) for 5 hours. (n = 3; two-way ANOVA). (C) ELISA
   data showing the release of IFN-γ, IL-2, and TNF-α after coincubation
   with target cells for 24 hours. (n = 3; two-way ANOVA). (D) Direct
   lysis of effector cells against target cells. Flow cytometry analysis
   of the percentage of CD19^+CD3^− cells. (n = 3; two-way ANOVA). (E and
   F) Representative bioluminescence imaging and statistical analysis of
   the bioluminescence intensity in nonobese diabetic/severe combined
   immunodeficient (NOD/SCID) mice. (n = 7; two-way ANOVA). (G) Body
   weight of the mice. (n = 7; two-way ANOVA). (H) Kaplan-Meier curve
   representing the percentage survival of the mice. (n = 7; log-rank
   test). (I) Specific fluorescence intensity (SFI) of CD19 on the bone
   marrow mononuclear cells (BMMNCs) in six patients with B acute
   lymphoblastic leukemia (B-ALL). (J) The proportion of CD107a^+ cells in
   effector cells after coincubation with primary B-ALL target cells at
   E:T = 1:1 for 5 hours. (n = 6; one-way ANOVA). (K) The release of
   IFN-γ, IL-2, and TNF-α after coincubation with primary B-ALL target
   cells for 24 hours. (n = 6; one-way ANOVA). (L) Direct lysis of
   effector cells against primary B-ALL target cells. (n = 6; two-way
   ANOVA). All values are means ± SEM. *P < 0.05, **P < 0.01, and ***P <
   0.001.

   To further investigate the cytotoxic efficacy of hPSC-CAR-T toward
   CD19^+ B acute lymphoblastic leukemia (B-ALL) in vivo, we established a
   B-ALL mouse model as previously described ([186]40). A total of 3 ×
   10^5 luciferase-expressing Nalm-6 cells (Nalm-6–luc2) were
   intravenously injected into nonobese diabetic/severe combined
   immunodeficient (NOD/SCID) mice. At days 4 and 11, 5 × 10^6 PB-VEC-T,
   hPSC-VEC-T, PB-CAR-T, or hPSC-CAR-T were administered intravenously
   ([187]Fig. 6A). Bioluminescent imaging showed that the injection of
   hPSC-CAR-T inhibited tumor progression ([188]Fig. 6, E and F), delayed
   weight loss ([189]Fig. 6G), and prolonged the survival of the treated
   mice ([190]Fig. 6H). Using the same protocol, we showed that the
   therapeutic performance of hPSC-CAR-T and PB-CAR-T was comparable
   ([191]Fig. 6, E to H).

   Last, we compared the cytotoxicity of hPSC-CAR-T and PB-CAR-T in
   primary B-ALL cells. Bone marrow mononuclear cells (BMMNCs) collected
   from six patients with B-ALL (table S1) were used as target cells. The
   CD19 expression level was detected ([192]Fig. 6I and fig. S6C). We
   showed that hPSC-CAR-T and PB-CAR-T shared similar cytotoxic
   characteristics against CD19^+ target cells, including degranulation
   ability ([193]Fig. 6J), cytokine release ([194]Fig. 6K), and true lytic
   capability ([195]Fig. 6L). These data demonstrate that, through
   engineered expression of anti–CD19-CAR, the hPSC-T generated from
   AE-primed HPCs potently inhibit the progression of leukemia both in
   vitro and in vivo.

DISCUSSION

   In the present study of hPSC differentiation, we constructed the first
   genome-scale gene expression landscape covering the entire course of
   the pluripotent-to-hematopoietic transition at the single-cell level.
   To avoid artificial bias and more precisely track hematopoietic fate
   choices, no cell sorting or marker enrichment was performed before
   capture of the single cells. By constructing the cellular and molecular
   expression profiles of human early hematopoiesis, we identified
   decreased oxygen metabolism as a previously unidentified molecular
   driver of EHT. Functionally, we showed that hypoxia enhanced
   hematopoietic differentiation of hPSCs by activating the arterial
   specification program that governed the generation of HPCs with
   committed T cell potential. The T cells generated from AE-primed HPCs
   showed a strong tumor killing potential both in vitro and in vivo
   following engineered expression of anti–CD19-CAR.

   A recent study of EHT has shown that HE cells are maintained in a
   quiescent state and reentry into the cell cycle is required to initiate
   EHT ([196]11). Our study confirmed active cell division before EHT
   ([197]Fig. 1, G and H, and fig. S1F). Furthermore, we revealed that
   once EHT occurred, gene expression by the cells was gradually silenced
   ([198]Fig. 1G), suggesting that a biphasic regulation of the cell cycle
   plays an important role in EHT. A previous report of the biphasic
   regulation of mesenchymal genes during EHT ([199]27) was also confirmed
   in our study at the single-cell level. We demonstrated that the
   expression of mesenchymal genes was up-regulated before EHT and
   attenuated during EHT ([200]Fig. 1, G and H, and fig. S1F). Beyond
   these initial findings, we demonstrated that oxidative metabolism
   played an important role in EHT ([201]Fig. 3, B to I).

   The effects of hypoxia on the maintenance and differentiation of adult
   HSCs have been widely studied ([202]41). However, because of the
   limitation of resources and ethical issues, little is known about the
   role of hypoxia in regulating early human hematopoiesis. In our study,
   we not only found that the process of EHT was accompanied by a decrease
   in oxygen metabolism but also showed that hypoxia enhanced HPC
   generation by promoting the arterial specification program. The
   mechanism by which hypoxia promotes arterial specification is not well
   characterized. However, a study of mouse embryonic stem cell
   differentiation indicated that hypoxia first induced the generation of
   EC progenitors by up-regulating the TF Etv2 and subsequently driving
   maturation to an AE fate via a Notch1-dependent signaling mechanism
   ([203]42).

   It is well-known that HE cells are derived from a subset of ECs;
   however, where and how the ECs with hematopoietic potential originate
   have not yet been fully elucidated. By dissecting the heterologous
   cellular components and their trajectories involved in EHT in this
   study, we provided evidence that the EC-HC lineages originated from
   early progenitors with mesenchymal features ([204]Fig. 1E).
   Furthermore, a small group of mesenchymal progenitors with EC-HC
   potential, designated EH-Mes, was identified in the D4-EC cluster
   ([205]Fig. 2, A to C and E). A similar population, designated
   mesenchymoangioblast (MB) cells, with endothelial and mesenchymal
   potential has also been identified in a study of hPSC differentiation
   ([206]43). In contrast to the ECs derived from EH-Mes, MB-derived ECs
   are capable of generating mesenchymal stem cells but not hematopoietic
   cells ([207]44).

   Because HSC–potential HE is largely associated with arteries
   ([208]45–[209]47), it has been assumed that this is represented by a
   subpopulation of arterial vascular ECs. This hypothesis was supported
   by a recent transcriptomic study of human embryonic hematopoietic
   development demonstrating that RUNX1 was exclusively expressed in a
   small population of cells in the AE cluster and characterization of
   RUNX1^+ HE cells revealed clear arterial features ([210]16). In our
   study, the D6-HE was transcriptionally similar to the HE cluster
   identified in human embryos and was more similar to arterial HE than
   the D4-HE. Arterial-type HE cells were also identified in a hPSC
   differentiation model with the CD34^+CD144^+CD43^−CD73^−CD184^+
   phenotype, and activation of the arterial specification program in HE
   cells was shown to drive the development of hematopoiesis with more T
   cell potential ([211]9, [212]35). Therefore, the arterial-type HE cells
   with CD34^+CD43^−CD73^−CD184^+ phenotype were applied in our functional
   studies ([213]Fig. 5A). It has been reported that the EHT of
   arterial-type HE cells is Notch dependent ([214]9, [215]35), which may
   explain why it is difficult to promote EHT in arterial-type HE cells
   without any cellular niche support ([216]Fig. 5B).

   We also conducted cell-cell interaction analysis to investigate the
   potential mechanisms by which the AE niche regulates hematopoiesis
   ([217]Fig. 2K and fig. S2, C to E). Notch signaling was validated as a
   critical regulator (fig. S5D); however, considering the many other
   interactions (e.g., PDGFB-PDGFRB) between AE and HE, it can be
   speculated that while Notch activation is required, it may not be
   sufficient for the maintenance of AE function. For example, we showed
   that PDGFB improved human HSC engraftment in our previous study
   ([218]48). Although the Mes and VE identified in our study have neither
   hematopoietic potential nor the ability to promote EHT, they may play a
   key role in the development of HE cells. The interaction between EC and
   Mes has been demonstrated to control the generation of HE cells with
   RUNX1 expression ([219]49).

   Instead of generation of CAR-T from CAR-expressing iPSCs as previous
   reports ([220]50, [221]51), the CAR-T in our study were done by
   producing T cells from hPSCs that were then transduced with the CAR
   construct, due to a concern that CAR expression in hPSCs may alter EHT
   and impair T cell differentiation. The iPSCs derived from T cells
   (T-iPSCs) have been reported with a higher efficiency in T cell
   differentiation than the non–T cell–derived iPSCs (non–T-iPSCs)
   ([222]52). It will be interesting to investigate in the future whether
   T-iPSCs and non–T-iPSCs (or hESCs) have same or different efficiencies
   in HPC generation and its differentiation to different hematopoietic
   lineages. Moreover, the function of the hematopoietic cells generated
   from T-iPSCs and non–T-iPSCs (or hESCs) may be different. In our study,
   the H1-derived T cells highly expressing CD45RA, CD62L, CD28, and CD27
   (fig. S5E) are closest to naïve T cells and T memory stem cells
   ([223]53, [224]54) and show a strong ability to eliminate leukemia
   cells following engineered expression of anti–CD19-CAR.It has been
   reported that naïve T cells and T memory stem cells have stronger
   proliferation and differentiation ability when stimulated by antigen
   and produce greater antitumor activity than the PB-T ([225]53,
   [226]54). This may explain our observation of the better effects of
   hPSC-CAR-T than PB-CAR-T in vivo. The better effects of hPSC-CAR-T are
   also consistent with a previous study that TCR-transduced iPSC-derived
   T cells (iPSC-TCR-T) showed better effects in inhibiting tumor
   progression and extending the survival of mice than the PB-isolated
   TCR-T ([227]55). As a growing number of studies have revealed multiple
   overlapping waves of hematopoietic progenitors generated from HE cells
   ([228]56, [229]57), it is unclear whether the AE-primed CD45^+ HPCs
   with T cell potential are a transient wave of HSCs or T-biased
   definitive progenitors.

   In summary, we construct comprehensive cellular and molecular networks
   toward hematopoiesis of hPSCs at the single-cell level, which provide
   benchmark datasets to understand the origins of human hematopoiesis.
   Moreover, finding the important role of the AE niche function in
   lymphoid development from hPSCs provides an approach that significantly
   improves T cell progenitor production and facilitates advances in the
   translation of hPSC-based immunotherapies to the clinic.

MATERIALS AND METHODS

Study design

   The goal of this study was to investigate the key cellular and
   molecular signals involved in human early hematopoiesis and identify
   strategies for functional hematopoietic cell generation from hPSCs. We
   performed extensive single-cell transcriptomic analyses to map fate
   choices and gene expression patterns during hematopoietic
   differentiation of hPSCs. To identify differences in the activation of
   signals that direct hematopoiesis in vivo and in vitro, we compared the
   transcriptomic differences during early hematopoiesis in hPSCs and
   human embryos. For functional study, a range of in vitro assays using
   cell lines and primary B-ALL cells and in vivo assays using a mouse
   model of B-ALL was performed. For in vitro assays, at least three
   independent experiments were performed. We obtained samples of patients
   with B-ALL from the Institute of Hematology and Blood Diseases
   Hospital, Chinese Academy of Medical Sciences. Informed consent was
   obtained from all participants in accordance with the Declaration of
   Helsinki. The procedure was approved by the ethical advisory board of
   the Institute of Hematology and Blood Diseases Hospital.
   Patient-related information is presented in table S1. For in vivo
   studies, female and age-matched animals were randomly assigned to
   experimental or control groups. The number of animals used in each
   study group is specified in the figure legends. All animal studies were
   performed according to procedures and protocols approved by the
   Institutional Animal Care and Use Committee of Peking Union Medical
   College.

Maintenance and differentiation of hPSCs

   The hESC (H1) and hiPSC (BC1, BC1-T, and JHUi181) lines were maintained
   on Matrigel-coated plates with E8 medium (Life Technologies) as
   described previously ([230]23–[231]25). For differentiation, single
   hPSCs were obtained for sequential EC-HC induction, as described
   previously ([232]23–[233]25) with minor modifications. Briefly, single
   hPSCs were plated at an optimized density at 6 × 10^3 cells per well
   onto vitronectin (PeproTech)–coated 12-well plates in the STEMdiff APEL
   2 Medium (STEMCELL Technologies) supplemented with 3 μM glycogen
   synthase kinase 3 inhibitor CHIR99021 [Applied Biological Materials
   (ABM)], activin A (2 ng/ml; PeproTech), BMP4 (10 ng/ml; PeproTech), or
   10 μM rho kinase inhibitor Y-27632 (STEMCELL Technologies) on day 0.
   After 48 hours (day 2), the medium was changed to STEMdiff APEL Medium
   supplemented with vascular endothelial growth factor (VEGF; 40 ng/ml;
   PeproTech). On day 3, fibroblast growth factor 2 (FGF2; 40 ng/ml; ABM)
   was added to the cultures without aspirating the old medium. From day
   4, the medium was changed to STEMdiff APEL 2 Medium supplemented with
   VEGF (40 ng/ml; PeproTech) and FGF2 (40 ng/ml; ABM) until day 6.
   Cultures were maintained at 37°C under normoxic (room air with 5%
   CO[2]) or hypoxic (1% or 5% O[2]) conditions as indicated. Where
   indicated, a MEK inhibitor (U0126; 3 μM; Selleck Chemicals) was
   included.

T cell differentiation

   For OP9-hDLL1 generation, OP9 cells were transduced with a lentiviral
   vector encoding full-length hDLL1. The highest 5% DLL1-expressing cells
   were sorted by fluorescence-activated cell sorting (FACS) using an
   anti-DLL1 antibody and passaged in minimum essential medium–α/10% fetal
   bovine serum (FBS). Stable expression was confirmed by flow cytometric
   and qRT-PCR analyses of DLL1 expression after several weeks of culture.
   For T cell differentiation, 1 × 10^6 OP9-hDLL1 cells were centrifuged
   with 5 × 10^5 CD45^+ cells to form small 3D aggregates, which were then
   plated onto a 0.4-mm Millicell transwell insert (EMD Millipore) placed
   in a six-well plate containing 1 ml of T cell differentiation medium
   consisting of RPMI 1640 (Gibco), 4% B27 supplement (Thermo Fisher
   Scientific), 30 mM l-ascorbic acid 2-phosphate sesquimagnesium salt
   hydrate (Sigma-Aldrich), 1% penicillin/streptomycin (Thermo Fisher
   Scientific), stem cell factor (20 ng/ml; PeproTech), FLT3L (5 ng/ml;
   PeproTech), and IL-7 (5 ng/ml; PeproTech). Medium was changed every 3
   to 4 days, and the cells were cultured for 8 weeks before harvesting
   for CD3^+ enrichment, followed by a further 1 week of stimulation with
   anti-CD3/CD28 beads (Thermo Fisher Scientific). Cells were harvested by
   adding TrypLE (Gibco) and disaggregating the 3D aggregate by pipetting,
   followed by passage through a 50-mm nylon strainer. The cells were then
   analyzed by flow cytometry on the days indicated.

In vitro function studies of hPSC-CAR-T

   Nalm-6, Raji, and Molm13 cells were maintained in RPMI-1640 medium
   supplemented with 10% FBS (Gibco). T cells were cultured in X-VIVO 15
   medium (Lonza) supplemented with 5% FBS and recombinant human IL-2 (50
   to 100 U/ml; PeproTech) (T cell culture medium). Primary BMMNCs were
   isolated from patients and cultured as previously described ([234]58).

Analysis of direct cytotoxicity

   hPSC-CAR-T, PB-CAR-T, hPSC-VEC-T, or PB-VEC-T were incubated with
   target cells at an effector-to-target (E:T) ratio of 1:1 in T cell
   culture medium without rhIL-2. After 8 and 24 hours, the cells were
   harvested and stained with phycoerythrin (PE)–conjugated anti-human
   CD19 antibody (BioLegend, catalog no. 302208, HIB19) and
   allophycocyanin (APC)-Cy7–conjugated anti-human CD3 antibody
   (BioLegend, catalog no. 300318, HIT3a) for 30 min at 4°C. The cells
   were then washed and resuspended in phosphate-buffered saline for flow
   cytometry analysis. The percentage of CD3^−CD19^+ cells represented the
   residual level of target cells.

Cytokine-releasing assay

   hPSC-CAR-T, PB-CAR-T, hPSC-VEC-T, or PB-VEC-T (2 × 10^5) were
   cocultured with target cells at E:T ratios of 1:1 in 1 ml of T cell
   culture medium without rhIL-2 for 24 hours. After 24 hours, IL-2,
   IFN-γ, and TNF-α in the culture supernatants were detected using ELISA
   kits (R&D Systems, USA) according to the manufacturer’s instructions.

Degranulation assay

   hPSC-CAR-T, PB-CAR-T, hPSC-VEC-T, or PB-VEC-T (1 × 10^5) were
   cocultured with 1 × 10^5 target cells in 200 μl of T cell culture
   medium with PE-conjugated anti-CD107a antibody and 100 U of rhIL-2.
   After 1 hour, monensin (100 μg/ml) was added into the cocultured
   system, and the cells were incubated for 4 hours. The cells were then
   labeled with APC-Cy7–conjugated anti-human CD3 antibody and analyzed by
   flow cytometry. All CD107a^+ cells in GFP^+ cells were regarded as
   degranulated T cells.

In vivo NOD/SCID murine studies

   NOD/SCID female mice (aged 8 weeks) were purchased from the Institute
   of Laboratory Animal Sciences (Chinese Academy of Medical Sciences and
   Peking Union Medical College, China). All animal experiments were
   approved by the Institutional Animal Care and Use Committee of Peking
   Union Medical College. Mice (n = 24) were intravenously inoculated with
   3 × 10^5 Nalm-6–luc2 cells. Four days after transplantation, mice were
   randomly divided into four treatment groups according to the average
   radiance of bioluminescent imaging: hPSC-CAR-T group, PB-CAR-T group,
   hPSC-VEC-T group, and PB-VEC-T group. The mice in these groups received
   intravenous injection of 5 × 10^6 hPSC-CAR-T, PB-CAR-T, hPSC-VEC-T, and
   PB-VEC-T, respectively, on days 5 and 10. Bioluminescent images were
   obtained using Caliper IVIS Lumina II (Caliper Life Sciences, USA), and
   the average radiance was calculated as described previously ([235]59).

Statistical analysis

   Analyses were performed using GraphPad Prism 6.0 software. Survival
   rates were analyzed by the Kaplan-Meier method. Data obtained from
   multiple experiments were reported as the means ± SEM. An unpaired t
   test was used to compare the means from two groups, and analysis of
   variance (ANOVA) was used to compare the means from three or more
   groups. Results with a value of P < 0.05 were considered statistically
   significant. *P < 0.05, **P < 0.01, and ***P < 0.001.

Acknowledgments