Hlf marks the developmental pathway for hematopoietic stem cells but not for erythro-myeloid progenitors

Hematopoietic stem cells (HSCs), which reside in the bone marrow, are the major source of mature blood lineages including myeloid cells and lymphocytes (Orkin and Zon, 2008). In the mouse embryo during midgestation, HSCs are generated from aortic endothelium through the transdifferentiation process known as endothelial-to-hematopoietic transition (EHT; Zovein et al., 2008; Chen et al., 2009; Medvinsky et al., 2011). During EHT, flat-shaped hemogenic endothelia transform into round-shaped hematopoietic cells (Boisset et al., 2010), which subsequently form cell aggregates, called hematopoietic clusters (Yokomizo and Dzierzak, 2010). At embryonic days (E) 10–11, these clusters are thought to contain HSC precursors (pre-HSCs; Taoudi et al., 2008; Zhou et al., 2016), which colonize the fetal liver before or after they mature into fully functional HSCs. It has been reported that HSCs are generated from regions other than the dorsal aorta, such as the yolk sac, placenta, and head (Samokhvalov et al., 2007; Rhodes et al., 2008; Li et al., 2012). Indeed, a recent tracing study revealed that 35%–40% of HSCs in the adult bone marrow may originate from the yolk sac (Lee et al., 2016).

Before the emergence of HSCs (∼E10), lineage-restricted progenitors, such as erythro-myeloid progenitors (EMPs) and lymphoid–myeloid progenitors, are detected in the mouse embryo (Böiers et al., 2013; McGrath et al., 2015a). These are referred to as HSC-independent lineages (Dzierzak and Bigas, 2018). EMPs are the dominant progenitor population at the midgestation stage and produce transient, definitive-type erythrocytes and long-lived, tissue-resident macrophages, which have the potential to proliferate and are maintained in some tissues until adult stage and thereafter (Perdiguero et al., 2015). Similar to aortic clusters, EMPs are generated from hemogenic endothelium at E8–9 in the yolk sac through the formation of hematopoietic clusters (Padrón-Barthe et al., 2014; Frame et al., 2016; Kasaai et al., 2017).

Based on the results of a transgenic rescue experiment that used the unique features of Ly6a and Tek regulatory elements, it has been proposed that HSCs and EMPs arise from distinct hemogenic endothelium (Chen et al., 2011). However, it is still difficult to identify HSC-producing endothelium due to the lack of specific markers, an issue that also hinders the generation of HSCs from pluripotent stem cells in vitro (Rowe et al., 2016). So far, HSC induction without the introduction of genetic materials has not been achieved, whereas EMPs are readily induced, suggesting that current standard culture conditions do not recapitulate the HSC-generating phase of hematopoiesis in the aorta-gonad-mesonephros (AGM) region but rather mimic the EMP-forming situation in the yolk sac (McGrath et al., 2015a). One way to circumvent this issue is the identification of nascent pre-HSC/HSC specific markers suitable for optimizing culture conditions (Li et al., 2017; Tober et al., 2018).

Hepatic leukemia factor (Hlf) encodes a proline- and acid-rich basic region leucine zipper (PAR-bZIP) transcription factor, and recent studies revealed that Hlf is specifically expressed in adult bone marrow HSCs and is a critical regulator of HSC quiescence (Komorowska et al., 2017; Wahlestedt et al., 2017). Patients with acute lymphoblastic leukemia have a reciprocal chromosomal translocation of Hlf with E2A gene (Inaba et al., 1992). In addition, several studies have shown through forced expression that Hlf expression is strongly associated with the acquisition of stem cell properties. Indeed, the ectopic expression of Hlf in HSCs/progenitors reinforces multipotency and self-renewal ability (Shojaei et al., 2005; Gazit et al., 2013). Six transcription factors, including Hlf, can reprogram blood progenitors into transplantable HSC-like cells (Riddell et al., 2014).

Here, using a novel reporter mouse, we analyzed Hlf expression during hematopoietic development in the embryo. Hlf expression begins in E10 aortic clusters during EHT, and Hlf^hi cell fractions in E14 fetal livers are enriched for HSCs that can reconstitute the adult hematopoietic system. In contrast, Hlf expression is not detected in EMPs or in hematopoietic clusters in E9 yolk sac. These results suggest that Hlf expression discriminates the HSC-producing pathway from the EMP-producing pathway in the mouse embryo.

To understand HSC specification during ontogeny and to search for nascent HSC markers, we performed single-cell microarray analysis of developing HSC populations. We previously showed that hematopoietic clusters in the major arteries can be detected and enriched by c-Kit and CD31 staining and that Runx1-GFP (eR1-GFP) transgene expression marks hemogenic endothelial cells (Ng et al., 2010; Yokomizo and Dzierzak, 2010). Using these markers, we collected 85 single cells from five different populations: endothelial cells (eR1-GFP⁻CD31⁺Ter119⁻CD45⁻CD41⁻), hemogenic endothelial cells (eR1-GFP⁺CD31⁺Ter119⁻CD45⁻CD41⁻), E10.5 hematopoietic clusters (c-Kit⁺CD31⁺CD45⁺SSEA1⁻), E12.5 hematopoietic clusters (c-Kit⁺CD31⁺CD45⁺SSEA1⁻), and E14.5 fetal liver HSCs (CD150⁺CD48⁻c-Kit⁺Sca-1⁺Lin⁻Mac-1⁺; Fig. S1). Subsequent cDNA synthesis and microarray analysis revealed a dynamic change in hematopoietic gene expression during HSC development (Fig. 1 A). Among HSC/hematopoietic stem and progenitor cell (HSPC)–related genes, four genes—Mecom, Hlf, Hoxa9, and Procr—showed differential expression patterns compared with other genes (such as Gata2, Myb, and Meis1) and seemed to be up-regulated during HSC development (Fig. 1 A). Because endothelial cells could be an obstacle for imaging or identifying rare cell populations, such as HSCs and pre-HSCs, we excluded Mecom, Hoxa9, and Procr from the list of candidate marker genes, as they are expressed in sorted hemogenic endothelial fractions (Fig. 1 A). Therefore, we focused on Hlf for further detailed analysis.

Hlf encodes the PAR-bZIP transcription factor and is expressed in adult HSCs (Gazit et al., 2013; Komorowska et al., 2017). To further investigate Hlf expression during HSC formation in the embryo, we generated an Hlf reporter mouse. For the Hlf-tdTomato allele, we inserted the sequence encoding T2A-tdTomato before the endogenous stop codon within exon 4 (Fig. 1 B). This should leave endogenous Hlf expression intact in the mice. Indeed, a similar level of Hlf protein expression was observed between Hlf^+/+ and Hlf^tdTomato/+ mice (Fig. S2 A). Blood cell analysis also showed normal hematopoietic differentiation in adult Hlf^tdTomato/+ mice (Fig. S2, B and C).

To investigate Hlf expression in embryonic hematopoiesis, we first analyzed the Hlf^tdTomato/+ fetal liver at E14.5. In flow cytometry analysis, 3.9% ± 1.0% of fetal liver cells were positive for Hlf-tdTomato (Hlf⁺). Virtually all Hlf^hi cells were positive for c-Kit, and lineage marker–positive cells (Lin⁺) were negative for Hlf (Fig. 1, C and E), indicating that the expression is restricted to HSCs/progenitor cells. Immunostaining analysis of E14.5 fetal liver also showed restricted Hlf expression within a subset of c-Kit⁺ cells (Fig. 1 D). Within the c-Kit⁺ fraction, CD150⁺CD48⁻c-Kit⁺Sca-1⁺Lin⁻ (CD150⁺CD48⁻KSL) HSCs had the highest expression of Hlf, and the intensity decreased along with differentiation (Fig. 1 C). A similar expression pattern was also observed in the adult bone marrow (Fig. S2, D and E).

To confirm the correlation between Hlf expression and stem cell potential, we isolated cells with the highest (0.03%) and low/negative Hlf expression level (Hlf^hic-Kit⁺ cells and Hlf^lo/−c-Kit⁺ cells, respectively) and transplanted them separately into irradiated adult mice (Fig. 1 E). Long-term repopulating activity was found for the Hlf^hi fraction but not the Hlf^lo/− fraction (Fig. 1 E). These data indicate that Hlf^hi cells include HSCs in the E14.5 fetal liver.

Embryonic HSCs are generated from hemogenic endothelium through the formation of hematopoietic clusters within the dorsal aorta (Chen et al., 2009). To examine the expression of Hlf in the AGM region, we performed whole-mount immunostaining analysis of E10.5 Hlf^tdTomato/+ embryos. Hlf-tdTomato signals were detected exclusively within the dorsal aorta (Fig. 2 A). Most (98%) of the c-Kit⁺ cells attached to the endothelial wall were positive for Hlf (Fig. 2, B–D). c-Kit⁺ hematopoietic clusters in the vitelline artery and the umbilical artery were also positive for Hlf (Fig. 2 E). On the other hand, endothelial cells (CD31⁺c-Kit⁻) did not express Hlf (Fig. 2, B and C), indicating that Hlf starts to be expressed in nascent hematopoietic cells. Of note, 95% of the c-Kit⁺ circulating cells were negative for Hlf (Fig. 2 D), which, as shown later, may indicate a specific marking of the HSC-producing pathway in the dorsal aorta.

We also examined the expression of Hlf-tdTomato in the E10.5 caudal half region by FACS analysis. As expected, a distinct Hlf⁺c-Kit⁺ subset (0.051% ± 0.014%) was detected, and endothelial cells (CD31⁺CD45⁻CD41⁻) were negative for Hlf (Fig. 2 F). The Hlf⁺c-Kit⁺ subset coexpressed CD41 and exhibited a heterogeneous expression of CD45, consistent with known characteristics of hematopoietic clusters (Yokomizo and Dzierzak, 2010). Because c-Kit⁺ circulating cells were not included in the Hlf⁺c-Kit⁺ subset (Fig. 2 D), this subset represents nearly a pure population of hematopoietic clusters.

We noticed that the intensity of tdTomato in the E10 hematopoietic cluster cells was apparently weak compared with that in the E14.5 HSCs. Given that HSC maturation occurs between E10.5 and E14.5 stages, we speculated that Hlf expression may be an indicator of HSC maturation. To explore this possibility, we compared Hlf-tdTomato expression between pre-HSCs and HSCs. A recent study showed that pre-HSCs are immunophenotypically defined by CD31⁺CD45⁻CD41^loc-Kit⁺EPCR^hi cells (Zhou et al., 2016; Fig. S3 A). As expected, pre-HSCs reside in the hematopoietic cluster subset and, similar to HSCs in the E14.5 fetal liver, exhibited the highest intensity of Hlf expression in the AGM region (Fig. 3 A). Hlf expression in the pre-HSCs was lower than that in the E14.5 HSCs (Fig. 3 A). An increase in Hlf expression in the HSCs was observed until the adult stage (Fig. S3 B). These data suggest that Hlf expression increases with HSC development.

Hematopoietic clusters detached from aortic endothelium colonize the fetal liver via the circulation (Zovein et al., 2008). Therefore, it is plausible that pre-HSC to HSC maturation occurs in the dorsal aorta or fetal liver. To determine which organ provides the HSC maturation niche, we analyzed the expression of Hlf-tdTomato within the AGM and fetal liver. As shown in Fig. 3 B, sequential up-regulation in Hlf was observed in the fetal liver, but not in the AGM after E12.5. Immunostaining analysis also showed an up-regulation in Hlf-tdTomato in the fetal liver between E11.5 and E14.5 (Fig. 3 C). A portion of the Hlf^hi cells in the fetal liver expressed Evi-1 and endothelial protein C receptor (EPCR), two markers of fetal liver HSCs (Iwasaki et al., 2010; Kataoka et al., 2011) that were both identified as embryonic HSC signature genes in our microarray analysis (Fig. 1 A). Both Hlf^hiEvi-1⁺ and Hlf^hiEPCR⁺ subsets were gradually apparent along with the elevation in Hlf expression between E11.5 and E14.5 (Fig. 3 D), supporting the notion that HSC maturation occurs within the Hlf^hi subset in the fetal liver. Localization of the developing HSCs within the Hlf^hi subset was confirmed by long-term repopulation assay. Hlf^hi cells isolated from the E12.5 fetal liver gave long-term multilineage reconstitution (Fig. 3 E). Taken together, we conclude that Hlf expression marks developing HSCs in the embryo.

Before the formation of HSCs, EMPs emerge in the yolk sac and migrate to other organs via the circulation (Lux et al., 2008). Most c-Kit⁺ circulating cells at E10.5 are EMPs (Frame et al., 2016). Immunostaining analysis of the Myb^GFP/+ knock-in reporter mouse, which presumably marks both HSCs and EMPs (Hoeffel and Ginhoux, 2015; Sakamoto et al., 2015), showed that c-Kit⁺ circulating cells in the dorsal aorta are positive for Myb-GFP (Fig. S4 A). On the other hand, analysis of Hlf^tdTomato/+ mice showed that only 5% of c-Kit⁺ circulating cells were positive for Hlf-tdTomato (Figs. 2 C and 4 A), suggesting that Hlf expression does not mark EMPs. To verify this finding, we performed FACS analysis on cells from E9.5 embryos. We found that phenotypic EMPs (c-Kit⁺CD41⁺CD16/32⁺; McGrath et al., 2015a) and their progenies (c-Kit⁻CD45⁺) at E9.5 were negative for Hlf (Fig. 4 B).

EMPs emerge from yolk sac endothelium through the formation of c-Kit⁺ hematopoietic clusters around E9 stage (Frame et al., 2016; Kasaai et al., 2017). To determine whether nascent EMPs were Hlf⁻, we further analyzed the yolk sac at the 19–35 somite pair (sp) stage (E9–10). At E10.5, an Hlf⁺c-Kit⁺ cluster subset (0.54% ± 0.17%) with lymphoid potential was detected in the yolk sac (Fig. 4 C, red rectangle; see also Fig. S4, B–D). At this stage, EMPs were enriched in the Hlf⁻c-Kit⁺ subset (1.8% ± 0.66%; Fig. 4 C, blue rectangle). However, at E9.5 (19–21 sp), the Hlf⁺c-Kit⁺ subset was almost absent (Fig. 4 C). Given that most EMPs are generated at E8–9 (Gomez Perdiguero et al., 2015), these data imply that EMPs are generated through the formation of Hlf⁻c-Kit⁺ hematopoietic clusters. Indeed, a portion of the Hlf⁻c-Kit⁺ subset in the E9.5 yolk sac contains CD31⁺CD41⁺CD45⁻ cells (Fig. 4 C, black rectangle), representing a nascent phenotype of hematopoietic cluster cells (Yokomizo and Dzierzak, 2010; Frame et al., 2016).

Our FACS analysis results suggested that EMPs are generated through the formation of Hlf⁻ hematopoietic clusters. To confirm the existence of Hlf⁻ hematopoietic clusters in the yolk sac, we performed whole-mount immunostaining analysis. Consistent with the FACS data, c-Kit⁺ hematopoietic clusters expressed Hlf at E10.5 (33 and 34 sp), whereas c-Kit⁺ clusters at E9.5 (19 sp) did not (Figs. 4 D and S4 E). Hlf⁻ cells were rarely observed within E10.5 Hlf⁺ hematopoietic clusters (Fig. 2, C and E; and Fig. 4 D). These data indicate that c-Kit⁺ hematopoietic clusters are subdivided into Hlf⁺ and Hlf⁻ types, and also suggest that there is a developmental switch from an Hlf⁻ to Hlf⁺ cluster-producing phase between E9 and E10.

To determine the precise stage of Hlf⁺ hematopoietic cluster emergence, we analyzed the E9 to E10 stages in detail. At 19–21 sp, we observed 0–24 Hlf⁺c-Kit⁺ cells in the yolk sac per embryo (Fig. 4 E), indicating that Hlf⁺ hematopoietic clusters emerge around the 20-sp stage. Collectively, these data suggest that the developmental pathway for EMPs is negative for Hlf.

To obtain mechanistic insight into HSC specification, we compared the gene expression profiles between Hlf⁺ and Hlf⁻ hematopoietic clusters. Hlf⁺ and Hlf⁻ nascent hematopoietic cluster cells (c-Kit⁺CD45⁻; Yokomizo and Dzierzak, 2010) were isolated from E10.5 (35 and 36 sp) caudal half and E9.5 (18–20 sp) yolk sac, respectively, and were subjected to RNA sequencing (RNA-seq) analysis (Figs. 5 A and S5, A and B). We identified 872 genes up-regulated (more than a twofold change) and 318 genes down-regulated in the Hlf⁺ fraction (Figs. 5 B and S5, C–F). Consistent with the specific marking of HSCs by Hlf expression, gene set enrichment analysis (GSEA) indicated that HSC and EMP signatures were highly enriched in Hlf⁺ and Hlf⁻ fractions, respectively (Fig. 5 C). In contrast, analysis with progenitors (HSPCs) and differentiated progeny (erythroid and myeloid) signatures showed similarity between the two fractions (Figs. 5 C and S5 D), suggesting that the Hlf⁻ fraction has characteristics of hematopoietic progenitors but lacks an activation of the HSC program. Accordingly, HSC genes (Hlf, Mecom, Procr, Hoxa9, and Vwf) were specifically detected in the Hlf⁺ fraction (Fig. 5 D). In addition, Hlf⁺ cluster cells expressed a broad range of Hox cluster genes, partly consistent with a previous report that HOXA genes are expressed in human AGM, fetal liver HSPCs, and AGM-like hematopoietic cells induced from human embryonic stem (ES) cells (Dou et al., 2016; Ng et al., 2016). GSEA also revealed enrichment of Notch, bone morphogenetic protein, and Wnt signaling genes in the Hlf⁺ fraction (Fig. S5 D), in line with their positive effects on HSC generation (Hadland et al., 2004; Sturgeon et al., 2014; Crisan et al., 2016). Among HSPC-related genes, factors with known roles in EHT (Runx1, Gata2, Tal1, Meis1, and Rac1; Azcoitia et al., 2005; Ghiaur et al., 2008; Chen et al., 2009; de Pater et al., 2013; Zhen et al., 2013) were comparably expressed in both fractions (Fig. 5 D), probably reflecting the common program for EHT progression in the yolk sac and the major arteries. Together, our RNA-seq data support the notion that acquisition of the HSC or EMP fate takes place at the formation of hematopoietic clusters or earlier.

Our RNA-seq analysis suggested that HSC-specifying genes can be distinguished from other HSPC-related genes (Fig. 5 D). One of the candidate HSC-specifying genes is Mecom (Evi-1). Immunological Genome Project (https://www.immgen.org) data also suggests that Mecom is most closely correlated with Hlf within the HSPC gene network. Given that functional HSCs are absent in Evi-1–deficient embryos (Goyama et al., 2008), we predicted that Evi-1 is involved in preHSC to HSC maturation but not in EMP formation or EHT progression. To clarify the function of Evi-1 in embryonic hematopoiesis and the relationship between Evi-1 and Hlf, we crossed Hlf^tdTomato/+ mice with Evi-1 mutant mice. As expected, Evi-1^−/− embryos at E10.5 had a normal Hlf⁻c-Kit⁺ EMP-enriched fraction (Fig. 6 A), and an analysis of E9.5 yolk sac confirmed absence of Evi-1 expression in EMPs and intact EMP formation in Evi-1^−/− embryos (Fig. 6, B and C). In contrast to Hlf⁻c-Kit⁺ EMP-enriched cells, there were obviously fewer Hlf⁺c-Kit⁺ hematopoietic cluster cells in Evi-1^−/− embryos at E10.5 (Fig. 6 A).

We next performed whole-mount immunostaining analysis to investigate whether hematopoietic clusters formed in Evi-1^−/− embryos. c-Kit⁺ hematopoietic clusters formed normally in the Evi-1^−/− dorsal aorta, but there was a decrease in Hlf-tdTomato expression in the hematopoietic clusters (Fig. 6, D and E), indicating that Evi-1 is not required for EHT but regulates Hlf expression in hematopoietic clusters. Because Hlf expression is indicative of HSC maturation status (Fig. 3), these data also suggest that Evi-1 is involved in HSC maturation. To examine this, we next analyzed E12.5 fetal liver of the Evi-1^−/−::Hlf^tdTomato/+ mouse. An up-regulation in Hlf was observed in Evi-1^+/+ fetal liver cells but not in Evi-1^−/− cells (Fig. 6 F). In addition, Hlf⁺EPCR⁺ and Hlf⁺Sca-1⁺ subsets did not emerge in Evi-1^−/− fetal liver (Fig. 6 G). These data support a role for Evi-1 in the maturation of preHSCs into HSCs and suggest that the Hlf-tdTomato reporter system is a useful tool for dissecting HSC-dependent and -independent lineage development in the mouse embryo.

It remains controversial how Myb is involved in the formation of HSC-independent hematopoiesis (Schulz et al., 2012; Hoeffel et al., 2015; McGrath et al., 2015b; Perdiguero and Geissmann, 2016). Although some c-Kit⁺ cells can be detected in the Myb-deficient embryo (Schulz et al., 2012), the developmental origin of those c-Kit⁺ cells is unclear. To explore the function of Myb in definitive hematopoiesis, we crossed the Myb mutant mouse with the Hlf^tdTomato/+ reporter mouse. At E10.5, the Hlf⁺ hematopoietic cluster subset (Hlf⁺c-Kit⁺) was normally formed in the Myb^−/− embryos (Fig. 7 A). Consistently, whole-mount immunostaining analysis also showed normal formation of c-Kit⁺ aortic clusters in Myb^−/− embryos (Fig. 7, B and C), indicating that Myb is not required for Hlf⁺ hematopoietic cluster formation. In contrast to the Hlf⁺c-Kit⁺ subset, there was a severe decrease in the EMP-enriched Hlf⁻c-Kit⁺ subset in E10.5 Myb^−/− embryos (Fig. 7 A). These results, combined with the Evi-1^−/− analysis in Fig. 6 A, indicate that the formation of Hlf⁺c-Kit⁺ and Hlf⁻c-Kit⁺ subsets is uncoupled. The near absence of EMPs in the E10.5 Myb^−/− embryo was confirmed by an analysis of the yolk sac and the embryo proper (Fig. 7 D). In agreement with this data, c-Kit⁺ circulating cells, which include numerous EMPs, were rarely observed in the Myb^−/− dorsal aorta at E10.5 (Fig. 7 E).

Given that Myb is not expressed in hemogenic endothelium (Fig. 1 A; Swiers et al., 2013), it is likely that Myb is involved in EMP maintenance or expansion but not in EMP formation from hemogenic endothelium. To examine this, we analyzed the yolk sac at E9, which is the major site and stage for EMP formation. Although we observed a slight reduction in the numbers of EMPs (P = 0.02), substantial numbers of EMPs were still formed in the Myb^−/− embryos (Fig. 7 F). Taken together, these data indicate that Myb is required for EMP maintenance or expansion but not for formation.

A detailed analysis of HSC specification has been hampered by the coexistence of HSC-independent lineages, such as EMPs, which have surface markers similar to embryonic HSCs and are abundantly present in the midgestation phase of embryogenesis (Dzierzak and Bigas, 2018). Here, we demonstrate that nascent HSCs and EMPs are distinguished by the expression of Hlf, and this knowledge enables us to access embryonic HSCs and their precursors more easily. Using this new reporter mouse combined with RNA-seq and confocal imaging analyses, we demonstrate that Hlf⁺ and Hlf⁻ hematopoietic clusters contribute to HSCs and EMPs, respectively. Moreover, HSC specification is already activated within Hlf⁺ nascent hematopoietic clusters. These results strongly suggest that HSCs and EMPs are generated from distinct populations of hemogenic endothelium. This inference is consistent with the model proposed in a previous genetic study, where uncoupling formation of HSCs and EMPs was seen in Cbfb^−/− mice rescued by Ly6a-Cbfb or Tek-Cbfb transgenes (Chen et al., 2011).

RNA-seq data showed that EHT-related genes, Runx1, Gata2, Tal1, Meis1, and Rac1, were comparably expressed in Hlf⁺ and Hlf⁻ subsets, probably reflecting the existence of common EHT machinery in HSC- and EMP-producing cluster formation. Moreover, we did not observe significant differences in the gene signatures of progenitor, erythroid, or myeloid genes between the Hlf⁺ and Hlf⁻ subsets, indicating that progenitor features are common to both subsets. We speculate that these progenitor features are the default state of nascent hematopoietic cells, with supplementary HSC program activation through Hlf expression endowing the cells with stem cell properties. Consistently, a recent study showed that polycomb group protein Ezh1 represses the HSC program, including Hlf expression, before the emergence of HSCs (Vo et al., 2018).

In this study, we identified Evi-1 as a factor specific to HSC development. In Evi-1–deficient embryos, Hlf expression was decreased but not completely lost, indicating that Evi-1 is an upstream regulator of Hlf. Given that Mecom is most closely correlated with Hlf in the HSPC gene network (Immunological Genome Project database), Evi-1 may directly regulate the Hlf gene. In contrast to Evi-1, it seems that Hlf is dispensable for HSC generation, as adult Hlf-deficient mice exhibited normal steady-state hematopoiesis (Komorowska et al., 2017). However, we speculate that Hlf is involved in the transcriptional circuitry of HSC specification for the following reasons. First, overexpression of Hlf influences the fate of stem cells or progenitors. For example, Hlf, together with five other transcription factors, can transform committed blood progenitors into functional HSCs (Riddell et al., 2014); this suggests that Hlf can induce stemness. Second, there may be a functional redundancy with other PAR-bZip family proteins (Dbp and Tef) or CCAAT/enhancer binding proteins, which share a similar binding motif with Hlf (Wahlestedt et al., 2017).

Interestingly, we found that Hlf⁺ hematopoietic clusters are also localized in the yolk sac. These cells first appear between E9 and E10 and show a similar surface marker phenotype with intra-aortic Hlf⁺ hematopoietic clusters, which express CD41, CD31, and EPCR (data not shown). Recently, Yzaguirre and Speck (2016) reported that progenitors with lymphoid potential are enriched in Ly-6a-GFP⁺ population in the yolk sac. Since both the lymphoid potential of Ly-6a-GFP⁺ population and Hlf⁺ cells in the yolk sac emerge at ∼E10, and the expression level of Sca-1, which is encoded by Ly-6a gene, correlates with that of Hlf (Fig. 3 D), it is likely that Hlf⁺ hematopoietic clusters in the yolk sac contain Ly6a-GFP⁺ cells. Furthermore, a Lyve1-based lineage tracing study showed that a subset of adult HSCs are generated from yolk sac hemogenic endothelium (Lee et al., 2016). We speculate that yolk sac–derived HSCs may be generated through the formation of Hlf⁺ hematopoietic clusters. Studies with lineage tracing of yolk sac Hlf⁺ cells are required to prove this hypothesis.

The AGM and fetal liver are proposed sites for pre-HSC to HSC maturation (Taoudi et al., 2008; Kieusseian et al., 2012; Rybtsov et al., 2016). We showed that Hlf expression levels correlate well with HSC maturation status. Moreover, we observed a continuous increase in Hlf expression in the fetal liver but not in the AGM, suggesting that the fetal liver is the major site for HSC maturation. Others have reported drastic increases (28–176-fold) in HSC number in fetal liver between E11 and E12 (Kumaravelu et al., 2002; Gekas et al., 2005). This may be explained by the simultaneous maturation of pre-HSCs in the fetal liver during this period, rather than by HSC expansion through rapid symmetric divisions.

The generation of engraftable HSCs from pluripotent stem cells is a long-standing challenge in regenerative medicine. The current typical culture conditions mimic yolk sac–phase hematopoiesis and result in the preferential generation of EMPs from pluripotent stem cells (McGrath et al., 2015a). A recent study has improved the culture conditions for human ES cells by stimulating retinoic acid signaling or modulating WNT and ACTIVIN signaling, leading to AGM-like gene expression in induced hematopoietic cells (Dou et al., 2016; Ng et al., 2016). The study also emphasized that HOXA expression is a landmark of AGM-like hematopoiesis (Ng et al., 2016). Consistent with those findings, we demonstrated that Hox cluster genes are specifically expressed in Hlf⁺c-Kit⁺ nascent hematopoietic cells. Therefore, similar to HOXA gene expression, Hlf expression could serve as a landmark of pre-HSC emergence. In addition to the detection of preHSC emergence, the Hlf-tdTomato reporter can monitor the maturation status of HSCs. Therefore, we believe that the Hlf-tdTomato reporter ES cell line is suitable for optimizing culture conditions for inducing both preHSC emergence and its maturation in vitro.

All mice were on the C57BL/6 background. Runx1-GFP (eR1-GFP), Evi-1^GFP/+, Evi-1^+/−, Myb^GFP/+, and Myb^+/− mice were described previously (Mucenski et al., 1991; Goyama et al., 2008; Ng et al., 2010; Kataoka et al., 2011; Sakamoto et al., 2015). C57BL/6-Ly5.2 (Ly5.2) and C57BL/6-Ly5.1 (Ly5.1) mice were purchased from Japan SLC and Sankyo-Lab Service, respectively. Embryos were generated by timed matings and were staged according to embryonic day, somite pairs, and Thelier criteria (http://www.emouseatlas.org/emap/home.html). For the bone marrow analysis, 8–13-wk-old mice were used. All animal experiments were performed in accordance with institutional guidelines and were approved by the Institutional Animal Care and Use Committee of the National University of Singapore; the Ethics Committees on Animal Experimentation, Juntendo University; and the Animal Care and Use Committee of Kumamoto University.

To generate the targeting vector, the right and left arms (1,393 and 815 bp, respectively) were cloned from mouse genomic DNA (Clontech). A cassette bearing a T2A/tdTomato/BGHpolyA was inserted before the endogenous terminal codon within exon 4.

Oocytes were collected from C57BL/6J female mice (Charles River Laboratories Japan) superovulated with 10 IU of pregnant mare’s serum gonadotropin (Peamex; Sankyo Lifetech) followed by 10 IU of human chorionic gonadotropin (hCG; Sigma-Aldrich) ∼48 h later. The females were sacrificed 14 h after the hCG injection to collect metaphase II oocytes. The complex of oocytes and cumulus cells was released from oviducts into 0.1% hyaluronidase (Sigma-Aldrich) in Hepes-mCZB for 2 min at 37°C to disperse cumulus cells. The cumulus-free oocytes were washed and maintained in modified Whitten’s medium at 37°C in 5% CO₂ air atmosphere until use. Cauda epididymides were removed from C57BL/6J males (Charles River Laboratories Japan). The epididymides were excised with a pair of fine scissors and compressed with forceps to release the dense mass of spermatozoa into 1 ml of Hepes-mCZB in 1.5-ml microtubes (96.7246.9.01; Treff). The spermatozoa dispersed in the microtubes for 5 min at 37°C. After gently pipetting, sperm-containing microtubes were then placed on ice until use. Sperm suspensions (10⁶ in 29 µl) were mixed with 1 µl of targeting vector solution (253 ng/µl) and clustered regularly interspaced short palindromic repeat (CRISPR) vector pX330 expressing Cas9 and Hlf-specific single guide RNA (5′-AGG​CAC​GGG​CCC​CTG​TAA​GA-3′; 192 ng/µl) and incubated at ambient temperature for 2 min. Intracytoplasmic sperm injection (ICSI) was performed according to the method of Kimura and Yanagimachi (1995) with slight modifications. 5 µl of the mixture of sperm-targeting vector and CRISPR vector were diluted with 10 µl of Hepes-mCZB containing 12% polyvinylpyrrolidone (M_r 360,000; Sigma-Aldrich) before use. The ICSI dish contained a drop of Hepes-mCZB, a sperm-targeting vector, CRISPR vector drop, and a Hepes-mCZB/12% polyvinylpyrrolidone needle-cleaning drop. Injections were performed by micromanipulators (Leica) with a PMM-150 FU piezo-impact drive unit (Prime Tech) using a blunt-ended, mercury-containing injection pipette with ∼6 µm inner diameter. A single spermatozoon was drawn, tail first, into the injection pipette and moved back and forth until the head-midpiece junction was at the opening of the injection pipette. The head was separated from the midpiece by applying one or more piezo pulses. After discarding the midpiece and tail, the head was redrawn into the pipette and injected immediately into an oocyte. Sperm-injected oocytes were washed with modified Whitten’s medium and cultured in 20-µl drops of the same medium covered with mineral oil at 37°C in 5% CO₂ in air. The oocytes were examined ∼6 h after ICSI for survival and activation. The oocytes with two well-developed pronuclei were recorded at 7 h after ICSI to assess fertilizing ability. Two-cell-stage embryos, cultured for ∼24 h and developed from oocytes fertilized by ICSI, were transferred to oviducts of the pseudopregnant ICR recipient females (Charles River Laboratories Japan) that were mated to vasectomized ICR male mice (Charles River Laboratories Japan). The embryo transfer to the pseudopregnant females was performed on the day that the vaginal plug was detected.

Generated Hlf^tdTomato/+ reporter mice were genotyped by PCR amplification using the following primers: forward (5′-TAG​TTG​CCA​GCC​ATC​TGT​TG-3′) and reverse (5′-TCC​CAT​TCT​GAG​ATA​CAC​CAG​TG-3′). The annealing location of the reverse primer was designed to be outside the homology arm. Wild-type Hlf allele was detected by PCR using forward (5′-CAG​GAG​GTG​GCT​GAT​TTA​AG-3′) and reverse (5′-TCC​CAT​TCT​GAG​ATA​CAC​CAG​TG-3′) primers.

Whole-mount immunostaining was performed as described previously (Yokomizo et al., 2012). Briefly, embryos were fixed for 20 to 30 min in 2% paraformaldehyde diluted in PBS on ice and dehydrated in graded concentrations of methanol/PBS (50%, 100%; 10 min each). For staining around the dorsal aorta region, the yolk sac, head, limb buds, and lateral body wall were removed in 100% methanol. When using streptavidin/biotin reaction, endogenous biotin activity was blocked with streptavidin/biotin blocking kit (Vector Laboratories) following the manufacturer’s instructions. After preincubation with PBS-MT (PBS containing 0.4% Triton X-100 and 1% skim milk) for ≥1 h, samples were incubated overnight with primary antibody diluted in PBS-MT and washed three times in PBS-MT throughout the next day (2–3 h per wash). Secondary antibody was diluted in PBS-MT, and samples were incubated overnight. Primary antibodies used were for c-Kit (rat, 2B8; BD Biosciences), biotinylated CD31 (rat, MEC 13.3; BD Biosciences), c-Kit (goat; R&D Systems), GFP (rabbit; MBL), RFP (rabbit; MBL), and RFP (rabbit; Rockland Immunochemicals). Secondary antibodies (or streptavidin conjugates) were goat anti-rat IgG-Alexa Fluor 555 (Invitrogen), goat anti-rabbit IgG-Alexa Fluor 647 (Invitrogen), donkey anti-rat IgG-Alexa Fluor 488 (Jackson ImmunoResearch), donkey anti-goat IgG-cyanine 3 [Cy3] (Jackson ImmunoResearch), and Cy3-streptavidin (Jackson ImmunoResearch). Nuclei were visualized with Hoechst 33258 (Invitrogen).

Freshly dissected femurs were fixed in cold paraformaldehyde-lysine-periodate buffer for 8 h at 4°C, rehydrated in 30% sucrose solution for 48 h, and snap frozen in optimum cutting temperature compound (Tissue Tek). Frozen femurs were sectioned using a Leica cryostat until the bone marrow cavity was exposed along the longitudinal axis. This procedure was repeated on the opposite side. The bone samples were washed with PBS and blocked overnight at 4°C in blocking solution (0.2% Triton X-100, 1% BSA, and 10% donkey serum in PBS). The bone marrow slices were stained with anti-laminin (rabbit; Sigma-Aldrich), Alexa Fluor 488–conjugated anti-Sca-1 (rat, E13-161.7; BioLegend), anti–c-Kit (goat; R&D Systems), and anti-RFP (chicken; Rockland Immunochemicals) overnight at 4°C in blocking solution, washed overnight in PBS, and stained with donkey anti-rabbit IgG-Alexa Fluor 488 (Jackson ImmunoResearch), donkey anti-goat IgG (heavy and light chains [H+L])–Alexa Fluor 647 (Jackson ImmunoResearch), and donkey anti-chicken IgY (IgG; H+L)-Cy3 (Jackson ImmunoResearch).

Immunostained caudal half, yolk sac, and fetal livers were mounted in a 1:2 mix of benzyl alcohol and benzyl benzoate to increase the transparency of tissues and were analyzed with a confocal microscope (Olympus FV-1200 equipped with gallium arsenide phosphide photomultiplier tube detectors, UPLSAPO 20×/NA 0.75). Immunostained bone marrow samples were incubated overnight in RapiClear1.52 (SunJin Lab Funakoshi). Samples were mounted on glass slides and analyzed with a confocal microscope (Leica SP8, HC PL APO CS2 20×/0.75 DRY). Three-dimensional (3D) reconstructions were generated from z-stacks with Olympus FluoView software or Imaris software (Bitplane).

Single-cell suspensions were prepared by treating tissues with collagenase (0.125% in PBS, 10% FCS, and 1% penicillin/streptomycin) for 1 h at 37°C. Bone marrow cells were prepared by crushing the bones using a mortar. Cell number was counted using a TC20 automated cell counter (Bio-Rad). Cells were stained with fluorescence-conjugated antibodies: c-Kit (2B8, BV421; BioLegend), c-Kit (2B8, PE-Cy7; BD Biosciences), c-Kit (2B8, allophycocyanin [APC]; BD Bioscience), CD31 (MEC13.3, FITC; BD Biosciences), SSEA-1 (MC-480, PE; eBiosciences), SSEA-1 (MC-480, V405; BD Biosciences), CD45 (30-F11, peridinin-chlorophyll protein [PerCP]-Cy5.5; BD Biosciences), CD45 (30-F11, PerCP-Cy5.5; BioLegend), CD45 (30-F11, BV510; BD Biosciences), Sca-1 (D7, PerCP-Cy5.5; eBioscience), Sca-1 (D7, PE-Cy7; BioLegend), Sca-1 (D7, APC-Cy7; BioLegend), CD150 (TC15-12F12.2, APC; BioLegend), CD150 (TC15-12F12.2, Alexa Fluor 488; BioLegend), CD48 (HM48-1, FITC; BD Biosciences), CD48 (HM48-1, APC-Cy7; BioLegend), CD41 (MWReg30, PerCP-eFluor710; eBioscience), CD16/32 (93, APC; BioLegend), EPCR (eBio1560, APC; eBioscience), Ter119 (TER-119, biotin; BD Biosciences), Ter119 (TER-119, PE; BD Biosciences), CD3e (145-2C11, biotin; BD Biosciences), CD3e (145-2C11, PE; BD Biosciences), CD4 (GK1.5, biotin; BD Biosciences), CD4 (GK1.5, PE; BD Biosciences), CD8a (53-6.7, biotin; BD Biosciences), CD8a (53-6.7, PE; BD Biosciences), B220 (RA3-6B2, biotin; BD Biosciences), B220 (RA3-6B2, PE; BD Biosciences), Gr-1 (RB6-8C5, biotin; BD Biosciences), Gr-1 (RB6-8C5, PE; BD Biosciences), Mac-1 (M1/70, biotin; BD Biosciences), Mac-1 (M1/70, PE; BD Biosciences), and Mac-1 (M1/70, APC-Cy7; BD Biosciences). Cells were analyzed by FACS AriaII, FACS SORPAria, or FACS Canto II (BD Biosciences). Mac-1 was used only in the lineage cocktail for bone marrow. Data were analyzed using FlowJo software (TreeStar).

Sorted cells from Hlf^tdTomato/+ (C57BL/6-Ly5.1/5.2) mice were coinjected intravenously with spleen cells (2 × 10⁵, Ly5.1) or bone marrow cells (2 × 10⁵, Ly5.1) into 9.0 Gy (4.5 Gy, each of two doses)-irradiated (C57BL/6-Ly5.2) mice. Secondary transplantation was performed using 5 × 10⁶ or 10⁷ bone marrow cells. Repopulation was assayed at 4, 8, 12, and 16 wk after transplantation. Peripheral blood was collected by retro-orbital bleeding under anesthesia. Cells were stained with antibodies to CD45.1 (A20, FITC), CD45.2 (104, APC), Mac-1 (M1/70, PerCP-Cy5.5), Gr-1 (RB6-8C5, PE), B220 (RA3-6B2, PE-Cy7), CD4 (GK1.5, APC-Cy7), and CD8 (53-6.7, APC-Cy7), all purchased from BioLegend. Cells were analyzed by FACS Canto II.

Cells were sorted directly into 4.1 µl lysis buffer in 96-well plates using the FACSAria Automated Cell Deposition Unit. Doublets were excluded by forward scatter width/height (FSC-W/FSC-H) and side scatter width/height (SSC-W/SSC-H) gating. To ensure single-cell isolation, we first sorted 220 single cells from an E10.5 embryo into a 96-well plate and examined them under a stereomicroscope. We observed 217 single cells (98.6%) and three vacant wells. Lysis buffer was transferred into a PCR tube, followed by the addition of spike RNA mixture (Lys, Dap, Phe, and Thr artificially polyA-tailed RNAs from Bacillus subtilis at 1,000, 100, 20, and 5 copies, respectively; American Type Culture Collection; 87482, 87483, 87484, and 87486). Single-cell cDNA amplification was performed as previously described (Kurimoto et al., 2007). Briefly, first-strand cDNAs were synthesized using a poly(dT)-tailed primer. The unreacted primer was eliminated by exonuclease (Takara), and the second strands were generated with a second poly(dT)-tailed primer after poly(dA) addition to the first-strand cDNAs. The quality of cDNAs was assessed by quantitative real-time PCR for housekeeping genes (Gapdh and β-actin) and spike RNAs. Template preparation for microarray from the amplified cDNAs was performed as previously described (Kurimoto et al., 2007) with slight modification (T7 promoter addition by 12-cycle PCR, instead of a nine-cycle PCR). Quantity and quality of fragmented cRNAs were assessed with a RNA 6000 Pico Kit (Agilent Technologies). The qualified cRNA was hybridized to the GeneChip Mouse Genome 430 2.0 Arrays (Affymetrix) according to the manufacturer’s instructions. Microarray data processing was performed using the guanine cytosine robust multi-array analysis algorithm implemented in the Bioconductor package affy. Because E14.5 HSCs contained less mRNA than E10.5 hematopoietic cluster cells, slight signal saturation was observed in E14.5 HSCs when equal amounts of amplified cRNA were applied to the array chips. This was corrected by the signal intensity of spike RNAs. A heatmap was generated using the matrix2png web interface. All microarray data were deposited in Gene Expression Omnibus under accession no. GSE113892.

Real-time PCR was performed using a 7500 Fast Real-Time PCR System (Thermo Fisher Scientific) and Power SYBR Green Master Mix (Applied Biosystems) according to the manufacturer’s instructions.

Random displacement amplification sequencing was performed as previously described (Hayashi et al., 2018). Using 100 sorted cells, first-strand cDNA was synthesized using PrimeScript RT reagent kit (Takara) and not-so-random primers (Ozsolak et al., 2010). Following synthesis of the first strand, the second strand was synthesized using Klenow Fragment (3′-5′ exo-; New England Biolabs) and complementary chains of not-so-random primers. After purification of the double-stranded cDNA, the library for RNA-seq was prepared and amplified with a Nextera XT DNA sample Prep Kit (Illumina), according to the manufacturer’s instructions. This prepared library was sequenced on a Next-Seq system (Illumina). Each obtained read was mapped to the reference sequence “GRCm38/mm10” using CLC genomic workbench v7.0.4 (Qiagen), and expression levels were normalized and subjected to statistical analyses based on EdgeR. A heatmap was generated using the matrix2png web interface. All RNA-seq data were deposited in Gene Expression Omnibus under accession no. GSE114164.

GSEA was performed using GSEA v3.0 (http://www.broadinstitute.org/gsea/). Gene sets were obtained from MSigDB (http://www.broadinstitute.org/gsea/msigdb), Chambers et al. (2007), Gazit et al. (2013), Mass et al. (2016), Mercher et al. (2008), and Wahlestedt et al. (2017). Gene ontology analysis was performed using the DAVID 6.7 website (https://david.ncifcrf.gov).

To make protein extracts for immunoblotting, magnetic-activated cell sorted (Miltenyi Biotec) c-Kit⁺ cells were lysed in radioimmunoprecipitation assay buffer (Cell Signaling). The protein concentration was determined with the BCA protein assay kit (Thermo Fisher Scientific). Equal amounts of protein were denatured, electrophoresed using 12.5% SDS-PAGE, and transferred to a polyvinylidene fluoride membrane. The following primary antibodies were used for immunoblotting detection: anti-HLF (mouse, 4D8; Santa Cruz Biotechnology), anti-RFP pAb (rabbit; Rockland Immunochemicals), and anti–β-actin (rabbit; Cell Signaling). Goat anti-rabbit IgG (Santa Cruz Biotechnology) and goat anti-mouse IgG (Jackson ImmunoResearch) were used as the secondary antibodies. Chemiluminescence was detected using the Pierce ECL Western Blotting Femto kit (Thermo Fisher Scientific), and images were captured using an FX7 camera (Fusion).

Peripheral blood was collected by retro-orbital bleeding under anesthesia, and complete blood cell counts were performed using an automated hematology analyzer, pocH-100i V Diff (Sysmex).

For the generation of lymphoid cells, sorted cells were cultured in a T25 culture flask for 11–13 d on an OP9 or OP9-DL1 stromal cell layer in αMEM medium (Gibco) containing 10% FBS and 5 × 10⁻⁵ M 2-mercaptoethanol, supplemented with 50 ng/ml of stem cell factor (R&D Systems), 50 ng/ml of FMS-like tyrosine kinase 3 ligand (Flt3L; PeproTech), and 20 ng/ml of IL-7 (R&D Systems). Hematopoietic cells were harvested and analyzed for the expression of surface markers (B cells; B220⁺CD19⁺ cells, T cells; and CD4⁺CD8a⁺ cells) using flow cytometry.

The statistical significance of differences between datasets was determined by Student’s t test using Prism 6 software (GraphPad). All error bars represent SD. Two-tailed P values are shown (significance levels: *, P < 0.05; **, P < 0.01; ***, P < 0.001; and ****, P < 0.0001).

Fig. S1 shows the gating strategy for single-cell microarray analysis. Fig. S2 shows characterization of adult Hlf-tdTomato mice. Fig. S3 shows elevation of Hlf-tdTomato expression during HSC development. Fig. S4 shows characterization of Hlf⁺c-Kit⁺ cells in the yolk sac. Fig. S5 shows RNA-seq analysis of Hlf⁺ and Hlf⁻ cells.

We thank Seiya Mizuno and Marito Araki for designing the targeting construct; Hiroshi Sakamoto, Takashi Moriguchi, and Masayuki Yamamoto for experimental materials; Saori Morino-Koga and Minetaro Ogawa for assistance with lymphoid potential assay; Toshiyuki Kobayashi for Kumamoto earthquake donation; and Takako Ideue for technical assistance.

This work was supported by Japan Society for the Promotion of Science Kakenhi Grant (26461454, 17K09958, and 26221309, Japan) and SENSHIN Medical Research Foundation, the Singapore Ministry of Health's National Medical Research Council under its Singapore Translational Research Investigator Award, and the National Research Foundation Singapore and the Singapore Ministry of Education under its Research Centres of Excellence initiative. We also acknowledge the Research Institute for Diseases of Old Age and the Center for Biomedical Research Resources, Juntendo University School of Medicine.

The authors declare no competing financial interests.

Author contributions: T. Yokomizo conceived the project and designed and performed the research. N. Watanabe made the Hlf-tdTomato targeting construct and performed peripheral blood cell analysis. T. Umemoto performed the RNA-seq analysis. J. Matsuo, A. Shimono, and S. Mori assisted with single-cell microarray analysis. Y. Kihara performed Western blotting analysis. R. Harai and H. Takizawa performed whole-mount immunostaining of the bone marrow. E. Nakamura and N. Tada performed ICSI for generating the Hlf^tdTomato/+ mouse. T. Sato and M. Kurokawa contributed analytical tools. N. Nakagata assisted with animal transport and performed in vitro fertilization. T. Yokomizo, T. Takaku, D.G. Tenen, M. Osato, T. Suda, and N. Komatsu participated in project planning. T. Yokomizo, M. Osato, and T. Suda wrote the manuscript.