A novel prospective isolation of murine fetal liver progenitors to study in utero hematopoietic defects

In recent years, highly detailed characterization of adult bone marrow (BM) myeloid progenitors has been achieved and, as a result, the impact of somatic defects on different hematopoietic lineage fate decisions can be precisely determined. Fetal liver (FL) hematopoietic progenitor cells (HPCs) are poorly characterized in comparison, potentially hindering the study of the impact of genetic alterations on midgestation hematopoiesis. Numerous disorders, for example infant acute leukemias, have in utero origins and their study would therefore benefit from the ability to isolate highly purified progenitor subsets. We previously demonstrated that a Runx1 distal promoter (P1)-GFP::proximal promoter (P2)-hCD4 dual-reporter mouse (Mus musculus) model can be used to identify adult BM progenitor subsets with distinct lineage preferences. In this study, we undertook the characterization of the expression of Runx1-P1-GFP and P2-hCD4 in FL. Expression of P2-hCD4 in the FL immunophenotypic Megakaryocyte-Erythroid Progenitor (MEP) and Common Myeloid Progenitor (CMP) compartments corresponded to increased granulocytic/monocytic/megakaryocytic and decreased erythroid specification. Moreover, Runx1-P2-hCD4 expression correlated with several endogenous cell surface markers’ expression, including CD31 and CD45, providing a new strategy for prospective identification of highly purified fetal myeloid progenitors in transgenic mouse models. We utilized this methodology to compare the impact of the deletion of either total RUNX1 or RUNX1C alone and to determine the fetal HPCs lineages most substantially affected. This new prospective identification of FL progenitors therefore raises the prospect of identifying the underlying gene networks responsible with greater precision than previously possible.

The fetal liver is seeded initially by EMPs, followed by definitive HSCs from E11.5, and forms an ideal niche for hematopoietic stem and progenitor cell (HSPC) expansion and maturation [21,29,30].By E14.5, a definitive hematopoietic stem and progenitor cell hierarchy is believed to be firmly established in the fetal liver [31,32].Equivalents of previously described BM HPCs, including MEPs, CMPs, Granulocyte/Monocyte Progenitors (GMPs) and Megakaryocyte Progenitors (MkPs), have been identified in E14.5 fetal liver [31,33].However, the resolution at which these fetal liver hematopoietic progenitor cells can be identified and isolated is far lower than for their bone marrow equivalents [9,31,[34][35][36][37].This hinders the interrogation of this fetal liver hierarchy, particularly in mouse models of hematopoietic disorders with in utero origins.Indeed, higher-resolution purification of myeloid hematopoietic progenitor cells would allow the execution of transcriptomic and clonogenic lineage analysis to investigate the disruption of specific gene regulatory networks and the resultant impact on hematopoietic output.We therefore endeavored to improve our resolution of the specification of fetal liver myeloid hematopoietic progenitor cells.

Results
Delineating CD150/CD55 expression allows enhanced purification of immunophenotypic MEP and CMP fractions in E14. 5
The observation that fetal liver HSCs are capable of establishing a hematopoietic stem and progenitor cell hierarchy does not preclude the possibility that yolk sac erythro-myeloid progenitors (EMPs) could also contribute to these cell populations.To determine whether EMPs could directly give rise to a hematopoietic progenitor cell profile comparable to that observed in the fetal liver, we isolated yolk sac cells from E9.5 embryos and cultured them in pro-myeloid medium for up to 24 hours (with or without prior explant culture (S1M and S1N Fig) ).We observed that a small proportion of cells (up to 15% in the explant cultures) had the LK immunophenotype.The majority of these cells were either immunophenotypic GMPs or were LK CD16/32 -CD150 -CD41 + .No CD150 + LK progenitors were produced in these cultures.This therefore suggests that yolk sac EMPs are incapable of directly establishing our observed FL myeloid hematopoietic progenitor hierarchy.
https://doi.org/10.1371/journal.pgen.1007127.g002expression in E14.5 fetal liver.The aim was to achieve the prospective isolation of these populations from wild type (WT) or other transgenic mouse lines in the absence of the Runx1-P2-hCD4 reporter.Firstly, we performed Single Cell RNA Sequencing on CD41 -CD150 + P2-hCD4 -and P2-hCD4 + MEPs and CMPs.By performing a principal component analysis, we observed a clear transition in transcriptomic activity from the P2-MEP to the P2+ CMP (S4A Fig) .In particular, erythroid gene expression (for example, Klf1) was highly upregulated in the P2-MEPs being downregulated with the upregulation of P2-hCD4 and also sharing an inverse relationship with the megakaryocytic transcription factor Fli1 and the granulocyte/monocyte transcription factor Spi1 (Pu.1).(Full lists of differentially expressed genes between P2-/+ MEPs and P2-/+ CMPs can be found in S5 In order to identify markers which may aid the isolation of P2-hCD4 -and P2-hCD4 + MEPs, we analyzed the expression of various cell surface markers, the aim being to identify genes which are upregulated or downregulated in some or all P2-hCD4 + MEPs compared to P2-hCD4 -MEPs.Promising candidates included Cd48, Pecam1 (Cd31), Ptprc (Cd45), Eng (Endoglin), Itgb1 (Cd29) and Tek (Tie2) (S4B Fig) .We therefore screened antibodies which had been raised against these markers and other heterogeneously-expressed markers in fetal liver ( S4C Fig).
All   CMPs produced BFUes and 54% of single cocultured cells yielded solely TER119 + erythroid cells.CD11b + Gr1 + granulocyte/monocyte and CD41 + megakaryocyte cell output was enhanced considerably in CD31 high CMPs.Interestingly, although CFU-GEMM activity was enriched in CD31 high CMPs, the proportion of positive co-cultured wells yielding megakaryocyte/erythroid/granulocyte/monocyte cells was higher for CD31 low CMPs.This was probably due to the overall plating efficiencies being higher for CD31 high CMPs.Therefore, both CD31 low CMPs and CD31 high CMP populations are heterogeneous, albeit skewed towards erythroid and GM/megakaryocytic specification respectively.
The short-term culture of these populations revealed hierarchical relationships; CD31 + MEPs gave rise to CD31 -MEPs and MkPs, whereas CD31 -MEPs rapidly downregulated CD150 (S4K and S4M Fig).The CD31 low CMPs are skewed to a pro-MEP fate, whereas CD31 high CMPs efficiently produce GMPs and CD31 + MEPs (S4L and S4N Fig).CD31 low CMPs therefore comprise a more advanced pro-erythroid fraction, but can also produce CD31 high CMPs, suggesting some granulocyte/monocyte commitment.
As previously indicated, the provenance of the fetal liver myeloid progenitors is not entirely clear.Following initial fetal liver colonization at E11.5, the HSCs expand exponentially and the numbers of repopulating units (per embryo equivalent) peak by E16, as fetal bone marrow colonization is underway (having begun from E15) [29,30,44].To add weight to the hypothesis that the CD150 + MEP and CMP populations are fetal liver HSC-derived, we analyzed these fractions in E16.5 fetal liver.Firstly, upon analyzing the LK hematopoietic progenitor compartment of the Runx1 P1-GFP::P2-hCD4 E16.5 fetal liver, we observed that the frequencies of P2 +/-CD41 -CD150 + MEPs and CMPs closely resembled that observed in E14.5 fetal liver (S5A and S5B Fig).Upon isolating these populations, we observed that the P2 + CD41 -CD150 + MEPs and CMPs displayed decreased erythroid and increased megakaryocytic/GM output compared to the P2 -fractions (S5C- S5H Fig).We also confirmed that Runx1 P2-hCD4 expression correlated well with CD31 expression in the E16.5 fetal liver LK fractions (S5I-J) and that wild type E16.5 CD31 +/-MEPs and CD31 low/high CMPs displayed similar lineage specificities, particular concerning erythroid output, compared to their P2 +/- equivalents (S5K-P).This therefore suggests that similar MEP and CMP populations can be discerned in E16.5 fetal liver as in E14.5 fetal liver and therefore that the hematopoietic hierarchy is maintained 5 days after fetal liver colonization, even after the shift to bone marrow colonization has begun.

Runx1-null CD31 low CMPs have impaired erythroid cell production
Defining restricted fetal liver hematopoietic progenitor cell compartments should provide the ability to identify the impact of genetic alterations with greater precision.For example, Runx1-flox::Vav1-Cre conditional knockout mice display impaired fetal liver erythroid and megakaryocytic maturation (S6A- S6H Fig).
The absence of RUNX1 protein in Runx1-flox/flox::Vav1-Cre (Runx1-del/del, S6A Fig) apparently caused a block in the upregulation of TER119 expression.Consequently, there was an accumulation of the immature S0-S2 erythroid lineage subsets, but a decrease in the CD71 high TER119 high S3 population (S6B-E) [45].Following megakaryocytic culture, Runx1-del/del fetal liver samples produced more CD41 high megakaryocytes than their WT and Runx1-del/+ littermates, but the majority of these did not upregulate CD42d, reflecting the previously described megakaryocytic maturation block (S6F and S6G Bulk OP9 co-culture revealed decreased mature TER119 + erythroid cell production by Runx1-del/del CD31 low CMPs and CD31 -/+ MEPs, but not CD31 high CMPs (Fig 5D and 5E,S7G and S7K Fig), confirming an erythroid maturation block.Runx1-del/del CD31 low CMPs had increased CD41 + megakaryocytic output, in line with the previously reported increased proliferation/reduced maturation in this lineage [48].However, this did not answer the question of whether lineage specification was altered by the absence of RUNX1 causing a perturbation of the numbers of pro-erythroid, pro-megakaryocyte and/or pro-granulocyte/monocyte progenitors in hematopoietic progenitor pools.
To address this question, we performed single cell OP9 co-cultures and observed similar total plating efficiencies for wild type, Runx1-del/+ and Runx1-del/del populations ( CMP fraction is diminished in Runx1-null E14.5 embryos, and that differentiation of these progenitors is impaired. A key advantage of identifying these highly purified CD31 low and CD31 high CMP fractions in Runx1 null fetal liver is the ability to analyze underlying cell-intrinsic changes driving the lineage specification and maturation defects with greater precision, particularly at a transcriptome level.We therefore analyzed the expression of key HSC and lineage-associated transcriptional regulators in wild type, Runx1-del/+ and Runx1-del/del CD31 low and CD31 high CMPs (S7Q Fig) .One key observation was that several key HSC and megakaryocyte/erythrocyteassociated transcription factors (Tal1, Gfi1b and Klf1) were downregulated in wild type CD31 high CMPs compared to wild type CD31 low CMPs, reflecting the megakaryocytic/erythroid lineage commitment which accompanies CD31 downregulation in the CMP compartment and demonstrating that CD31 low and CD31 high CMPs represent transcriptionally distinct progenitor subsets.
Notably, expression of Tal1, Gfi1b and Klf1 did not differ between wild type, Runx1-del/+ and Runx1-del/del CD31 high CMPs.However, they were substantially downregulated in Runx1-del/del CD31 low CMPs compared to their wild type and Runx1-del/+ equivalents.In fact, the transcription factors Tal1, Gfi1b, Klf1 and Gata2, plus the megakaryocytic/erythroid lineage markers Itga2b, Pf4 and Epor (S7R Fig), were all expressed at comparable levels in Runx1-del/del CD31 low CMPs to CD31 high CMPs (of all genotypes).This supports the hypothesis that the absence of RUNX1 results in a differentiation block between the CD31 high and CD31 low CMPs.Additionally, it indicates that the megakaryocytic/erythroid maturation defects are established even at this early stage, with a failure to upregulate key maturation-associated transcripts.
Interestingly, unlike the pro-erythroid transcription factor Klf1, the pro-megakaryocytic Fli1 and the pro-GM Spi1 and Gfi1 were not significantly different in CD31 low and CD31 high CMPs; nor were they impacted by the absence of RUNX1.This may indicate that commitment to the erythroid lineage is the default position in the FL CD31 high -to-CD31 low CMP transition, which was hindered in the absence of RUNX1.We therefore clearly demonstrate the benefits of isolating highly purified myeloid progenitors to aid in the understanding of the mechanistic basis of congenital hematopoietic defects in the fetal liver.

Runx1-P1-MRIPV CD31 high CMPs have impaired Mk specification
We recently reported that deleting the RUNX1C isoform in adult mice, whilst maintaining total RUNX1 expression, resulted in mild thrombocytopenia, due to impaired megakaryocytic specification but with normal megakaryocyte maturation.To achieve this we utilized the Runx1-P1-MRIPV line, in which the P1-encoded RUNX1C isoform is replaced by the P2encoded RUNX1B isoform [39].Given the contrast between this phenotype and that of the Runx1-del/del adult model, we decided to interrogate P1-MRIPV fetal liver hematopoiesis.
This hypothesis was confirmed through short-term culture of the CMP populations (S9J-S9M Fig), as P1-MRIPV/MRIPV CD31 high CMPs produced fewer MkPs and more GMPs than their wild type counterparts, but the lineage specification of CD31 low CMPs was largely unaffected.Therefore, in contrast to the impaired erythroid specification of Runx1-del/del CD31 low CMPs, P1-MRIPV/MRIPV CD31 high CMPs display impaired megakaryocytic specification.This demonstrates that the RUNX1B and RUNX1C isoforms have distinct roles during fetal liver hematopoiesis, akin to adult bone marrow hematopoiesis, which can be uncovered using our enhanced fetal liver HPC purification strategies.

Discussion
Bone marrow myeloid hematopoietic progenitor cells have been extensively characterized, including the recent demonstration that immunophenotypically similar CMPs can be separated into distinct PU.1-eYGP high GATA1-mCherry -pro-granulocyte/monocyte and PU.1-eYFP low GATA1-mCherry + pro-megakaryocyte/erythroid fractions [40].This therefore confirms that lineage fate decisions had commenced upstream of the CMPs in ancestral HSPCs.By comparison, the delineation of myeloid lineage-restricted hematopoietic progenitors in fetal liver lags critically behind.We therefore attempted to further compartmentalize the immunophenotypic MEP and CMP fractions to provide a more detailed hematopoietic hierarchy (summarized in Fig 7A).CD55 and CD150 were obvious lead candidates, as CD55 was successfully utilized by Guo et al to subdivide adult CMPs into pro-megakaryocyte/erythroid and pro-granulocyte/monocyte fractions [41]; additionally, CD150 is a well-established pro-megakaryocytic/erythroid and HSC marker [5,32,43,49,50].We observed that the entirety of megakaryocytic/erythroid bipotential and granulocyte/monocyte/megakaryocyte/erythrocyte multipotential CFU-C output resides respectively in the CD55 + CD150 + MEP and CD55 + CD150 + CMP fetal liver fractions.However, the CFU-Mk and CFU-MkE output of CD55 + CD150 + MEPs remained low (~5%), highlighting the need for further refined hematopoietic progenitor subfractionation.
For this, we identified distinct myeloid hematopoietic progenitor subsets on the basis of P2-hCD4 expression in our Runx1 P1-GFP::P2-hCD4 reporter mouse.To broaden the application of this finding, and not rely on the P1-GFP::P2-hCD4 reporter mouse, we searched for cell surface markers which correlated with P2-hCD4 expression.Following a screen of markers associated with heterogeneous fetal liver expression or multipotency/lineage specification in bone marrow hematopoietic progenitors [51], we identified CD31, CD45 and CD48 as strong candidates.Downregulation of the pan-leukocyte marker, CD45 [52,53], and the bone marrow hematopoietic progenitor, lymphocyte and macrophage-associated CD48 [49,[54][55][56] upon fetal liver erythroid commitment concurs with adult bone marrow erythroid progenitor specification [51].CD31, meanwhile, is expressed in endothelial progenitor cells and their mature progeny [57][58][59][60].CD31-null mice are viable and do not exhibit obvious vascular defects; nonetheless CD31 plays significant roles in vascular remodeling and tumor metastatic progression as well as in adhesion, survival, migration and activation of hematopoietic cells [61][62][63][64][65][66][67][68][69][70].In E14.5 fetal liver, CD31 is predominantly expressed in cells lining the hepatic vessels, but also in some Runx1 + hematopoietic stem and progenitor cells.We observed CD31 expression was highly enriched in the fetal liver myeloid progenitors with the greatest granulocyte/monocyte and megakaryocytic output.This concurs somewhat with the observation that in fetal liver and bone marrow, the entire multilineage LSK hematopoietic stem and progenitor cell fraction expresses CD31 [57,71,72].Contrastingly, the bone marrow LK CD31 + fraction was deficient in granulocyte/monocyte output, possessing chiefly short-term erythroid repopulating cells, whereas we demonstrate here that erythroid lineage commitment in fetal liver coincides with CD31 downregulation.This suggests CD31-expressing short-term progenitors are not equivalent throughout mouse ontogeny.The shift from CD31 + pro-GM/megakaryocytic fetal liver hematopoietic progenitors to bone marrow erythroid progenitors may reflect distinct interactions with their respective niches.CD31-null adult mice have more steady state circulating progenitors, as hematopoietic progenitors fail to migrate across the bone marrow vasculature [68].This phenotype was observed whether CD31 was deleted in hematopoietic or endothelial cells or both.The retention of the numerous, highly clonogenic CD31 -fetal liver erythroid progenitors in their niche may be less crucial than for CD31 + bone marrow erythroid progenitors, which are replaced less frequently by more quiescent precursors.
High CD31 expression in fetal liver pro-GM/megakaryocytic progenitors reinforces a phenotypic link between megakaryocytes and endothelial cells, which co-express numerous receptors, transcription factors and other signaling-associated factors [73].The similarities become even more pronounced when considering hemogenic endothelium, which produces HSCs through EHT, due to the elevated expression of hematopoietic regulators which drive this process [74][75][76].Megakaryocytes and endothelial cells are spatially close in hematopoietic vascular niches, their interactions conducted partially by CD31 [77].Indeed, CD31's absence impacts multiple aspects of megakaryopoiesis.It would therefore be worthwhile to determine whether CD31 deficiency impacts ancestral HPCs as well as their megakaryocytic progeny.Hematopoietic progenitor cell retention in the fetal liver vascular niche may consequently be severely impaired [68], potentially adding a new functional dimension to CD31 expression on CMPs and MEPs, as well as an immunophenotyping application.
One of the questions raised by our studies was whether the immunophenotypic CMP compartment actually contains single progenitor cells with the ability to produce granulocyte/monocyte, megakaryocyte and erythroid cells, thereby being defined as true Common Myeloid Progenitors.The alternative is that the CMP compartment solely comprises a heterogeneous population of monopotent or bipotent progenitors.Our single cell myeloid OP9 co-culture assays offer evidence that a small minority (<20%) of single isolated CMPs yield megakaryocytes, erythrocytes and granulocytes/monocytes, as would be expected for a true CMP.Therefore, we provide evidence that supports the existence of the CMP as a rare population within the fetal liver.
The CMP appears to be far scarcer than previously understood and it is likely that the LSK Multipotent Progenitor fractions may dominate in terms of common myeloid ancestry, particularly as they yield greater numbers of CFU-GEMMs than immunophenotypic CMPs.Indeed, we observed that fetal liver LSK HSCs and MPPs appeared to be the ancestors of the LK hematopoietic progenitors, at least in vitro.Contrastingly, we were unable to reproduce a similar myeloid progenitor hierarchy following culture of yolk sac cells.Nonetheless, several studies have suggested that at least some definitive hematopoietic stem and progenitor cells located in the embryo proper do not arise de novo, but instead originate from the yolk sac [75,[78][79][80].Therefore, it is possible that EMPs could be responsible to some extent for the establishment of a fetal liver hematopoietic progenitor hierarchy, including our populations of interest: the CD150 + MEPs and CMPs.Our results would suggest that this is only achieved after fetal liver colonization and differentiation in this supportive niche, yielding fetal liver hematopoietic stem and progenitor cells which, in turn, produce myeloid-restricted progenitors.
Our intent in this study was to delineate different fetal liver myeloid progenitor compartments, in order to provide a method to examine homeostatic developmental hematopoiesis and hematopoietic disease models.Indeed chromosomal translocations such as AML1-ETO, PML-RARA and CBFβ-MYH11, which cause childhood Acute Myeloid Leukemia (AML), frequently arise in utero as demonstrated by the high prevalence of such mutations in neonatal blood samples [34][35][36], and may therefore impact fetal hematopoiesis.Furthermore, Ye et al demonstrated that the initiation of AML requires partial myeloid differentiation by transformed CMPs to GMPs, highlighting how an understanding of the myeloid progenitor hierarchy facilitates examination of the origins and progression of malignant hematopoietic disorders [16].To demonstrate the application of our fetal liver myeloid progenitor scheme, we analyzed the impact of deleting Runx1 (a mutation which causes a megakaryocytic differentiation block comparable to FPD/AML [38,46,48]) on the specification and differentiation of these compartments.Akin to Behrens et al [81] in the adult, we observed deleting Runx1 causes a block in fetal liver erythroid differentiation and pinpointed this block to CD150 downregulation in immunophenotypic MEPs (Fig 7B).We also observed decreased erythroid specification in the Runx1-del/del CD150 + CD31 low CMP fraction, and determined that this may be due to a failure to upregulate an erythroid transcriptional network in the transition from CD31 high to CD31 low CMP.We are therefore able to gather a substantial amount of phenotypic information and also gain mechanistic insights from transcriptome analyses.We also evaluated the impact of removing the dominant RUNX1C isoform on fetal liver myelopoiesis.As we previously described in the adult, RUNX1C-null P1-MRIPV/MRIPV mice do not appear to have impaired megakaryocytic/erythroid differentiation [39].Nonetheless, the fetal liver CD150 + CD31 -MEP fraction was expanded, partially recapitulating the Runx1null erythroid lineage phenotype.However, P1-MRIPV/MRIPV fetal liver had reduced megakaryocytic specification, with the CD150 + CD31 high CMP-to-MkP transition being the most impaired pathway.We did not observe this in Runx1-null fetal liver; in fact megakaryocytic output increased in the absence of total RUNX1, as observed in adult hematopoiesis [39,81].This difference may explain the apparently contradictory finding by Kuvardina et al [82] that RUNX1 promotes megakaryocytic specification; RUNX1C may have a non-redundant function in megakaryocytic specification, whereas RUNX1B is either necessary or sufficient for megakaryocytic and erythroid maturation.Our myeloid hematopoietic progenitor scheme has allowed us to identify the cells of interest which perpetuate this hematological imbalance.Such a technique could therefore be applied to understanding the cells of origin in familial thrombocytopenia, or a recently described case of RUNX1-deleted Congenital Amegakaryocytic Thrombocytopenia [83].Moreover, our new protocol for prospective isolation of myeloid hematopoietic progenitors could be applied more broadly to other congenital or somatic genetic disorders which manifest in utero, as well as to analyzing lineage fate decisions in normal fetal liver hematopoiesis.
All animal work was performed under regulations governed by UK Home Office Legislation under the Animals (Scientific Procedures) Act 1986 and was approved by the Animal Welfare and Ethics Review Body of the Cancer Research UK Manchester Institute.

Fluorescence Activated Cell Sorting (FACS)
Details of FACS reagents and combinations used for each analysis are listed in S2 and S3 Tables.Prior to flow sorting or analysis of HPCs in E12.5, E13.5, E14.5 or E16.5 fetal liver, red blood cell lysis was performed as described [18].Dead cells were excluded using 1μg/ml Hoechst 33258 (ThermoFisher Scientific); gates were positioned based on Full Minus One controls.Cells were analyzed using a LSR-II or LSR-II Fortessa analyzer (BD Biosciences) or a NovoCyte (ACEA Biosciences Inc.).Cells were sorted using a FACSAria-II, FACSAria-III or Influx cell sorter (BD Biosciences).
Short-term culture of yolk sac cells.Yolk sacs were dissected from E9.5 embryos and either cultured as explants as described [88] or dissociated with 1mg/ml Collagenase-Dispase/ Phosphate Buffered Saline (PBS) (Roche) at 37˚C for 20 minutes.Yolk sac cell suspensions were then either FACS analyzed for hematopoietic stem/progenitor markers as described in S3 Table (day 0) or cultured in pro-myeloid medium (see above) for 24 or 48 hours.After 48 hours, explant-cultured yolk sac cells were dissociated by trypsinization and either FACS analyzed for hematopoietic stem/progenitor markers (see S3 Table) immediately or following culture in pro-myeloid medium for a further 24 hours.
Megakaryocyte culture of fetal liver.Following red blood cell lysis, unfractionated fetal liver cells were cultured at 37˚C in 5% CO 2 and atmospheric O 2 for 7 days at a density of 10 5 cells per ml in pro-megakaryocyte medium (IMDM supplemented with 10% FBS for Mouse Myeloid Colony-Forming Cells, 2mM L-glutamine, Penicillin-Streptomycin, 0.45mM MTG, 20ng/ml IL6, 50ng/ml IL11 and medium conditioned by cell lines producing IL3 and TPO (1% final concentration).Cells were then harvested and FACS analyzed for CD41 and CD42d expression as described in S3 Table.

Cytospin analysis
Up to 50,000 cultured fetal liver cells were suspended in 150μl PBS and immobilized on twin frosted glass microscope slides (Fisher Scientific) by cytospin at 200rpm, low acceleration for 5 minutes in a Shandon Cytospin3 (Thermo Scientific).Air-dried slides were submerged in May-Gru ¨nwald Eosin methylene blue Q Path stain (VWR) for 3 minutes, rinsed in tap water and submerged in 5% Giemsa's stain (VWR) for 20 minutes.Slides were subsequently air dried, mounted and scanned using the Pannoramic 250 Flash III (3DHISTECH).Images were acquired with the Pannoramic 250 software and analyzed with the Pannoramic Viewer software (3DHISTECH).

Immunofluorescent staining and imaging of fetal liver sections
Dissected fetal livers were fixed in 4% Paraformaldehyde (PFA) overnight, before they were soaked in 30% sucrose and mounted in OCT compound.10μm sections were prepared using a cryostat.The sections were incubated in blocking buffer (PBS with 10% FBS, 0.05% Tween20 and 10% goat serum (DAKO)) for 1 hour before the sections were incubated with primary antibodies at 4˚C overnight in blocking buffer.
Sections were washed three times in PBST (PBS with 0.05% Tween20) for 15 minutes each and then incubated with fluorochrome-conjugated secondary antibody at room temperature for 1 hour.
Sections were further washed three times in PBS and mounted using Prolong Gold antifade medium with DAPI (Life Technologies).Images (of Alexa Fluor 488, Alexa Fluor 647, DAPI and endogenous RFP) were taken using a low-light time lapse microscope (Leica) using the Metamorph imaging software and processed using ImageJ.

Western blot analysis
Whole cell protein extracts, purified using RIPA lysis buffer, were quantitated by the Bradford assay (Protein Assay Dye Reagent, Bio-Rad) on the Glomax Multi Detection System (Promega) and loaded alongside the SeeBlue Plus2 Prestained Standard (ThermoFisher Scientific) for electrophoretic separation on NuPAGE 4-12% Bis-Tris gels using the Novex Mini Cell system (ThermoFisher Scientific).Protein transfer was performed to nitrocellulose membranes using the iBlot system (ThermoFisher Scientific) and membranes were probed with anti-RUNX (EPR3099, Abcam) and anti-beta-actin (AC-15, Sigma-Aldrich) primary antibodies, followed by Horseradish Peroxidase (HRP)-conjugated goat anti-rabbit and goat anti-mouse (Thermo-Fisher Scientific) secondary antibodies respectively, in the iBind Western System (Thermo-Fisher Scientific).HRP activity was detected using Amersham ECL Prime Western Blotting Detection Reagent, imaged using the BioRad ChemiDoc Touch Imaging System and analyzed with BioRad Image Lab Software Version 6.

Gene expression by quantitative PCR
RNA was extracted using the RNeasy Plus Micro Kit (QIAGEN) and complementary DNA was synthesized using the High-Capacity cDNA Reverse Transcription Kit (ThermoFisher Scientific).Quantitative PCR (qPCR) was performed using Universal ProbeLibrary assays (Roche); primers and probes are listed in S4 Table.Expression values were normalized to beta-actin (Actb).
Single cell RNA sequencing E14.5 P1-GFP::P2-hCD4/+ fetal livers were prepared for flow sorting as described above.Single P2-hCD4 -/+ MEPs and CMPs were sorted into 384-well plates containing lysis buffer and snap frozen.Libraries were prepared using a modified version of the Smart-Seq2 protocol [89].Briefly, cDNA was prepared using a Mantis platform (Formulatrix) and quantified with quan-tIT picogreen reagent (Thermo Fisher).Dual indexed sequencing libraries were prepared from 0.1ng cDNA using an Echo525 automation system (Labcyte) in miniaturized reaction volumes.The library pool was quantified by qPCR using a Library Quantification Kit for Illumina sequencing platforms (Kapa Biosystems).Paired-end 75bp sequencing was carried out by clustering 1.5pM of the library pool on a NextSeq 500 sequencer (Illumina).
Base call files generated from the NextSeq 500 sequencing run were converted to the FASTQ format with the bcl2fast converter (Illumina).All FASTQ files corresponding to the same sample (derived from separately sequenced lanes) were then merged into a single FASTQ file (one per sample).Read trimming was performed with trimmomatic (v0.36) with the following settings "CROP:75 HEADCROP:5 SLIDINGWINDOW:20:20 MINLEN:36".The mouse reference genome GRCm38 (version M12, Ensembl release 87) and the ERCC reference sequence (Thermo Fisher) were combined and used as the reference genome for sequencing alignment, performed using STAR (version 2.4.2a)[90].The expression levels of 49,585 features annotated in the GENCODE mouse genome and 92 ERCC features were determined using HTSeq (version 0.6.1p1)[91].The following parameters were specified for the HTSeq quantification: '-format = bam-stranded = no-type = exon'.
Single cell count data were loaded into the R environment (R version 3.4.0)as a SCESet object using the Scater package (version 1.4.0)[92].Normalized gene expression values were taken from the default normalization performed by scater.Cells with fewer than 250,000 sequencing reads, more than 25% unmapped reads, and more than 15% ERCC content were removed.Low abundance genes (mean count <1) were excluded, as were overrepresented genes (>20% of total sequencing reads).Differentially expressed genes were identified using DESeq2 (version 1.16.1,Bioconductor).Prior to differential expression analysis, the data were filtered to remove genes with a dropout rate of higher than 75%; differential expression analysis was then performed using the function "DESeqDataSetFromMatrix" and by specifying the contrast of interest.The full scripts for the analysis of these data are available at https://github.com/m-zaki/CRUKMI_github/tree/master/JuliaDraper_PLOSgenetics.
The data discussed in this publication have been deposited in the NCBI Gene Expression Omnibus [93,94] and are accessible through GEO series accession number GSE107653.

Statistical analysis
For FACS purified populations (MEPs, CMPs) a sample size of n = 1 refers to tissues pooled from embryos from one litter.For total fetal liver analyses (Western blot, total FL culture), a sample size of n = 1 refers to one embryo.
Data were evaluated using an Ordinary 2-way ANOVA and expressed as mean ± standard error of the mean (SEM).P<0.05 was considered statistically significant.
Fig 5F).Whereas >50% of wild type and Runx1-del/+ CD31 low CMPs solely produced TER119 + Erythroid cells, this was reduced 4-fold in Runx1-del/del CD31 low CMPs (Fig 5G and 5H, S7L Fig), with a concurrent increase in pro-GM progenitors.Erythroid cells were more modestly reduced in CD31 high CMP cultures (S7M-S7O Fig), but this was not very impactful as the population has a lower pro-erythroid pool in wild type fetal liver compared to its CD31 low counterpart.Moreover, the Median Fluorescent Intensity of the TER119 + erythroid cells was decreased in Runx1-del/del CD31 low CMP cultures compared to that of wild type littermates (S7P Fig).This confirmed that the proportion of pro-erythroid progenitors residing in the fetal liver CD31 low