Interplay between SOX7 and RUNX1 regulates hemogenic endothelial fate in the yolk sac

Despite being the focus of intense research over recent years, the molecular mechanisms regulating the emergence of hematopoietic stem and progenitor cells (HSPCs) from hemogenic endothelium (HE) remains poorly understood. Clearly, this process of endothelial-to-hematopoietic transition (EHT) involves complex changes in gene expression that are regulated by an intricate balance of endothelial and hematopoietic transcription factors. It is now widely accepted that the hematopoietic transcription factor RUNX1 is a key master regulator of this process (Chen et al., 2009; Lancrin et al., 2009; Lie-A-Ling et al., 2014). RUNX1 is expressed in HE and emerging HSPCs both in the yolk sac and in the embryo proper (Frame et al., 2015; Lancrin et al., 2009; North et al., 1999), and controls the downregulation and upregulation of endothelial and hematopoietic transcriptional programs, respectively (Lancrin et al., 2012; Tanaka et al., 2012; Thambyrajah et al., 2016). RUNX1 has been shown to act in a heptad of interacting transcription factors in hematopoietic progenitor cells (HPCs) comprising FLI1, ERG, TAL1, LYL1, LMO2 and GATA2 (Wilson et al., 2010). Furthermore, it has been demonstrated that RUNX1 interacts with core binding factor β (CBFβ), which is thought to be key in enhancing RUNX1 binding to its DNA-binding sites, as well as enhancing RUNX1 protein stability (Huang et al., 2001; Link et al., 2010; Wang et al., 1993).

The SOXF family of transcription factors comprises SOX7, SOX17 and SOX18 (Bowles et al., 2000; Lilly et al., 2016; Wright et al., 1993). We and others have previously demonstrated that SOX7 is expressed in HE cells and its downregulation is necessary for the subsequent emergence of hematopoietic progenitors (Costa et al., 2012; Gandillet et al., 2009; Moignard et al., 2015). Therefore, SOX7 and RUNX1 appear to have contrasting effects in regulating opposing transcriptional programs in HE. Recent data have demonstrated that the interplay between SOX17 and RUNX1 regulates the emergence of HSPCs from the dorsal aorta (Bos et al., 2015; Lizama et al., 2015). It was further suggested that SOX17 binds to regulatory regions at the Runx1 locus and inhibits Runx1 expression, therefore maintaining endothelial identity of the arterial HE cells (Lizama et al., 2015). It was also proposed that, as HE differentiates, RUNX1 downregulates Sox17 expression (Lizama et al., 2015). However, the molecular and biochemical details of the interaction between RUNX1 and SOX17 were not investigated.

Although the role of SOX17 in regulating the expression of RUNX1 has been demonstrated in arterial endothelial cells of the dorsal aorta at E10.5, its role in regulating the emergence of HPCs from the yolk sac has not been studied. HE is detected earlier in the yolk sac than in the dorsal aorta (from E8.5 to E9.5), and therefore arises before, during and after vascular remodelling (Frame et al., 2015). In contrast to the embryo proper, in the yolk sac HPCs emerge from a spectrum of endothelial cells in both arterial and venous regions, and in both large and small blood vessels (Frame et al., 2015). As the endothelial expression of SOX17 is restricted to arteries (Corada et al., 2013), this factor cannot control RUNX1 function and EHT in non-arterial vasculatures. Other endothelial transcription factors must therefore be important for regulating HE in the yolk sac. SOX7 is a prime candidate for this role, as previous studies have demonstrated that this SOXF factor is expressed from E7.5 in the yolk sac and is widely expressed in the developing vascular network: in both arteries and veins, and in large and small blood vessels (Gandillet et al., 2009; Sakamoto et al., 2007; Zhou et al., 2015).

Earlier studies have documented the interplay between SOX and RUNX factors. SOX9 and RUNX2 have been demonstrated to physically interact via their respective HMG and RUNT domains to regulate the differentiation of osteo/chondro progenitors (Akiyama et al., 2005; Zhou et al., 2006). SOX9 inhibits RUNX2 DNA binding and RUNX2 transcriptional activity, thereby promoting chondrogenesis; however, in progenitors expressing low levels of SOX9, RUNX2 is free to promote osteogenesis (Zhou et al., 2006). Therefore, the balance between the two factors regulates cell fate decisions that promote either osteogenesis or chondrogenesis (Zhou et al., 2006).

In this study, we show that enforced expression of SOX7 in HE inhibits the upregulation of the hematopoietic transcriptional program, including RUNX1 and TAL1 target genes. Surprisingly however, in contrast to what has been reported with SOX17, we found that SOX7 does not repress Runx1 expression. We found that SOX7 inhibits the transcriptional activity of RUNX1 via direct protein-protein interactions through their respective HMG and RUNT domains, in a manner akin to that of SOX9 and RUNX2. Through this interaction, we show that SOX7 inhibits RUNX1 DNA binding, as well as its interaction with its pro-hematopoietic co-factor CBFβ. Finally, we demonstrate using immunofluorescence and single-cell qPCR profiling that SOX7 and RUNX1 are co-expressed in yolk sac HE at E8.5 and E9.5. Collectively, these data demonstrate for the first time how the balance between interacting endothelial and hematopoietic transcription factors regulates the dynamically changing transcriptional programs during HE differentiation and EHT.

In order to investigate the effect of SOX7 on hematopoietic gene expression in HE, we used a doxycycline-inducible Sox7 (iSox7) ESC line, which was differentiated via embryoid body (EB) formation. FLK1⁺ cells sorted from iSox7 EBs were cultured as a monolayer for 48 h to induce HE differentiation and cells were pulsed with doxycycline for 6, 12 and 24 h before harvesting and RNA-seq analysis (Fig. 1A). Principle component analysis (PCA) identified clear distinctions in the transcriptional programs of untreated and doxycycline-treated cells at 6, 12 and 24 h (Fig. S1A). At 6 h post-doxycycline induction, 130 genes were differentially expressed. Of these, 91 genes were downregulated, including key hematopoietic transcriptional regulators such as: Nfe2, Gfi1b, Spi1, Gata1 and Myb (Fig. 1B and Table S1). Ingenuity pathway analysis (IPA) of these genes identified a significant enrichment for hematopoietic GO terms such as: hematological system development and function, hematopoiesis, and development of hematopoietic cells (Fig. 1C,D). At 12 and 24 h post-doxycycline induction, 341 and 566 genes, respectively, were differentially expressed (Tables S2 and S3). Ingenuity pathway analysis (IPA) of the genes downregulated by SOX7 at 12 and 24 h again identified a significant enrichment for genes involved with hematopoietic GO terms (Fig. S1B,C; data not shown).

To examine whether these transcriptional changes following SOX7 enforced expression translate to functional changes in HE differentiation, FLK1⁺ cells sorted from iSox7 EBs were cultured as a monolayer for 48 h before treatment with and without doxycycline for 24 h. Cells were replated into hematopoietic progenitor cell CFU media with and without doxycycline and analysed after 8 days. Doxycycline treatment resulted in a decrease in the numbers of primitive and definitive hematopoietic colonies, whereas there was an expansion of poorly differentiated blast-like colonies, as previously described (Costa et al., 2012) (Fig. S1D). Harvesting the cells from the CFU assays identified that doxycycline treatment expanded the numbers of cells expressing the endothelial marker TIE2, while the numbers of CD45⁺/CD41⁻ hematopoietic cells decreased (Fig. S1E,F). These data, together with the RNA-seq data set, indicate that the enforced expression of SOX7 inhibits the full process of EHT.

In order to uncover the upstream positive regulators of the set of genes rapidly downregulated by SOX7 at 6 h, we further examined this gene set using ingenuity upstream analysis. RUNX1 and TAL1 were identified as upstream activators of these genes (Fig. 1E); the expression levels of RUNX1 and TAL1 themselves were not affected by SOX7-enforced expression. Furthermore, we observed that 53.8% of the genes downregulated by SOX7 at 6 h were bound by RUNX1 and/or TAL1 in HPC-7 hematopoietic progenitors, whereas 40.7% of the genes were bound by both RUNX1 and TAL1 (Fig. 1F). Collectively, these data indicate that in HE, SOX7 inhibits the expression of hematopoietic genes, many of which are direct transcriptional targets of RUNX1 and TAL1.

Previous studies have suggested that SOX7 is expressed from the onset of endothelial differentiation in the yolk sac at E7.5, whereas SOX17 is expressed in developing arterial endothelium at later stages of development (Corada et al., 2013; Gandillet et al., 2009; Sakamoto et al., 2007). In order to define the expression of the three SOXF factors in yolk sac HE, we sorted FLK1⁺/RUNX-RFP⁺/CD41⁻ HE cells from Runx1-rfp transgenic E8.5 embryos and performed single-cell qPCR. Substantial overlap for the expression of the three SOXF factors was evident; however, SOX7 was the most broadly expressed across the HE population: SOX7 was expressed in 80% of cells and 20% of cells expressed only SOX7 (Fig. 2A). This is in stark contrast to SOX17, which was expressed in fewer than 50% of HE cells (and as little as 4% of these cells expressed only SOX17). These data suggest that, in yolk sac HE, SOX7 is likely to be the most important SOXF factor regulating its development.

In order to further investigate SOX7 and RUNX1 co-expressed in yolk sac HE, we performed immunofluorescence for RUNX, SOX7 and PECAM1 in E8.5 and E9.5 embryos. RUNX staining in yolk sac at E8.5 showed similar patterns to proximal RUNX1-RFP expression (Fig. S2A,B), confirming previous studies demonstrating that RUNX1 is the prevalent RUNX factor expressed in the yolk sac at this stage of development (Landry et al., 2008; Otto et al., 1997). SOX7 was expressed in the yolk sac at E8.5, including in endothelial cells on the edges of blood island regions (Fig. S2B and Fig. 2B,C). RUNX expression was found mainly around the blood island regions, as previously described (Fig. S2B and Fig. 2B,C) (Frame et al., 2015). Importantly, and in agreement with our single-cell PCR analysis, we observed RUNX and SOX7 co-expressing endothelial cells on the edge of the blood island (Fig. 2B,C).

By E9.5, SOX7 was expressed throughout the yolk sac, but higher SOX7 expression was evident in endothelial cells of the larger blood vessels (Fig. S2C). At this stage after vascular remodelling, the blood islands disappear and RUNX is expressed in clusters throughout the yolk sac, as previously described (Fig. S2C) (Frame et al., 2015). RUNX and SOX7 co-expressing endothelial cells were more abundant than at E8.5, and these cells generally expressed lower levels of RUNX than cells in emerging hematopoietic clusters (Fig. 2D). SOX7 was also expressed in endothelial cells lining the dorsal aorta at E9.5, and endothelial cells co-expressing SOX7 and RUNX were clearly detected (Fig. 2E). Collectively, these data reveal that, in yolk sac HE, SOX7 is the predominant SOXF factor, whereas in the dorsal aorta it is likely that both SOX17 and SOX7 regulate the behaviour of HE. Furthermore, SOX7 and RUNX are found co-expressed in subsets of endothelial cells throughout development at sites of blood cell emergence.

To investigate whether SOX7 also inhibits the expression of RUNX1 target genes in vivo, we analysed the expression of some of these targets in enriched yolk sac HE sorted on the FLK1⁺/CD41⁻ immuno-phenotype from E8.5 Sox7^+/+ and Sox7^−/− embryos. QPCR analysis identified no change in the levels of RUNX1 transcripts, whereas there were significantly higher expression levels of Nfe2, Myb and Spi1 in Sox7^−/− cells (Fig. 3A). These data support the notion that SOX7 does not affect RUNX1 expression, but modulates the expression of potential RUNX1 target genes in yolk sac HE. In order to examine the role of SOX7 in regulating yolk sac HE differentiation in vivo, we isolated yolk sacs from E9.5 doxycycline-inducible iSox7 or wild-type embryos, which were cultured either as explants or dissociated and cultured on OP9 cells (Fig. 3B). Doxycycline addition to iSox7 yolk sac explant cultures resulted in a decrease in the number of CD45⁺ hematopoietic cells compared with wild-type yolk sacs, indicating that the enforced expression of SOX7 inhibited the emergence of hematopoietic cells (Fig. 3C). In yolk sac dissociation cultures, SOX7-enforced expression expanded or maintained the TIE2⁺/CD41⁺ and TIE2⁺/cKIT⁺ HE populations compared with wild-type yolk sac cultures (Fig. 3D,E). Together, these data indicate that SOX7 inhibits the differentiation of HE and the subsequent emergence of hematopoietic progenitor cells.

The fact that SOX7 inhibits the expression of RUNX1 target genes, but not RUNX1 itself, suggested that SOX7 may inhibit the transcriptional activity of RUNX1 in a similar manner to that of SOX9, which inhibits RUNX2 transcriptional activity during osteo-chondrogenic differentiation (Akiyama et al., 2005). To investigate this hypothesis, we performed transcriptional activity assays with the osteocalcin 6×OSE2-luc RUNX luciferase reporter. RUNX1 strongly activated the transcription of the reporter construct, but the addition of SOX7 completely inhibited this RUNX1-induced activation, with levels returning to baseline (Fig. 4A). The RNA-seq data presented above revealed that the hematopoietic transcriptional regulator Nfe2 was rapidly downregulated upon doxycycline-induced SOX7 expression; furthermore, the promoter region of Nfe2 has been shown previously to be bound by RUNX1 in hematopoietic progenitors (Goode et al., 2016; Wilson et al., 2010). Therefore, we cloned the NFE2 promoter region (−3562/+58 TSS) containing RUNX1 and FLI1 ChIP-seq peaks (Goode et al., 2016; Wilson et al., 2010) and performed transcriptional activity assays. The addition of RUNX1 and FLI1 individually produced minimal activation of the Nfe2 promoter; however, when added together, a stronger activation was observed (Fig. 4B). The addition of SOX7 inhibited the activation of the Nfe2 promoter by RUNX1 and FLI1, with levels returning to baseline (Fig. 4B). Deletion of two RUNX1-binding sites within the Nfe2 promoter region abrogated the activation by RUNX1 and FLI1, confirming that the Nfe2 promoter is directly activated by RUNX1 (Fig. 4C and Fig. S3).

In order to characterise the important domains of SOX7 that are required for its inhibitory effect on RUNX1 transcriptional activation, we performed mutagenesis of SOX7 followed by luciferase transcriptional analysis (Fig. 4D). Deletion of the HMG domain of SOX7 completely abrogated its inhibitory effects on RUNX1, whereas deletion of the transactivation domain had no effect (Fig. 4E). Furthermore, the R60E point mutation in the HMG domain reversed the inhibitory effect of SOX7 on RUNX1 (Fig. 4E). HMG domain deletion and the R60E mutation also abrogated the inhibitory effects of SOX7 on RUNX1 and FLI1 transcriptional activity at the Nfe2 promoter region (Fig. 4F).

The fact that SOX7 can inhibit RUNX1 transcriptional activity suggests that the two proteins can physically interact. To investigate this, we performed co-immunoprecipitation assays using HEK293T cells co-transfected with SOX7- and RUNX1-expressing plasmids. The subsequent isolation of SOX7 revealed that RUNX1 could be immunoprecipitated with SOX7 (Fig. 5A). The reverse experiment, pulling down RUNX1 and detecting SOX7, also showed that both proteins could be immunoprecipitated together (Fig. 5A), demonstrating that SOX7 and RUNX1 can physically interact.

In order to investigate further the interaction between SOX7 and RUNX1, we used Duolink proximity ligation assay (PLA) as a more-sensitive and quantitative method of assessing protein interactions, where each Duolink puncta represents a single protein-protein interaction. HEK293T cells were co-transfected with RUNX1- and SOX7-expressing plasmids before Duolink analysis. Duolink puncta were detected in cells transfected with wild-type RUNX1 and SOX7 plasmids, confirming an interaction between the two proteins (Fig. 5B,C and Fig. S4). However, deletion of the HMG domain of SOX7 or deletion of the RUNT domain of RUNX1 substantially decreased the number of Duolink puncta per cell, indicating a decrease in the interaction between RUNX1 and SOX7 (Fig. 5B,C and Fig. S4). Collectively, these data demonstrate that SOX7 and RUNX1 physically interact via their respective HMG and RUNT domains.

The RUNT domain of RUNX1 provides key contacts for interacting with its co-factor CBFβ (Zhang et al., 2003), resulting in a RUNX1-CBFβ complex with much enhanced transcriptional activity. To test whether SOX7 can inhibit the interaction of RUNX1 with CBFβ, we performed a SOX7 competition assay using Duolink PLA. HEK293T cells were transfected with plasmids expressing FLAG-RUNX1 and HA-CBFβ, with or without untagged SOX7 competitor and a plasmid expressing GFP only (Fig. 6A). Transfected cells were enriched by sorting for GFP before Duolink PLA (Fig. 6A). Duolink puncta were detected between RUNX1 and CBFβ, as expected (Fig. 6B and Fig. S5). However, when untagged SOX7 was added at an equal concentration to RUNX1 and CBFβ, it resulted in a significant decrease in the number of Duolink puncta per cell between RUNX1 and CBFβ (Fig. 6B and Fig. S5A). Furthermore, the addition of untagged SOX7 at three times the concentration of RUNX1 and CBFβ completely abolished the interaction between RUNX1 and CBFβ (Fig. 6C and Fig. S5B). Deletion of the HMG domain of SOX7 significantly blunted the inhibitory effects of SOX7 on the interaction between RUNX1 and CBFβ, indicating that the HMG domain is important for this process (Fig. S6A,B). In the reverse experiment, CBFβ could outcompete SOX7 for RUNX1 binding, especially when CBFβ was added at three times the concentration of SOX7 and RUNX1 (Fig. S6C,D). These data therefore suggest that SOX7 and CBFβ compete for RUNX1 binding and that the balance between these transcription factors could be important for regulating HE differentiation.

As the RUNT domain is the DNA-binding domain of RUNX1, we investigated whether SOX7 can inhibit RUNX1 binding to its target DNA sites. FLK1⁺ cells were isolated from day 3.5 iSox7 EBs and reaggregated in HE differentiation media for 48 h. CD41⁺ hematopoietic progenitor cells were isolated and cultured for a further 12 h with or without doxycycline to induce SOX7 expression. RUNX1 ChIP and analysis of the Nfe2 promoter region identified enrichment for RUNX1 binding, which was significantly decreased upon SOX7 expression when the cells were cultured with doxycycline (Fig. 7A). To further investigate the effect of SOX7 on RUNX1 DNA binding, we performed a biotin-DNA-RUNX1 pull-down assay using nuclear extracts from iSox7 CD41⁺ cells cultured with or without doxycycline. The expression of SOX7 upon doxycycline addition resulted in a dramatic decrease in the amount of RUNX1 pulled-down by its DNA-binding site (Fig. 7B). Collectively, these data indicate that SOX7 inhibits the DNA-binding capacity of RUNX1.

In this study, we have dissected the mechanism by which SOX7 regulates the emergence of hematopoietic cells from HE in the yolk sac. We uncovered that SOX7 inhibits the expression of genes positively regulated by RUNX1 and TAL1, while having no effect on the expression of these key transcriptional regulators. Furthermore, we demonstrate that SOX7 can strongly inhibit the transcriptional activity of RUNX1 and transcriptional complexes containing RUNX1. The mechanism described herein for SOX7 therefore contrasts with that proposed for SOX17 in HE. Lizama and colleagues showed that SOX17 inhibits EHT by binding to the Runx1 locus and repressing Runx1 transcription (Lizama et al., 2015). It is interesting to speculate therefore that there is multi-layered regulation of RUNX1 in HE by SOX factors: SOX17 at the transcriptional level and SOX7 at the protein level. Another possibility, given the similarity of the HMG domain between the three SOXF factors (Bowles et al., 2000), is that SOX17 and SOX18 can also physically interact with RUNX1, although this remains to be proven.

Several studies have demonstrated that SOX17 expression during endothelial differentiation is restricted to arteries (Corada et al., 2013; Frame et al., 2015; Matsui et al., 2006). By contrast, SOX7 is expressed from the onset of endothelial differentiation in the yolk sac at E7.5, and is expressed in arterial and venous vasculature (Gandillet et al., 2009; Sakamoto et al., 2007; Zhou et al., 2015). In agreement with these data, we discovered that SOX7 is more broadly expressed across the yolk sac HE at E8.5 compared with SOX17. Therefore, it is likely that SOX7 plays a more important role than SOX17 in yolk sac HE, especially before and during vascular remodelling and in smaller blood vessels. As we also detect RUNX and SOX7 co-expressing cells in the AGM region of the dorsal aorta at E9.5, it is likely that both SOX7 and SOX17 regulate RUNX1 behaviour in this context. The role of SOX18 has not yet been investigated in HE; however, given the partial redundancy and compensation between the SOXF factors during endothelial development, it will be an interesting topic for future study.

Collectively, our data presented here and previous studies demonstrate that SOX7 is expressed in HE but is rapidly switched off during EHT (Costa et al., 2012). The expression of the RUNX1b isoform is low in HE but it increases during EHT before the subsequent rapid upregulation of RUNX1c in emerging hematopoietic progenitors (Sroczynska et al., 2009a; Swiers et al., 2013). RUNX1 is a master regulator of EHT: activating the expression of hematopoietic genes, while switching off the endothelial transcriptional program via activation of GFI1 expression (Lancrin et al., 2012; Lie-A-Ling et al., 2014; Tanaka et al., 2012; Thambyrajah et al., 2016). We propose that in the early stages of HE differentiation, SOX7 can outcompete CBFβ for RUNX1 binding, while also preventing RUNX1 binding to its hematopoietic target genes (Fig. 7C). As the endothelial transcriptional program is switched off, SOX7 expression decreases and RUNX1 is free to interact with CBFβ and to bind its target sites, resulting in switching on of the hematopoietic transcriptional program (Fig. 7C). In this study, we therefore propose a novel model where the balance of protein interactions between opposing endothelial and hematopoietic transcription factors fine-tunes HE differentiation and temporally regulates the emergence of hematopoietic progenitors.

The mechanism by which SOX7 regulates hematopoietic differentiation by interacting with RUNX1 is strikingly similar to the roles of SOX9 and RUNX2 in osteo-chondrogenic differentiation (Akiyama et al., 2005; Zhou et al., 2006). In this scenario, SOX9 inhibits RUNX2 DNA binding and RUNX2 transcriptional activity, which promotes chondrogenesis (Zhou et al., 2006). However, in progenitors expressing low levels of SOX9, RUNX2 is free to promote osteogenesis (Zhou et al., 2006). Consequently, the balance between SOX9 and RUNX2 regulates the cell fate decisions of osteo-chondrogenic progenitors. It will be interesting to investigate whether interplay between SOX and RUNX factors is an important developmental mechanism for regulating cell fate decisions during differentiation of other cell lineages.

ESCs were cultured and differentiated as previously described (Sroczynska et al., 2009b). Embryoid bodies (EBs) were routinely maintained up to day 3.5, and FLK1⁺ cells were isolated and cultured as previously described for HE differentiation (Fehling et al., 2003; Lancrin et al., 2009). The iSox7 cell line was generated as previous described (Costa et al., 2012), authenticated and regularly checked for contamination. Doxycycline was added to cultures at 1 µg/ml where indicated.

Differentiating HE cultures were replated into methylcellulose-based hematopoietic progenitor cell differentiation media as previously described (Gandillet et al., 2009; Sroczynska et al., 2009b). Doxycycline was added at 1 µg/ml where indicated; primitive erythroid colonies were scored on day 4 and definitive colonies were scored on day 8. At the end of the culture period, colonies were harvested and analysed for hematopoietic differentiation by flow cytometry.

E9.5 yolk sac explants were cultured for 2 days in explant medium adapted from Malik et al. (2013), containing: IMDM, 16% BIT serum substitute (Stem Cell Technologies), 50 µg/ml ascorbic acid (Sigma), 0.5% PFHMII (Invitrogen), 15% serum replacement (Invitrogen), 5% platelet-derived serum, 2 mM L-glutamine (Gibco), 150 µM MTG (Sigma), 2 U/ml EPO (Eprex, Janssen-Cilag) and 100 ng/ml SCF (Peprotech) (protocol kindly provided by Professor James Palis, University of Rochester Medical Center, Rochester, USA). For dissociated yolk sac cultures, E9.5 yolk sacs were dissociated with collagenase dispase (Roche) diluted to 1 mg/ml in PBS for 20 min at 37°C. Yolk sac cell suspensions were cultured for 4 days on OP9 cells in HE differentiation media previously described for AGM culture (Thambyrajah et al., 2016). Doxycycline was added to all yolk sac cultures at 1 µg/ml.

Targeted Sox7 ESC clone B9 (International Knockout Mouse Consortium) was injected into mouse blastocysts. Resultant chimaeras were crossed with C57BL/6 mice. Subsequent generations were crossed with PGK-Cre mice to excise the neomycin cassette and the critical exon 2 of the Sox7 gene, resulting in the generation of a LacZ-tagged null allele (Sox7^−/+). The Runx1b::rfp transgenic line was generated as previously described (Sroczynska et al., 2009a), except that the hCD4 reporter was replaced with RFP. The iSox7 transgenic line was generated as previously described (Gandillet et al., 2009).

Timed matings were set up between: male and female Sox7^−/+ mice, male Runx1^RFP/WT and female ICR^WT/WT mice and between male iSox7⁺/rtTA⁺ and female ICR^WT/WT mice. The morning of vaginal plug detection was embryonic day (E) 0.5. All animal work was performed under regulation governed by the UK Home Office Legislation under the Animal Scientific Procedures Act (ASPA) 1986.

HEK293T cells were seeded at 2×10⁶ cells/plate in a 100 mm plate containing IMDM 10% FBS media. The following day, cells were transfected using gene juice (Novagen) according to the manufacturer's instructions, with 2 µg of each of the pcDNA-transcription factor plasmids and 200 ng of pcDNA-GFP plasmid. After 48 h, GFP⁺ cells were sorted and 1×10⁴ cells were attached to microscope slides by cytospsin before fixation with 4% PFA and permeabilisation with 0.3% Triton X-100. Subsequent steps were performed using the Duolink PLA red detection kit, according to the manufacturer's instructions (Sigma). For competitive interaction assays, HEK293T cells were transfected either with equal amounts of each plasmid (1 µg each) or with the competitor plasmid (3 µg) at three times the concentration of the other plasmids (1 µg each). A GFP (200 ng)-expressing plasmid was used to enrich for transfected cells for both transfections, and the rest of the steps were carried out as above.

Duolink images were analysed using Image J software. The nuclear DAPI channel images were overlaid with binary Duolink puncta images and protein interactions were quantified by counting the number of Duolink puncta per nucleus for each cell in the field of view.

Embryos were stained as previously described (Yokomizo et al., 2013) using a goat anti-SOX7 antibody (1:200) (R&D systems; AF2766) and a donkey anti-goat AF647 antibody (1:2000) (Invitrogen; A-21447). Subsequently, embryos were incubated with a rat anti-CD31 antibody (1:1000) (BD Biosciences; 553370) and a rabbit anti-pan-RUNX antibody (1:1000) (Abcam; ab92336) before staining with a goat anti-rat AF488 and a goat anti-rabbit antibody (both 1:2000) (both Invitrogen). Yolk sacs were isolated and flat mounted with DAPI, as previously described (Frame et al., 2015), before imaging. All primary antibodies have been used in previous studies (Yokomizo et al., 2013; Yzaguirre and Speck, 2016; Zhou et al., 2015). Specific labelling of primary antibodies was determined by comparison with no primary antibody-treated controls.

All assays were performed using the Dual-Luciferase Reporter System (Promega) and luciferase activities were measured with the GloMax-Multi Detection System (Promega) according to the manufacturer's instructions. The osteocalcin 6×OSE2-luc RUNX luciferase reporter has been described previously (Zhou et al., 2006).

RNA was isolated from sorted FLK1⁺ cells cultured for 48 h in HE media before pulsing with or without 1 µg/ml of doxycycline for 6, 12 and 24 h. Indexed polyA libraries were prepared using 200 ng of total RNA and 14 cycles of amplification in the Agilent Sure Select Automated Strand Specific RNA library prep for Illumina Multiplexed sequencing (G9691-90030) on an Agilent Bravo-BenchCel-MiniHub Automated Liquid Handling Platform. Libraries were sized on an Agilent Bioanalyzer (G2940CA) using a High Sensitivity DNA Kit (5067-4626) and quantified by qPCR using a Kapa Library Quantification Kit for Illumina sequencing platforms (KK4835). Pooled libraries were sequenced at 1×75 bp on an Illumina NextSeq500 system (SY-415-1001).

Single end sequencing (75 bp) were aligned to the mouse reference genome GRCm38 (Ensembl v76) using Bowtie2 (version 2.2.1) with default settings. Lane-wise and technical replicates of the same samples were merged into one BAM file. The expression levels of 41,228 annotated features were determined using the feature Counts function from the Bioconductor package Rsubread (version 1.13.13).

The Bioconductor package edgeR (version 3.8.5) was used to identify genes that showed statistically significant variation in expressions level in the following doxycycline treatment. The data were filtered to include only genes with at least 1 count-per-million reads and differential expression analysis was performed using the function exactTest in edgeR. Genes differentially expressed 1.5-fold following doxycycline treatment (P-value <0.05, FDR<0.1) are shown.

Single RUNX1-RFP⁺/FLK1⁺/CD41⁻ cells were sorted from E8.5 RUNX1^RFP/WT embryos, then cDNA synthesis and specific target amplification (STA) was performed as previously described (Moignard et al., 2013). Single-cell qPCR was performed with 1 µl of STA mix (diluted 5 fold), 0.1 µM probes from the universal probe library (Roche), 0.2 µM primers (see Table S4) and 1× TaqMan master mix (Applied Biosystems). The number of cells expressing each SOXF factor above the limit of detection was calculated.

FLK1⁺/CD41⁻ HE enriched cells were sorted from SOX7^−/− and SOX7^+/+ E8.5 embryos and total RNA was isolated using an RNeasy Micro plus Kit (Qiagen) according to the manufacturer's instructions. cDNA synthesis and STA was performed using 1 µl of isolated RNA, and qPCR was performed as above using 1 µl STA mix (diluted 1000-fold). Gene expression was calculated relative to β-actin using the ΔΔCt method.

FLK1⁺ cells were sorted from iSox7 EBs, reaggregated and cultured for 2 days (as described previously) to generate HPCs (Goode et al., 2016). CD41⁺ cells were sorted and cultured with or without 1 µg/ml doxycycline for 12 h, before harvesting. Chromatin immunoprecipitation was performed using the HighCell# ChIP kit (Diagenode) according to the manufacturer's instructions. Briefly, cells were crosslinked for 8 min in 1% formaldehyde and the nuclei were then sonicated for 15 cycles (30 s on/30 s off) with the Bioruptor (Diagenode). Five-million cells were used for each immunoprecipitation with either the control immunoglobulin or the anti-RUNX1 antibody (4 µg) (Abcam). RUNX1 binding to the Nfe2 promoter was assessed by SYBR green qPCR using 0.25 µM primers (see Table S4) and 1×SYBR green mix. For data analysis, ChIP signals were normalised first to input then to IgG reference levels.

Biotin-DNA pull-down assays were performed as previously described (Costa et al., 2012), using a biotinylated 30 bp probe taken from the osteocalcin promoter containing a single RUNX-binding site (see Table S4 for sequence).

Cells were dissociated by trypsinisation, washed with IMDM media containing 10% FBS and stained with the following antibodies for 30 min on ice: CD41 PE/Cy7 (1:400), CD41 APC (1:400), CD45 PerCP-Cy5.5 (1:200), cKIT APC eflour780 (1:200), TIE2 PE (1:200) (all eBioscience) and FLK1-PE/Cy7 (1:300) (Biolegend). Cells were washed and analysed using an LSRII flow cytometer (BD) or sorted on an ARIAIII FACs machine (BD). Subsequent analysis was performed using FlowJo software V9 (Tree Star), gating first on the viable cells from the forward and side scatter.

Sample sizes were chosen based on pilot studies. Student's t-test was used to assess the differences between two populations in both cell line and embryo experiments. A hypergeometric probability test was used to assess the significance of overlap between two gene lists. Chi-squared test was used to assess frequency differences between single cell populations. *P<0.05, **P<0.01, ***P<0.001.

We thank Professor James Palis for provision of the yolk sac flat mounting and yolk sac explant culture protocols. We thank the staff at the Flow Cytometry, Advanced Imaging and Molecular Biology Core facilities of CRUK Manchester Institute for technical support.