Murine AGM single-cell profiling identifies a continuum of hemogenic endothelium differentiation marked by ACE

: In vitro generation and expansion of hematopoietic stem cells (HSCs) holds great promise for the treatment of any ailment that relies on bone marrow or blood transplantation. To achieve this, it is essential to resolve the molecular and cellular pathways that govern HSC formation in the embryo. HSCs first emerge in the aorta-gonad-mesonephros region (AGM) where a rare subset of endothelial cells, hemogenic endothelium (HE), undergoes an endothelial-to-hematopoietic transition (EHT). Here, we present full-length single-cell-RNA-sequencing of the EHT process with a focus on HE and dorsal aorta niche cells. By using Runx1b and Gfi1/1b transgenic reporter mouse models to isolate HE, we uncovered that the pre-HE to HE continuum is specifically marked by Angiotensin-I converting enzyme (ACE) expression. We established that HE cells begin to enter the cell cycle near the time of EHT initiation when their morphology still resembles endothelial cells. We further demonstrated that RUNX1 AGM niche cells consist of smooth muscle cells and PDGFRa + mesenchymal cells and can functionally support Overall, our study provides new insights into HE differentiation towards HSC and the role cells this Our expansive scRNA-seq represents powerful Abstract In vitro generation and expansion of hematopoietic stem cells (HSCs) holds great promise for the treatment of any ailment that relies on bone marrow or blood transplantation. To achieve this, it is essential to resolve the molecular and cellular pathways that govern HSC formation in the embryo. HSCs first emerge in the aorta-gonad-mesonephros region (AGM) where a rare subset of endothelial cells, hemogenic endothelium (HE), undergoes an endothelial-to-hematopoietic transition (EHT). Here, we present full-length single-cell-RNA-sequencing of the EHT process with a focus on HE and dorsal aorta niche cells. By using Runx1b and Gfi1/1b transgenic reporter mouse models to isolate HE, we uncovered that the pre-HE to HE continuum is specifically marked by Angiotensin-I converting enzyme (ACE) expression. We established that HE cells begin to enter the cell cycle near the time of EHT initiation when their morphology still resembles endothelial cells. We further demonstrated that RUNX1 AGM niche cells consist of vascular smooth muscle cells and PDGFRa + mesenchymal cells and can functionally support hematopoiesis. Overall, our study provides new insights into HE differentiation towards HSC and the role of AGM RUNX1+ niche cells in this process. Our expansive scRNA-seq datasets represents a powerful resource to investigate these processes further. can a role in DA EHT. these reveal the identity of as muscle cells and


Introduction
Hematopoietic stem cells (HSCs) sit at the apex of the blood system and are powerful treatment modalities for cancer and blood malignancies 1,2 . Understanding the molecular programs underpinning their formation during embryogenesis is critical for the development of efficient protocols to generate and amplify HSCs in vitro. In mice, the first transplantable HSCs are generated intra-embryonically in the region of the aorta-gonad-mesonephros (AGM) between embryonic day (E) E10.5 to E11.5 [3][4][5] . Within this narrow time window, specialized endothelial cells lining the dorsal aorta (DA), known as hemogenic endothelium (HE), transition to non-adherent hematopoietic cells via a process termed the endothelial-to-hematopoietic transition (EHT) [6][7][8][9][10] . Morphologically, the EHT produces intra-aortic hematopoietic clusters (IAHC) protruding from the endothelium into the lumen of the DA and other major arteries [11][12][13] .
Although HE is established as the cellular source of the first blood cells in vivo and in vitro 8,10,14 , our knowledge of the molecular and cellular mechanisms regulating HSC emergence from HE remains incomplete.
The transcription factor RUNX1 controls the initiation and completion of EHT and is essential for establishing definitive hematopoiesis [15][16][17][18][19] . Downstream RUNX1 target genes Gfi1/Gfi1b are critical regulators of EHT that recruit histone-modifying complexes to silence the endothelial program in HE 20,21 . The sequential expression of the transcriptional repressors GFI1/1B marks distinct stages of the EHT. GFI1 is expressed in HE, while GFI1B is mainly found in IAHC 21 . In contrast, RUNX1 expression is detected during all stages of EHT 22,23 . Additionally, RUNX1 expression is also found in subaortic mesenchyme 5,22,24 . The exact identity and the potential role of these mesenchymal RUNX1 + cells are currently unknown.
Single-cell RNA sequencing (scRNA-seq) is a powerful tool to profile developmental pathways. Although this can now be done globally on whole organs and organisms 25-29 , understanding pathways involving rare cell populations still greatly benefits from targeted approaches using enriched cell populations. We and others have previously taken the latter Downloaded from http://ashpublications.org/blood/article-pdf/doi/10.1182/blood.2020007885/1822187/blood.2020007885.pdf by guest on 16 September 2021 approach to start resolving molecular events leading up to HSC formation in the murine AGM [30][31][32][33][34][35] . Much emphasis has been on the final steps of HSC commitment within the IAHC 11,31,33,35 .
However, the intricacies of DA HE differentiation before the initiation of EHT are not well defined. Reasons for this are the rarity of the HE population but also a lack of suitable cell surface markers for HE purification.
To gain insight into how DA HE progresses towards EHT, we used two transgenic reporter mouse models (Runx1b RFP and Gfi1 Tomato /Gfi1b GFP ) 20,21,23,36,37 to isolate and profile a wide variety of phenotypic HE. We reasoned that targeted scRNA-seq of Runx1b:RFP + and Gfi1:Tomato + phenotypic HE populations is essential to profile them in-depth as they represent only a small fraction of the total CDH5 + endothelial populations in the AGM (approximately 6% and 0.6% respectively, Supplemental Figure 1). We also profiled Runx1b:RFP + AGM mesenchymal cells. To maximize data recovery we adopted a full-length scRNA-seq protocol 38,39 . The resulting dataset contains nearly 1,200 FACS sorted cells, covering nine E10.5 mouse AGM populations.
Our data capture a detailed HE differentiation continuum, covering pre-HE and HE stages. We found that this continuum is marked by Angiotensin-I converting enzyme (ACE) expression and that pre-HE and HE can be discerned based on cell cycle status. Additionally, we establish that the rare Runx1b:RFP + sub-aortic mesenchymal population supports hematopoiesis and consists of smooth muscle and PDGFRa + cells. This high-resolution single-cell data set covering HE and its surrounding niche in the mouse DA can be queried at http://shiny.cruk.manchester.ac.uk/AGM_scRNA/.

Mouse embryos
Mouse work was performed following the U.K. Animal Scientific Procedures Act 1986 and was approved by the Animal Welfare and Ethics Review Body of the CRUK-MI. Gfi1 GFP /Gfi1b GFP , Gfi1 Tomato /Gfi1b GFP and Runx1b RFP lines have been described previously 20,21,23,36 . Embryos were genotyped and processed as previously described 20,21,23 . Genotyping primers are listed in Supplemental Table 1.

Zebrafish experiments
Zebrafish work was performed by Azelead (France) following the E.U. guidelines for handling laboratory animals and was approved by the Animal Care and Use Committee (APAFIS#21063).
Antibodies are listed in Supplemental Table 1.

Ex-vivo AGM assays
Reaggregation experiments: AGM cells were FACS depleted of the target population and reaggregated for 4 days in hanging drop culture as previously described 21 . Aggregates were trypsinized and analyzed in colony-forming unit assays (CFU-assays) as previously described 43 .
Co-culture experiments: FACS sorted AGM cells were co-cultured with vascular-niche-cells for 10 days as previously described 21,44,45 .
Single-cell RNA sequencing and data processing Cells were sorted into lysis buffer and snap-frozen. Libraries were prepared using a modified Smart-Seq2 protocol described previously 46

Immunofluorescence
Embryos fixed in 4% formaldehyde were embedded in paraffin or snap-frozen. Staining and processing of frozen sections has been described previously 21,50 . Staining of paraffin-embedded sections was performed on the Leica BOND RX as previously described 51

Separation of E10.5 AGM cells into transcriptionally defined clusters
Having established that E10.5 FACS-HE Gfi1-het possesses a strong hemogenic transcriptional profile, we next sought to create a detailed scRNA-seq dataset representing early EHT in the E10.5 DA. We focused on capturing different stages of HE differentiation. Therefore, we expanded our E10.5 panel of FACS populations with phenotypic endothelial and HE cells from both heterozygote and homozygote Runx1b RFP as well as homozygote  In HE, the Runx1b isoform precedes the expression of Gfi1 21 and both genes are essential for the initiation of EHT 8,15,21 . To illustrate this developmental relationship between RUNX1 and GFI1, we isolated phenotypic Runx1b:RFP + Gfi1:GFP -E10.5 AGM HE (CD41 -CD45 -TER119 -CDH5 + KIT -Runx1b:RFP + Gfi1:GFP -) from Runx1b RFP /Gfi1 GFP reporter embryos. Following 24h of aggregation culture the Runx1b:RFP + Gfi1:GFP -HE gave rise to Runx1b:RFP + Gfi1:GFP + HE and progenitors (Supplemental Figure 6).

Trajectory analysis of E10.5 EHT uncovers a detailed HE maturation continuum
To identify changes associated with HE differentiation, we focused on clusters C4-C8, which cover the continuum from pre-HE cells to IAHC. Unlike the Runx1b KO Figure 12A). Although the pre-HE clusters (C4 & C5) did not cluster together with the endothelial clusters (C1-C3) they still displayed robust arterial gene expression (Vwf, Bmx, Hey2 and Jag1) (Supplemental Figure 8C). This is expected as the cells in these clusters do not express significant amounts of Gfi1, which is instrumental for initiating the downregulation of the endothelial program 20,21 . In both Runx1 KO and Gfi1/1b KO mice, IAHC are absent 21,22,59 and development arrests before EHT (C7). This is illustrated by the absence of KIT + cells in E10.5 AGMs (Supplemental Figures 12B-C). Interestingly, the KO clusters (C4:pre-HE KO , C6:HE KO ) sit at different points in the pseudotime trajectory, illustrating that they are arrested at different pre-EHT stages. C4:pre-HE KO sits within the EHT-continuum, while C6:HE KO seems to divert away from this trajectory ( Figure 3A, Supplemental Figure 12D).
Altogether these data show that we captured a detailed HE differentiation trajectory.  Table 6). Pathways associated with AE identity were enriched in C5:pre-HE. These included integrin and cadherin signaling, which are important for cell-cell and cell-matrix interactions 67,68 , and Notch and Wnt signaling, which both need to be downregulated to drive EHT forward [69][70][71] . Strikingly, the upregulated genes in C6:HE were dominated by cell cycle-related pathways. We, therefore, analyzed the cell cycle status of all clusters (C4-C8), including the KO cells. In line with previous findings, most C8:IAHC cells were in the G2/M or S phase 72 ( Figure 3D, Supplemental Figure 13C). However, an increase in cycling cells could already be detected in C6:HE. Interestingly, C6:HE KO phenocopies its heterozygous counterpart C6:HE with regards to cell cycle status ( Figure 3D). We confirmed these results by immunofluorescence analysis of CCNB1 (marking G2/M) 73 and KI67 (marking G1, S, G2 and M) 74 in E10.5 Gfi1-reporter embryos. CD31 + Gfi1:Tomato + expressed higher levels of CCNB1 and KI67 relative to CD31 + Gfi1:Tomato -AE in the DA (Figures 3E-H). Similarly, we observed increased entry into the cell cycle at the late stages of HE differentiation upon in vitro hematopoietic differentiation of FUCCI (fluorescent, ubiquitination-based cell cycle indicator) mouse embryonic stem cells 75 (Supplemental Figure 13D). Together, these results indicate that as HE cells get closer to EHT, they start downregulating their arterial identity while entering the cell cycle. The latter does not appear to rely on GFI1.

Ace: a novel HE cell surface (CS) enrichment marker
Currently, HE isolation is heavily reliant on transgenic reporter models 21,23,33,35,[76][77][78] . To identify HE CS markers, for use in a wild-type context, we compared each cluster along the EHT trajectory to C3:ENDO-AE (Supplemental Table 7). The resulting list of surface markers was screened for no/low expression in C1:ENDO-VE and antibody availability.
Neurl3 expression was observed in C6:HE-C8:IAHC, while CD27 and Flt3/CD135 were highly expressed in C8:IAHC. Neurl3 is a recently identified HE/pre-HSC marker 33 , CD27 has been shown to mark pre-HSC/HSC 79 and Flt3 marks early hematopoietic progenitors 80,81 . CD44, Procr/CD201 and the Angiotensin-I converting enzyme Ace/CD143 were highly expressed in the pre-HE and HE populations ( Figure 4A). Both CD44 and Procr, were expressed throughout the AGM CDH5 + population from C1:ENDO-VE to C8:IAHC. In contrast, Ace was mainly restricted to the pre-HE to HE continuum (C4-C6) in particular to the pre-HE clusters (C4-C5). The Runx1b KO (C4:pre-HE KO ) and Gfi1/1b KO HE (C6:HE KO ) showed comparable Ace expression levels to their EHT competent counterparts (C5:pre-HE and C6:HE). Unlike CD44 82 and PROCR 31 , ACE has not been previously identified as a murine HE marker, therefore we analyzed it further.
Immunofluorescence for ACE on E10.5 AGM sections correlated well with the scRNA-seq data.
ACE expression was exclusively seen in CD31 + DA cells within the AGM (Figure 4B), including the RUNX1 + and Gfi1:Tomato + subsets ( Figure 4C, Supplemental Figure 14A) embedded within the CD31 + lining (likely HE). In contrast, CD31 + cells budding into the DA lumen (likely IAHC) showed reduced/absent ACE staining. Overall, we found that on average 27% of the CD31 + cells within the E10.5 DA lining co-expressed ACE ( Figure 4B). We observed a similar ACE expression pattern in the E10.5 vitelline artery (Supplemental Figure 14B), another known site of hematopoietic emergence 13,83 . Next, we assessed the ability of ACE to enrich for Gfi1:Tomato + AGM HE cells from the Gfi1 Tomato /Gfi1b GFP transgenic reporter. In line with the scRNA-seq, ACE marked a smaller subpopulation of the LIN -KIT -CDH5 + cells compared to CD44 (Supplemental Figures 15A-C). Consequently, gating on LIN -KIT -CDH5 + ACE + cells resulted in an improved enrichment of Gfi1:GFP + cells compared to both LIN -KIT -CDH5 + and LIN -KIT -CDH5 + CD44 + gating ( Figure 4D).

RUNX1 + mesenchymal cells support AGM hematopoiesis
Currently, little is known about the subpopulation of mesenchymal cells in the AGM that express the hematopoietic master regulator Runx1. To determine their potential role in hematopoiesis, we performed aggregation cultures and CFU-assays with FACS sorted CD45 -TER119 -CDH5depleted, or not, of Runx1b:RFP + E10.5 AGM cells ( Figure 5A). Depletion of Runx1b:RFP + mesenchymal cells resulted in an over 5-fold decrease in CFU potential (n=7 embryos/group). These results indicate that Runx1b:RFP + mesenchyme positively influences hematopoiesis.

RUNX1 + sub-aortic cells contribute to the smooth muscle lineage and PDGFRa mesenchyme
ScRNA-seq analysis revealed two distinct Runx1b:RFP + mesenchymal clusters (C9:MES-PDGFRa, C10:MES-SMA, Figures 2C-E), which contributed equally to the total Runx1 + mesenchymal population (Supplemental Figure 16A). Upon Runx1b KO, the distribution shifted towards C10 (Supplemental Figure 16A). Index FACS sorting and scRNA-seq data revealed that in C9:MES-PDGFRa, the Runx1b RFP locus is less active than in C10:MES-SMA (Supplemental  Figure 16G) immunofluorescence showed that RUNX1 + smooth muscle cells were predominantly detected on the ventral side of the DA below the CD31 + endothelium. We also observed a thickening of the SMA and RFP layer around the DA in Runx1b KO sections (Supplemental Figure 17A-B). GSEA revealed ribosomal transcripts to be enriched in C10 Runx1b-KO versus C10 Runx1b-het (Supplemental Figure 17C). Subsequent screening of publicly available RUNX1 binding data [84][85][86][87][88] suggests that at least some ribosomal genes (11/66) can be directly bound by RUNX1 (Supplemental Table 10), as previously reported [89][90][91] . Unbiased clustering on C10 revealed two sub-clusters (C10-A and C10-B, Supplemental Figure 17D). C10-A consisted of both Runx1b het (41%) and Runx1b KO (59%) cells, while C10-B predominantly  Figure 17F). However, we did observe that Ptn (Pleiotrophin) expression was not only significantly upregulated in C9 as a whole, but the highest expression was seen in C9 Runx1bhet ( Figure 5F). RUNX1 binding analysis using published data 84-88 revealed the Ptn locus as a potential direct RUNX1 target (Supplemental Table 11). PTN has been implicated in HSC selfrenewal and retention in the bone marrow vascular niche 92 . To determine if PTN can play a role during AGM hematopoietic ontogeny, we performed ptn knockdown (KD) experiments in developing zebrafish. ptn KD resulted in a substantial decrease of emerging hematopoietic cells (kdrl + cd41 + ) in the DA at the 35hpf stage followed by a significant reduction of HSPCs in the caudal hematopoietic tissue (CHT) at 55hpf ( Figure 5G-H, Supplemental Figure 18). This indicates that PTN can play a role in the DA EHT. Together, these results reveal the identity of the RUNX1 + subaortic cells as smooth muscle cells and PDGFRa mesenchyme.

Discussion
Unraveling the molecular programs driving HSC formation during embryogenesis would be invaluable for developing protocols supporting in vitro generation and amplification of HSC to treat blood malignancies 1,2 . Here we present a high-resolution scRNA-seq dataset that encompasses sequential stages of blood formation in the mouse AGM region at the time of HSC formation. We focused on in-depth profiling of HE development and RUNX1B mesenchyme by analyzing FACS sorted populations from Runx1 and Gfi1/1b reporter mouse models.
By combining full-length scRNA-seq with the functional and phenotypic characterization of the sequenced FACS populations, we were able to precisely stage E10.5 pre-HE and HE populations within the EHT continuum. These assignments differ from those previously published 34 . Our data indicate that cells previously labeled as HE are more mature EHT cells and that cells previously labeled as pre-HE consist of both pre-HE and HE. The re-assignment of HE cells to cells undergoing EHT suggests that the previously described differentiation bottleneck 34 between pre-HE and HE could actually be a threshold between HE and the initiation of hematopoietic commitment (EHT). Our transcriptomic analyses also demonstrated that E11.5 HE has a strongly reduced hemogenic profile and is consequently less equipped to clear a potential bottleneck between HE and EHT. This unifies the seemingly contradictory observations that the AGM HE activity peaks around E10.5 93 while the frequency of phenotypic hemogenic endothelium (LIN -CDH5 + Gfi1:Tomato + ) is similar in the E10.5 and E11.5 AGM 21 .
We also found cell cycle status to be a distinguishing feature between pre-HE and HE and identified ACE as an enrichment marker for both populations. We uncovered that Ace transcript expression is mainly restricted to pre-HE/HE and compares favorably to the previously identified HE markers CD44 and Procr whose expression is detected in a wider range of AGM CDH5 + cells (Endothelial to IAHC). In human, ACE marks pre-hematopoietic mesoderm, aortic endothelial cells and HSPC 94,95 . In the E10.5 mouse AGM, we observed ACE expression exclusively in the endothelial lining of the DA and VA, which contain pre-HE and HE, while no robust ACE expression was observed anywhere else including in the more committed hematopoietic cell in IAHC.
The exclusive expression of ACE in the DA and VA lining suggests that it may contribute to a unique microenvironment promoting hematopoiesis. ACE converts angiotensin-1 into angiotensin-2, which in turn activates angiotensin-2 receptors (AT1, AT2) 96 . Two recent reports 97,98 have implicated the Endothelin-Renin-Angiotensin system (which includes ACE) in AGM hematopoiesis. In human and mouse, Atgr1 (coding for AT1) expression has been detected in subaortic DA cells 97,98 . In line with this, we observed Atgr1 expression in the RUNX1 + mesenchymal population, particularly in the smooth muscle fraction (C10:SMA, Supplemental Figure 19). Julien et al. 98 have recently shown that angiotensin-2 mediated activation of AT1 can promote the emergence of hematopoietic progenitors in mouse paraaortic splanchnopleure (P-Sp) cultures. Therefore, angiotensin-2 mediated signaling through AT1 on C10:SMA cells could, partially, explain our observation that Runx1b:RFP + mesenchyme can support hematopoietic activity.
The PDGFRa + mesenchymal (C9:MES-PDGFRa) fraction of the Runx1b:RFP + mesenchyme, can also play a role in stimulating hematopoiesis. We found that PTN, a secreted growth factor that plays a role in adult hematopoiesis 92 , is expressed by Runx1b:RFP + PDGFRa + mesenchymal cells in a RUNX1 dependent way. Our observation that ptn knockdown, in developing zebrafish, significantly inhibits HSPC emergence in the AGM suggests that local mesenchymal PTN production in the AGM plays a role in the EHT process.
Finally, our findings were greatly facilitated by employing high coverage full-length scRNA-seq on defined cell populations. Phenotypic knowledge of the sequenced populations regarding protein expression and transcriptionally active loci (as determined using reporter models) provided an invaluable extra layer of information to aid interpretation of the scRNAseq data. Overall, our data constitute a comprehensive atlas of EHT, which can be interrogated by the community to further our understanding of this process.