Repression of arterial genes in hemogenic endothelium is sufficient for haematopoietic fate acquisition

Changes in cell fate and identity are essential for endothelial-to-haematopoietic transition (EHT), an embryonic process that generates the first adult populations of haematopoietic stem cells (HSCs) from hemogenic endothelial cells. Dissecting EHT regulation is a critical step towards the production of in vitro derived HSCs. Yet, we do not know how distinct endothelial and haematopoietic fates are parsed during the transition. Here we show that genes required for arterial identity function later to repress haematopoietic fate. Tissue-specific, temporally controlled, genetic loss of arterial genes (Sox17 and Notch1) during EHT results in increased production of haematopoietic cells due to loss of Sox17-mediated repression of haematopoietic transcription factors (Runx1 and Gata2). However, the increase in EHT can be abrogated by increased Notch signalling. These findings demonstrate that the endothelial haematopoietic fate switch is actively repressed in a population of endothelial cells, and that derepression of these programs augments haematopoietic output.

T he first haematopoietic stem cells (HSCs) emerge in the embryo from a specialized subset of endothelial cells (ECs), collectively termed as hemogenic endothelium (HE). The concept of endothelial-derived HSCs has broad clinical implications as it may open new avenues for in vitro blood production. However, the hemogenic capacity of the endothelium is transient and its precise regulation remains unknown. During a narrow developmental time period (approximately embryonic day (E)10-12 in the mouse 1,2 , and 4-6 weeks in the human 3 ), hemogenic ECs acquire cell morphology and gene expression consistent with haematopoietic identity, in a process called endothelial-to-haematopoietic transition (EHT) [4][5][6] . In the mammalian system, the 'hemogenic window' is short lived and typified by groups (or clusters) of rounded cells that are observed within the vascular wall. The haematopoietic cell clusters have been demonstrated to contain both haematopoietic stem and progenitor cells (HSPCs) 7,8 . Regions known to harbour HE include the aorta-gonado-mesonephros (AGM) region 1,[9][10][11][12] , vitelline and umbilical arteries 9,13,14 , yolk sac 15,16 , placenta 17,18 and others 19,20 , but generally encompass arterial vascular beds, as opposed to the veins or capillaries 21 .
Interestingly, regulators of arterial fate, including the transcription factor Sox17 (ref. 22) and Notch1 (ref. 23), are implicated in haematopoietic emergence from HE, as early loss of either results in haematopoietic defects 24,25 . Sox17 positively regulates Notch1 for both arterial fate acquisition and hemogenic endothelial specification 22,26 . How these arterial fate specifiers function in endothelial to haematopoietic conversion, separate from their role in artery-vein specification, is unclear.
Here we present data that demonstrates after artery-vein specification, Sox17 actively prevents the transition to haematopoietic fate by repression of key haematopoietic transcription factors, thereby maintaining endothelial identity. The loss of Sox17 promotes haematopoietic conversion, and its dynamic expression imparts a previously unappreciated, but critical step, in endothelial to haematopoietic cell fate transition.

Results
Haematopoietic clusters and endothelial gene expression. We first evaluated the expression patterns of Sox17, Notch1, Runx1 and Gata2 in the embryonic dorsal aorta (AGM) as all four factors are shown to be required for HSC emergence. The endothelium of this region can be identified by immunofluorescence of the pan-EC surface marker PECAM-1 (CD31), and HSPC clusters are easily apparent through their rounded morphology and shared endothelial marker expression ( Fig. 1a-d). RUNX1 (ref. 27) and GATA2 (ref. 28), two transcription factors known to be required for HSPC emergence from HE, are localized to HSPC clusters, as compared with the adjacent endothelium ( Fig. 1a,b,e). When known regulators of the arterial program including Notch signalling 23,29 (visualized by the TP1-Venus reporter mouse line 30,31 ) and SOX17 (ref. 22) are evaluated, immunofluorescence is localized to the endothelium and not the HSPC clusters ( Fig. 1c-e). The appearance of HSPC clusters along the aortic wall is coincident with changes in cell surface marker expression, as cluster cells acquire c-Kit (CD117) 7,32,33 and CD41 (refs 34,35) markers ( Supplementary Fig. 1a,b), in addition to maintaining endothelial markers CD31 and VE-cadherin (CD144) 36 (Fig. 1a-e). Eventually, HSPCs also acquire CD45, a pan-haematopoietic surface marker ( Supplementary Fig. 1c). Sox17 expression is largely undetectable in cluster cells, but rarely can be seen in a perinuclear pattern with co-expression of Golgi markers ( Supplementary Fig. 2d-f), suggesting that it no longer functions as a transcription factor in the cluster cell population. As arterial markers can be flow sensitive [37][38][39] , we also evaluated the expression patterns of SOX17 and RUNX1 in Mcl2a À / À circulation mutants 40 (Supplementary Fig. 1g), and found that the segregation of SOX17 immunofluorescence to the endothelium and RUNX1 to haematopoietic cell clusters is preserved. The differential expression of surface markers allows for separation of endothelial and haematopoietic populations, as well as HSPC clusters (CD31 þ CD117 þ ), by fluorescent-activated cell sorting (FACS) (Supplementary Fig. 1h,i). Transcriptional analyses of sorted populations demonstrate that endothelial subsets (CD31 þ CD117 À CD45 À ) exhibit lower Runx1 and Gata2 transcript levels when compared with HSPC cluster populations (CD31 þ CD117 þ CD45 À ), or as compared with differentiated haematopoietic cells (CD31 À CD45 þ ) ( Fig. 1f; Supplementary Table 1). In contrast, genes associated with arterial identity (Sox17 and Notch1) are decreased in HSPC clusters as compared with the endothelium. Sox17 (refs 26,41) and Notch1 (refs 24,42-45) are known to be important for hemogenic endothelial specification. Thus, the finding that their transcripts and protein levels are actually decreased in HSPC clusters is intriguing. As relatively small populations of primordial germ cells can express CD31 and CD117 (refs 32,46), we also evaluated populations based on CD41 expression and found that the same trend is observed when we identify hemogenic cluster cells with the marker CD41 (refs 34,35) ( Supplementary  Fig. 1b,i-j). Together the data suggest that endothelial to haematopoietic fate conversion may require downregulation of critical arterial genes.
Sox17 negatively regulates haematopoietic fate. To evaluate the impact of Sox17 on EHT, we undertook both loss-and gain-offunction approaches. In vivo endothelial genetic deletion of Sox17 during EHT (induction at E9.5, evaluation at E11; Fig. 2a) was evaluated using a endothelial-specific Cre recombinase (Cdh5(PAC)-CreERT2 (ref. 47)) mouse line crossed to a Sox17 floxed line 25 with a ROSA26Cre reporter 48 (RTom, tdtomato, Td þ ). The induction strategy is similar to that used in fate-tracing studies 49 and allows for timing of Sox17 endothelial recombination early in the hemogenic window and during EHT. Transcript analysis of sorted ECs after in vivo induction uncovered a significant increase in Runx1 and Gata2, two haematopoietic transcription factors known to be critical for HSC development during EHT 27,28,50 (Fig. 2a). Notch1 transcripts are also notably decreased (Fig. 2a), in agreement with previous studies that show Sox17 positively regulates the Notch pathway 22,26 . In addition, other members of the SoxF family (Sox7 and Sox18) were increased, possibly due to a compensatory response (Fig. 2a). There were no observed differences in endothelial labelling or cell number across homozygous Sox17 f/f , heterozygous Sox17 f/ þ , or control animals ( Supplementary Fig. 2a). Immunohistochemical analysis demonstrates the presence of HSPC clusters in the aorta with a marked decrease of endothelial SOX17 in Sox17 f/f mutants (Fig. 2b). Also, we did not observe any obvious changes in endothelial morphology as evaluated by scanning electron microscopy ( Supplementary Fig. 2b).
Currently, it is not possible to predict which specific EC within a hemogenic vascular bed will transition to a haematopoietic fate. Also, not known is whether ECs comprising the same hemogenic site are all capable of EHT. So whether the actual cell fate conversion is a stochastic event or a predetermined fate change remains to be seen. To circumvent the current obstacles of EHT prediction, we adopted a fate-tracing strategy 49 Table 2). Tamoxifen induction in vitro with the active metabolite 4-hydroxytamoxifen at E11.0 allows immediate ablation in AGM explants during EHT, and the calculation of a HE ratio, which we define as traced haematopoietic cells (HCs) compared with traced ECs. Using this assay to temporally and conditionally ablate Sox17, we demonstrate that timed loss of endothelial Sox17 promotes conversion to haematopoietic cell fate in situ (Fig. 2c-f). Sox17 f/f mutants exhibit a significant threefold increase in HE ratios indicating increased haematopoietic output, in addition to significantly increased labelled hemogenic cluster populations  immunofluorescence is noted in the cell nuclei of the endothelial layer (arrowheads), as compared with the associated cell cluster. CD31 in red, and DAPI in blue. (d) Notch pathway activation (green) as measured in the TP1 Venus mouse line is notable in the endothelial layer (arrow) but less so in the associated haematopoietic cell cluster, CD31 in red. DAPI in blue. (e) CD144 (red) labels the endothelium and haematopoietic cluster cells (arrowhead), Sox17 in grey, and Runx1 in green. (f) Embryos at E10.5 were sorted based on cell surface markers to isolate endothelial cells (CD31 þ CD117 À CD45 À ), haematopoietic cluster cells (CD31 þ CD117 þ CD45 À ), maturing cluster cells and HSPCs (CD31 þ CD117 þ CD45 þ ) and mature haematopoietic cells (CD31 À CD45 þ ). Bar graphs depict transcript expression (RT-PCR) in each subgroup for Runx1, Gata2, Sox17, Notch1 and Cdh5 (CD144). Differing letters represent significance between groups where a versus b, or b versus c, or a versus c, is significant to a P valueo0.01 or less, n ¼ 3 litters, 24 embryos.
(CD31 þ CD41 þ Td þ ) and maturing HSPC populations Fig. 2c-f). The observed increase in HE ratios and HSPC number is not due to proliferation effects (as measured by BrdU incorporation; Supplementary Fig. 2d) nor is the higher HE ratio due to changes in cell death (Annexin-V staining; Supplementary Fig. 2d). We also observe increases in other haematopoietic populations Supplementary Fig. 2e). In addition, when a similar strategy is applied to earlier explants (E9.5) before haematopoietic cell cluster emergence, we observe similar trends in the HE ratio ( Supplementary Fig. 2f). So while Sox17 has been shown to be critical for HE specification before EHT, the loss of Sox17 actually promotes haematopoietic fate over endothelial fate during EHT. To further evaluate the role of   Table 2). (d) The HE ratio of Sox17 homozygous (f/f) and heterozygous Sox17 in this process, we undertook gain-of-function studies in wild-type AGM explants using adenoviral-mediated overexpression of human SOX17 (AdhSOX17-GFP; Fig. 2g). Green fluorescent protein (GFP) expression in explants overlapped with SOX17 co-staining (Fig. 2h), allowing for cell sorting of AGM ECs (CD31 þ ) that were either successfully infected (GFP þ ) or not infected (GFP À ) by AdhSOX17-GFP (Fig. 2i). Transcript analysis of ECs with SOX17 overexpression demonstrates significant increases in Sox17 and Notch1 transcripts with significant reduction in Runx1, Gata2, Sox7 and Sox18 transcripts (Fig. 2j). The data altogether suggest that Sox17 negatively regulates haematopoietic fate through repression of Runx1 and Gata2. We also show the known positive regulation of Notch1 by Sox17, and regulation of other SoxF family members, Sox7 and Sox18.
Sox17 represses Runx1 and Gata2. To determine whether the observed changes in Runx1 and Gata2 were due to regulation by SOX17, chromatin immunoprecipitation (ChIP) was carried out in sorted ECs at E11 (Fig. 3a), as well as in human umbilical arterial EC lines, HUAECs (Supplementary Fig. 3a; Supplementary Table 3). Two predicted SOX17-binding sites upstream of Runx1 and Gata2 5 0 -untranslated regions showed significant enrichment (Fig. 3a). To demonstrate whether SOX17 was capable of direct DNA binding of specific sequences in vitro, electrophoretic mobility shift assays (EMSAs) were conducted for sites with high species homology between human and mouse (Supplementary Fig. 3b; Supplementary Table 4). Specific areas within ChIP-enriched regions were capable of outcompeting the known SOX17 regulatory site in the Lef1 promoter 51 (Fig. 3b). We further analysed the regulation of Runx1 and Gata2 using d e E10.5 Sox17 Fold mRNA change Fold mRNA change IgG Sox17 IgG Sox17 Fold enrichment Fold enrichment luciferase reporter assays, which demonstrate derepression of both Runx1 and Gata2 activity after Sox17 short interfering RNA (siRNA) knockdown (Fig. 3c). In vivo loss of Sox17 demonstrates intact haematopoietic clusters with normal localization of RUNX1 and GATA2 expression (Fig. 3d,e). To investigate how Sox17 may regulate Runx1 and Gata2 in mature endothelium in the human system, we conducted in vitro gain-and loss-offunction experiments. SOX17 siRNA inhibition of human umbilical arterial cell lines resulted in significantly elevated RUNX1 transcripts, at similar levels to the control LEF1 (ref. 51), a SOX17 repressive target (Fig. 3f). In addition, genes important in arterial and venous identity are altered with decreased arterial gene transcripts (DLL4) 52,53 and elevated transcript levels of COUP-TFII, an important determinant of venous fate 21 (Fig. 3f). In contrast, when SOX17 is overexpressed after adenoviral infection, RUNX1 and GATA2 transcript levels are significantly decreased (Fig. 3g). SOX17 overexpression also altered levels of DLL4 (increased) and COUP-TFII (decreased) (Fig. 3g). The data suggest a novel role of Sox17 as a repressor of haematopoietic fate, while confirming Sox17 as a pro-arterial fate regulator.
Intersecting roles of Sox17, Runx1 and the Notch pathway. As Sox17 was previously shown to promote arterial identity upstream of the Notch pathway 22 , we evaluated SOX17 regulation of Notch pathway members in our system. SOX17 ChIP demonstrates enriched occupancy upstream of the Notch1 5 0 -untranslated region, and of the Notch ligand Dll4 (Fig. 4a). In addition, we also observe occupancy upstream of Coup-TFII, which has not been previously described (Fig. 4a). Similar enrichment of these sites was observed in HUAECs ( Supplementary Fig. 4a). We further validated direct binding of SOX17 within the enriched ChIP sites via EMSA, and demonstrated multiple SOX17-binding sites are capable of outcompeting Lef1 controls ( Fig. 4b; Supplementary Fig. 4b,c). To understand whether Notch1, a putative downstream target of SOX17, also plays a repressive role in EHT, we evaluated Notch1 loss of function. Similar to Sox17, loss of Notch1 in AGM explants increased the HE ratio, as well as populations of HSPCs (Fig. 4c-f; Supplementary Fig. 4d; Supplementary Table 5). We also observed increased HE ratios after AGM explants were exposed the g-secretase inhibitor DAPT (Supplementary Fig. 4e). However, when BrdU incorporation was evaluated in Notch1 mutant explants, significantly higher levels of incorporation occurred in the haematopoietic compartment (Fig. 4g), suggesting the observed changes may be due to haematopoietic cell proliferation, and not due to an increase in EHT. Annexin-V levels were not notably changed ( Supplementary Fig. 4f,g).   loss of Notch1 (induction at E9.5) demonstrates expected changes in arterial and venous identity genes (EfnB2 and EphB4) 23 within sorted ECs (Fig. 4h). No changes in Runx1 transcripts were noted, while expectedly Hes1 transcripts were decreased (Fig. 4h). There were no observed differences in endothelial labelling or cell number across homozygous Notch1 f/f , heterozygous Notch1 f/ þ or control animals ( Supplementary Fig. 4h). Interestingly, we also noted expected changes in endothelial morphology 54 ( Supplementary Fig. 4i).
To understand the role of Notch1 signalling in the context of Sox17 loss, we bred R26RNotch1IC-nEGFP lines 55 ( þ mNICD-GFP) that overexpress the Notch1 intracellular domain (NICD) upon Cre activation into our temporal endothelial-specific Sox17 loss-of-function models (Fig. 5a). Increased Notch activation in E11.0 AGM explants was capable of abrogating the observed EHT increase in Sox17 mutants (Fig. 5b) with normal appearing HSPC clusters in vivo after induction of Sox17 loss and NICD overexpression at E9.5 (Fig. 5c). Thus, the conversion to haematopoietic fate in HE requires loss of arterial identity programs in addition to derepression of haematopoietic genes by SOX17. While our data have shown the regulation of Runx1 by SOX17, previous reports suggest that RUNX1 may directly bind and repress Sox17 (ref. 56). To evaluate whether there may be bidirectional regulation in the endothelium, we performed RUNX1 ChIP of conserved sites upstream of Sox17 transcriptional start sites and found multiple areas of enrichment ( Fig. 5d; Supplementary Fig. 4j,k). Adenoviral overexpression of RUNX1 (AdhRUNX1-GFP) in HUAECs demonstrates decreased SOX17 and SOX18 transcripts (Fig. 5e). Overall, the data present a complex regulatory network for the maintenance of EC fate and the conversion to a haematopoietic fate (Fig. 5f). Once haematopoietic fate is achieved, both Sox17 and Notch1 have known roles in haematopoietic cell survival 25 and lineage differentiation 57 , which is also evident in our haematopoietic colony assay evaluation of mutant haematopoietic cells ( Supplementary Fig. 5).

Discussion
An important obstacle in recapitulating HE in culture for in vitro blood production is identification of possible activators and silencers of the hemogenic program. Here we demonstrate important altering requirements for Sox17 and Notch1, which highlights the refinements needed for translational models recapitulating EHT. Previous studies have identified Runx1 (ref. 27), Gata2 (ref. 28), Notch1 (refs 24,58) and Sox17 (refs 25,26) as critical for EHT. However, dissecting the contributions of these pathways to vascular development versus the process of haematopoietic emergence from the endothelium has not been previously reported. Notch1 and Sox17 both have important roles in arterial specification 22,23,43 . As the major vessels that harbour HE are arterial sites 9,13 , it may be that arterial identity is a prerequisite to hemogenic endothelial activity. However, hemogenic activity also occurs in yolk sac and placental vascular beds that are not overtly arterial [15][16][17][18] . In addition, recent evidence in human ESC cultures suggest that while hemogenic ECs incorporate into arterial vascular walls, they have differential surface marker expression profiles than arterial cells 59 . There is also evidence that arterial identity can be uncoupled from hemogenic capacity 58,60,61 . So it may be that hemogenic endothelial specification requires the same pathways mobilized in the acquisition of arterial identity, but not arterial identity per se. However, for the direct transition to haematopoietic fate, the expression levels of arterial/hemogenic specifiers need to be reduced. The complex temporal requirements, elucidated here, explains previous data where continued or overexpression of Sox17 was noted to prevent haematopoietic conversion in culture 26,62 . In addition, the reciprocal repression of Sox17 by RUNX1 introduces another unique aspect of fate determination where once endothelial Sox17 levels decrease, Runx1 levels can rapidly rise during the fate switch, and together they function as a classical bistable system; similar to those described in mesodermal progenitors 63 . Last, the data also demonstrate that the EHT program can be manipulated for increased haematopoietic output, suggesting that hemogenic EC number may not be a fixed entity. If EHT is not restricted to a fixed number of ECs within a hemogenic vascular compartment, but instead occurs as a more global transient stochastic process of developing endothelium, it allows for the possibility of endothelial expansion for HSC production.
Flow cytometric analyses and cell sorting. Whole embryos or AGMs underwent mechanical dissociation by pipetting to single cell suspension in Hank's Balanced Salt Solution with 2% fetal bovine serum, 1% penicillin/streptomycin and 10 mM HEPES, pH 7.2 (ref. 65) and stained for 30 min at 4°C with agitation. Single cell suspensions were sorted in a BD FACS Aria III. Flow cytometric analyses were performed on a FACS Verse or FACS Aria III using the FACSDiva 8.0 software (BD Biosciences) and data analysed using the FlowJo v10.0.7 (Tree Star). Gating strategy in Supplementary  Figs 1h,i, and 2c, see Supplementary Table 7 for a list of antibodies.
Real-time RT-PCR expression analysis. For in vivo transcriptional characterization of the induced endothelium, lineage traced CD31-APC þ ,CD41-FITC À , CD45-FITC À DAPI-excluded cells were sorted (for full list of antibodies please see Supplementary   ) was used to perform ChIP in HUAECs. The magnetic bead coated by the antibody was washed (PBS, 0.1% Triton X-100) then incubated with the precleared sample at 4°C overnight. The precipitates were washed, and the chromatin complexes were eluted. After reversal of cross-linking (65°C for 4 h), the DNA was purified using QIAquick PCR purification kit (Qiagen, 28104) and 100 pg was used as a template in each quantitative PCR reaction for quantitative analysis. Oligonucleotides used in PCR for quantitative ChIP are listed in Supplementary Table 3. Antibody dilutions are listed in Supplementary Table 7.
Non-radioactive electrophoretic mobility shift assay. Recombinant SOX17-Flag and Flag alone (pcDNA3 vector (Promega)) were expressed in 293T cells. Plasmids were transfected using Lipofectamine 2000 Transfection Reagent (Life Technologies, 11668019) 36 h before cells were lysed in RIPA buffer containing protease inhibitors. Recombinant protein was immunoprecipitated from lysate overnight at 4°C with Anti-FLAG M2 magnetic beads (Sigma, M8823) and the recombinant protein eluted with excess FLAG peptide. 5-7 ml of the first eluate was used in a binding reaction along with 0.3 pmol of complementary annealed 3 0 Biotin-labelled oligonucleotides (Integrated DNA Technologies), 300-fold excess competitor probes, 0.02U Poly(dG-dC) (Sigma, P9389) and binding buffer (100 mM HEPES pH 8.0, 50 mM KCl, 500 mM dithiothreitol, 50 mM EDTA, 1 mM MgCl 2 and 5% glycerol by volume) 66 . DNA-protein complexes were resolved on 7% native polyacrylamide gel, transferred to neutrally charged nylon membrane, incubated with Streptavidin-POD (Roche, 11089153001) and imaged by chemiluminescence. See Supplementary Table 4 for probe sequences.
Luciferase reporter assay. Putative regulatory sequences (700-850 bp) including Sox17 ChIP-enriched regions and EMSA-competent SOX17-binding sites were synthesized and cloned (Integrated DNA Technologies) based on UCSC genome browser murine sequences (see Supplementary Methods for fragment sequences). The fragments were amplified by PCR (Phusion, New England Biolabs) with appropriate linkers. The pGL4-TK vector (pGL4.54, Promega), containing the gene encoding Firefly luciferase driven by a TK minimal promoter, was digested using kpnI restriction enzyme (New England Biolabs) and mung bean nuclease (New England Biolabs) followed by ligation using Gibson Assembly Master mix (New England Biolabs) and confirmatory sequencing. C166 murine yolk sac ECs (30,000, ATCC, CRL-2581) were reverse cotransfected with 400 ng of reporter vector along with 10 ng of a Renilla luciferase transfection control plasmid (pRL, Promega) and 30 pmol of a Sox17-targeted or non-targeted 'scramble' siRNA pool (ON-TAR-GETplus siRNA SMARTpool, GE Dharmacon) using Lipofectamine 3000 (Life Technologies) according to manufacturer's recommendations. After 48 h of culture, cells were lysed and luciferase activity assessed using the Dual-Luciferase Reporter Assay System reagents (Promega) in a GloMax 96 Microplate Luminometer with dual injectors. In technical triplicate, relative luciferase activity was calculated by dividing Firefly readings by Renilla readings for each well and then normalized according to baseline values for each treatment condition after transfection of pGL4-TK without a fragment added.
Statistical analyses. Student's t-test, one-way and two-way analysis of variance analyses were performed as indicated in all experiments where nZ3 unless otherwise noted. Mean and standard error were calculated and graphed using GraphPad Prism 6 software. All statistical measurements are listed in Supplementary Tables 2 and 5.