Distinct myeloid progenitor differentiation pathways identified through single cell RNA sequencing

Single cell RNA sequencing identifies pre-GM heterogeneity

To address if the same myeloid differentiation pathways emerge from CMPs and LMPPs we performed global single cell gene profiling of bone marrow (BM) pre-GMs⁶, the earliest progenitor predicted to be shared between the two pathways^{6, 7, 8}. RNA sequencing of 63 single pre-GMs was followed by unsupervised clustering using the 100 genes with greatest variance across the entire cell population. This led to the identification of two distinct pre-GM clusters (Supplementary Fig. 1a), and 55 genes showing differential expression between them (P<0.05; FDR<0.05; Supplementary Fig. 1b) as potential sub-population classifiers. Clustering using this gene set identified three distinct pre-GM subpopulations (Gata1⁺Flt3⁻, Gata1⁻Flt3⁻ and Gata1⁻Flt3⁺) defined by expression of Gata1 and Flt3 (Fig. 1a), a pattern that was validated by targeted single cell gene expression analysis (Fig. 1b). This analysis also identified Gata1 expression as an optimal classifier, homogeneously and selectively expressed in a distinct subpopulation of Flt3⁻ preGMs, but not detected in Flt3⁺ preGMs. Notably, Gata1 and Flt3 expression have been used to define CMPs¹¹ and LMPPs¹², respectively, within the CD34⁺LSK population, suggesting they have the potential to identify pre-GM subsets derived from these distinct upstream progenitors.

Gata1 expression defines distinct myeloid progenitors

The Gata1-EGFP transgene used to define CMPs¹¹ lacks key regulatory elements^{19, 20, 21}, and is not expressed in GMPs²². We therefore generated the Gata1-EGFP reporter mouse line using a BAC transgenic construct containing all known Gata1 regulatory sequences, in which an enhanced green fluorescence protein (EGFP) expression cassette replaced the coding part of the second exon of the Gata1 gene (Supplementary Fig. 1c). In these Gata1-EGFP reporter mice distinct GE⁺ and GE⁻ subsets were observed in pre-GMs and GMPs (Fig. 1c; Supplementary Fig. 1d). Within the BM multi-potent Lin⁻Sca-1⁺c-Kit⁺ (LSK) compartment, HSCs are CD150⁺LSK²³, whereas LMPPs are Flt3^hiLSK¹². CD150⁺LSK cells contained a significant fraction of GE⁺ cells (Fig. 1e; Supplementary Fig. 1e), but only GE⁻ CD150⁺LSK cells possessed HSC function in vivo (Supplementary Fig. 2a-d). HSCs are therefore defined herein as CD150⁺GE⁻LSKs. Similarly, a small fraction (2-3%) of LMPPs had low Gata1-EGFP expression (Fig. 1d). Previous studies showed that approximately 2% of LMPPs possess detectable Mk potential¹², and we found this Mk potential to be confined to the small GE⁺ sub-fraction of Flt3^hiLSK cells (Supplementary Fig. 2e), supporting the existence of an LMPP population devoid of platelet-forming capacity^{12, 24, 25}, and leading us to stringently define and to sort LMPPs as Flt3^hiGE⁻LSK in this study. Importantly, Gata1-EGFP expression mirrored endogenous Gata1 mRNA expression both at the population level (Fig. 1e) and in analysis of single pre-GMs (Supplementary Fig. 2f). The Gata1-EGFP reporter therefore identifies transcriptional heterogeneity within the phenotypic HSC, LMPP, preGM and GMP populations.

Early separation of macrophage and mast cell potentials

Quantitative PCR of lineage-specific gene expression showed that megakaryocyte-erythroid–affiliated genes (Itga2b, Klf1, Epor, Gata2) were expressed preferentially in GE⁺ preGM and GMPs, preMegE (Lin⁻Sca-1⁻c-Kit⁺CD41⁻FcγRII/III⁻CD150⁺CD105⁻), MkP (Lin⁻Sca-1⁻c-Kit⁺CD41⁺) and preCFU-E (Lin⁻Sca-1⁻c-Kit⁺CD41⁻FcγRII/III⁻CD150⁺CD105⁺) progenitor cell populations (Fig. 1f), as were mast cell genes (Cma1, Mcpt1; Fig. 1g), whereas macrophage (Cd68, Emr1; Fig. 1h), neutrophil (Csf3r, Ctsg; Fig 1i) and lymphoid genes (Rag1, Cmah; Fig. 1j) were selectively expressed in LMPPs, GE⁻ pre-GMs and GE⁻ GMPs. The divergent expression of macrophage-neutrophil and mast cell genes between GE⁺ pre-GMs and GMPs and GE⁻ pre-GMs and GMPs, suggested that these progenitors could generate different myeloid cell types. We therefore cultured GE⁺ pre-GMs and GE⁻ pre-GMs under pan-myeloid conditions (IL-3, IL-5, IL-9, SCF and GM-CSF) for 8 days, and examined the resulting cultures by cytospin morphology and gene expression analysis. Morphological analysis showed that LMPPs and GE⁻ pre-GMs generated monocytes, but no mast cells, whereas GE⁺ pre-GMs generated mast cells, but no monocytes (Fig. 1k). Polymorphonuclear (PMN) cells were observed in both cultures, with neutrophil morphology (small cell size, highly condensed and segmented nucleus) evident in those derived from GE⁻ preGMs and LMPPs, whereas those derived from GE⁺ pre-GMs were generally larger with less condensed nuclei (Fig. 1k). Consistent with the observed separation of mast cell and monocyte-macrophage potential, mast cell-specific genes were exclusively expressed in the GE⁺ pre-GM-derived cultures (Fig 2a), and monocyte-macrophage–specific genes selectively in those derived from LMPPs and GE⁻ pre-GMs (Fig. 2b). Finally, while the genes encoding the myeloid PU.1 and C/EBPα transcription factors were expressed in all cultures, Gata1 and Gata2 expression was confined to cultures derived from GE⁺ pre-GMs (Fig. 2c,d).

To determine the frequencies and distribution of granulocyte and monocyte-macrophage lineage potentials within the different progenitor populations at the single cell level we individually cultured sorted progenitor cells and analyzed their myeloid lineage output at several time points. While single LMPPs, GE⁻ pre-GM and GE⁻ GMPs produced large numbers of monocytes, mast cell potential was never detected in cultures of single LMPPs or GE⁻ GMPs, and was very rare (<2% of cultures) in GE⁻ pre-GMs single cell-derived cultures, at all time points investigated. In contrast, monocytes were generated only very rarely (<2% of single cell cultures), whereas mast cell potential was highly abundant, from GE⁺ pre-GM or GE⁺ GMPs (Fig. 2e), Combined monocyte and mast cell morphology was exceptionally rare, seen only in 2 of >1000 single cell-derived clones of all progenitor analyzed. Importantly, this was not due to any inability of the culture system to support development of these two cell types simultaneously, as culture of multi-potent HSCs or co-culture of GE⁺ and GE⁻ pre-GMs generated combined mast cells and monocytes with high frequency (Supplementary Fig. 2g,h). Both monocytes and mast cells were observed in conjunction with other granulocytes with high frequency in single cell-derived clones from GE⁺ pre-GM cells (22-33%), GE⁻ pre-GM (ca. 50%) and LMPPs (65%), and also in GE⁺ GMPs and GE⁻ GMPs at lower frequencies (Fig. 2e). The polymorphonuclear cells associated with monocytes showed neutrophil morphology, whereas those associated with mast cells appeared larger with less condensed nuclei (Fig. 2f). These results show that GE⁺ and GE⁻ myeloid progenitors have distinct lineage potentials.

Eosinophil potential is found in GE⁺ preGMs and GE⁺ GMPs

While the above data clearly showed that mast cell and monocyte-macrophage potentials were separated prior to the formation of pre-GMs and GMPs, the nature of the additional granulocyte lineage potentials associated with these progenitors remained unclear. To address this issue we performed Affymetrix-based global gene profiling of pre-GMs, GMPs, HSCs, LMPPs, CLPs (Lin⁻B220⁻Sca-1^loc-Kit^loFlt3⁺IL-7Rα⁺)²⁶ and preMegEs. Consistent with their similar granulocyte-monocyte lineage readouts, LMPPs, GE⁻ pre-GMs and GE⁻ GMPs clustered together in principal component analysis, in close association with CLPs, whereas GE⁺ pre-GMs and GE⁺ GMPs formed a separate cluster, associated with preMegEs (Fig. 3a). Direct comparison of GE⁺ GMPs and GE⁻ GMP transcriptomes showed that GE⁺ GMPs were highly enriched for mast cell and eosinophil-specific gene expression, whereas numerous monocyte-macrophage and neutrophil-associated genes were more highly expressed in GE⁻ GMPs (Supplementary Table 1). We therefore examined the possibility that the polymorphonuclear cells derived from GE⁺ pre-GMs and GE⁺ GMPs were eosinophils. First, we cultured GE⁺ GMPs and GE⁻ GMPs in the presence of SCF and IL-5 (Supplementary Fig. 3a), and examined the resulting cultures by morphology and gene expression. Under these conditions GE⁺ GMPs did not generate mast cells, but only large polymorphonuclear cells with uncondensed nuclear morphology, while GE⁻ GMPs generated cells with monocyte and neutrophil morphology (Fig. 3b). Gene expression analysis indicated the selective expression of eosinophil-specific genes (Il5ra, Ccr3, Siglec5, Prg2) in GE⁺ GMP cultures (Fig. 3c), and of monocyte-macrophage-specific genes (Cd68, Lpl, Lrp1, Mpeg1) in GE⁻ GMP cultures (Fig. 3d). In both cases the cell-type specificity of gene expression was validated by analysis of macrophages and eosinophils sorted from the peritoneal cavity as controls. Cytospins indicated that the morphology of the in vitro GE⁺ GMP-derived eosinophil-like cells was similar to that of purified peritoneal eosinophils (Supplementary Fig. 3b). These results indicated the selective development of eosinophils from GE⁺ GMPs.

We also performed flow cytometric analysis of the cells differentiated from GE⁺ pre-GMs and GE⁻ pre-GMs, as well as GE⁺ GMPs and GE⁻ GMPs following culture under pan-myeloid cytokine conditions. Cells with a FcεRIα⁺c-Kit⁺GE⁺ mast cell surface phenotype were abundant in GE⁺ pre-GM and GE⁺ GMP cultures, whereas Mac-1⁺Ly6G^loGE⁻ monocytes and Mac-1⁺Ly6G^hiGE⁻ neutrophils were the principal cell types generated from GE⁻ pre-GMs and GE⁻ GMPs (Fig. 4a,b), indicating that neutrophil potential was selectively associated with GE⁻ pre-GMs and GE⁻ GMPs. Eosinophil, neutrophil and mast cell identities were further confirmed by analysis of the expression of lineage-specific genes (Supplementary Fig. 3c-f). FcεRIα⁻c-Kit⁻SiglecF⁺GE⁺ eosinophils were inefficiently generated under the pan-myeloid conditions (Fig. 4a,b), likely because of overgrowth by mast cells. To circumvent this, we cultured GMP in the presence of IL-5 and SCF, and observed the selective development of eosinophil-like cells in GE⁺ GMP cultures, while GE⁻ GMPs gave rise to monocytes and neutrophils, but few or no eosinophils, even under conditions favoring eosinophil development (Fig. 4c). As such, based on analyses of morphology, surface markers and gene expression we concluded that GE⁺ GMPs gave rise to eosinophils and mast cells, but did not produce detectable neutrophils or monocytes. Finally, to accurately co-localize lineage potentials in single progenitors we combined morphological analysis of cytospins of single cell-derived cultures with immunocytochemistry against EGFP to distinguish GE⁺ eosinophils from GE⁻ neutrophils (Fig. 4d). GE⁺ pre-GMs and GE⁺ GMPs contained a high proportion of bi-potent mast cell-eosinophil progenitors, whereas GE⁻ LMPPs, GE⁻ pre-GMs and GE⁻ GMPs were abundant in bi-potent neutrophil-monocyte progenitors (Fig. 4e). In addition, under pan-myeloid conditions LMPPs, GE⁺ pre-GMs and GE⁻ pre-GMs we observed the emergence of FcγRII/III⁺GE⁺ cells from GE⁺ pre-GMs and FcγRII/III⁺GE⁻ cells from GE⁻ pre-GMs and LMPPs in both cases displaying the clonogenic lineage potentials of the corresponding GE⁺ and GE⁻ GMP population (Supplementary Fig. 4a,b), further supporting the existence of independent and functionally distinct GE⁺ and GE⁻ progenitor hierarchies. The comparison of gene expression profiles of GE⁺ GMPs and GE⁻ GMPs identified genes encoding cell surface markers that were preferentially expressed in the GE⁺ subsets, such as Cd55, Il1rl1, or the GE⁻ subsets, such as Ly6c (Supplementary Table 1). While cell surface expression of Il1rl1 was not detectable on pre-GMs or GMPs with available antibodies, CD55 and Ly6C were expressed selectively on GE⁺ pre-GMs and GMPs, and GE⁻ pre-GMs and GMPs, respectively (Supplementary Fig. 4c). In agreement with this selective expression, they allowed the separation of progenitors with a restricted monocyte-neutrophil potential (Ly6C⁺ pre-GMs and GMPs) or restricted mast cell-eosinophil potential (CD55⁺ pre-GMs and GMPs), without the use of the Gata1-EGFP transgene (Supplementary Fig. 4d). Overall, these results indicate that mast cell-eosinophil potential is found in GE⁺ preGMs and GE⁺ GMPs, whereas monocyte-neutrophil potential resides in LMPPs, GE⁻ preGMs and GE⁻ GMPs.

Gata1 expression defines two progenitor domains

We further examined the extent to which the identified progenitors possessed other lineage potentials. Gene Set Enrichment Analysis-based²⁷ comparison of lineage programming in ⁻ pre-GMs and GE⁺ pre-GMs showed strong enrichment of Mk-E-associated genes in GE⁺ pre-GMs, and of lymphoid genes in GE⁻ pre-GMs (Fig. 5a). All the identified pre-GM and GMP populations generated colonies when cultured under myeloid conditions (Fig. 5b). In contrast, megakaryocytic and erythroid colony forming ability was primarily associated with GE⁺ pre-GMs, which showed a Mk-E potential comparable to that seen in preMegEs (Fig. 5c,d), and contained multi-potent cells with combined granulocyte and Mk-E potential, but no monocyte-macrophage potential (Fig. 5e,f). Conversely, B- and T-lymphoid potential was high in LMPPs, but also significant in GE⁻ pre-GMs, but not GE⁺ pre-GMs (Fig. 5g; Supplementary Fig. 4e).

Antagonistic transcription factor pairs, including PU.1–GATA-1 and C/EBP–FOG-1, have been proposed to be co-expressed in multi- or bi-potent progenitors, and resolution of the antagonism to underlie lineage bifurcations²⁸. We therefore measured the expression of these and other key transcription factors in GE⁺ pre-GMs and GMPs, as well as GE⁻ pre-GMs and GMPs, as well as in HSCs, LMPPs, pre-CFU-Es, pre-MegEs and CLPs, to determine if their expression was consistent with the associated lineage potentials. Most notably, Cebpa, Cebpb and Zfpm1 (encoding FOG-1) were all expressed in GE⁺ pre-GMs, with Zfpm1 down-regulated and Cebpa and Cebpb up-regulated in GE⁺ GMPs, The converse change, Zfpm1 upregulation and Cebpa and Cebpb downregulation, was seen in pre-MegEs. The FOG-1 target gene Trib2²⁹, which mediates C/EBPα and C/EBPβ degradation^{30, 31}, was up-regulated in preMegEs, consistent with their proposed role in resolving the multi-potency of GE⁺ pre-GMs²⁹, and with the observed ability of FOG-1 to suppress eosinophil and mast cell lineage differentiation^{32, 33}. Sfpi1 (encoding PU.1) and Gata1 were expressed in GE⁺ pre-GMs, with Sfpi1 down-regulated in Mk-E-committed pre-MegEs. Gata1 and Sfpi1 were co-expressed in GE⁺ GMPs, indicating that down-regulation of GATA-1 is not decisive for commitment of GE⁺ pre-GMs to granulocyte lineages, and consistent with PU.1 being required for mast cell differentiation³⁴. Finally, key transcriptional regulators of monocyte-macrophage (Irf8, Klf4)³⁵ and neutrophil (Gfi1)³⁶ differentiation were highly expressed in LMPP and GE⁻ pre-GM and GMP populations compared to GE⁺ pre-GMs and GMPs (Fig. 6). These data identify LMPPs and GE+ preGMs as distinct multi-potent progenitors with non-overlapping lineage potentials, the former producing lymphocytes, neutrophils and monocuytes-macrophages, and the latter erythrocytes, platelets, eosinophils and mast cells.

Functional heterogeneity of GE⁻ pre-GMs

Lymphoid lineage potential has been associated with Flt3⁺ pre-GM³⁷. By flow cytometry we observed that Flt3⁺ pre-GMs were exclusively GE⁻ (Fig. 7a). The selective presence of lymphoid potential in Flt3⁺GE⁻ pre-GMs, but not Flt3⁻GE⁻ pre-GMs raised the possibility that other lineage potentials also differed between these two populations. Bulk and single cell cultures showed that both Flt3⁺ and Flt3⁻GE⁻ pre-GMs generated monocytes and neutrophils (Fig. 7b,c), with the Flt3⁻GE⁻ pre-GMs perhaps more biased towards neutrophil production. While mast cells did not develop from either progenitor population, a small number of eosinophils developed in the bulk GE⁻Flt3⁻ pre-GM cultures (Fig. 7b), and 7% of clones derived from single GE⁻Flt3⁻ pre-GMs showed combined neutrophil and eosinophil morphology (Fig. 7c). These results indicate that a minor, but distinct fraction of GE⁻Flt3⁻ pre-GM progenitor are restricted to neutrophil and eosinophil differentiation. qRT-PCR profiling of single pre-GMs confirmed expression of Gata1 in GE+ preGMs and of Flt3, with Sfpi1 mRNA expression being similar across the 3 pre-GM subsets (Fig. 7d-f), consistent with their myeloid progenitor identity. Mk-E and mast lineage-specific genes were co-expressed in >90% of single GE⁺ pre-GMs and antagonistic transcription factor pairs, such as Sfpi1-Gata1 and Cebpa-Zfpm1 were co-expressed with high frequency (32/48 cells and 17/48 cells, respectively) in single GE⁺ pre-GMs, suggesting the molecular properties of multi-potent progenitors. Co-expression of Sfpi1-Gata1 and Cebpa-Zfpm1 was not observed in GE⁻Flt3⁺ or GE⁻Flt3⁻ pre-GMs, which instead had high expression of monocyte-macrophage and neutrophil genes such as Irf8, Csf3r, consistent with the presence of these lineage potentials in both GE⁻ pre-GM sub-populations (Fig. 7g). In addition, the proportion of Flt3⁺ pre-GMs was decreased in Flt3 ligand null mice (Supplementary Fig. 5a,b), suggesting that Flt3 signaling regulates the Flt3⁺ pre-GM subset. In contrast, GE⁺ pre-GMs and GE⁺ GMPs were maintained in Gata1^fl/Y;Mx1-Cre conditional knockout mice after poly(I-C)–induced deletion (Supplementary Fig. 5c,d). Consistent with previous findings³⁸, Gata1 null GE⁺ progenitors differentiated into mast cells in vitro (Supplementary Fig. 5e,f), possibly due to redundancy with Gata2 in these progenitors. The gene expression of the identified myeloid progenitors is therefore consistent with their observed lineage potentials.

In vivo myeloid potentials of GE⁺ and GE⁻ progenitors

Finally, we investigated in vivo lineage potentials by co-transplanting Gata1⁺ and Gata1⁻ progenitor populations (CD45.2 and CD45.1/2) intrafemorally into non-irradiated, c-Kit hypomorphic W⁴¹/W⁴¹ recipients (CD45.1) (Supplementary Fig. 6a). Flow cytometry analysis (after 7 and 10-11 days) indicated that Lin⁻CD11b^loLy6G^hi neutrophils and Lin⁻CD11b^hiLy6G⁻c-Kit⁻FcεR1α⁻SiglecF⁻ monocytes were generated by Gata1⁻ GMPs and GE⁻Flt3⁺ pre-GMs, and Lin⁻CD11b^loLy6G⁻c-Kit⁻FcεR1α⁻SiglecF⁺ eosinophils by GE⁺ GMPs and GE⁺ pre-GMs, in both the spleen and bone marrow of reconstituted mice (Fig. 8a and Supplementary Fig. 6b,c). However, overall engraftment levels, in particular of eosinophils, were low, and no Lin⁻Ly6G⁻c-Kit⁺FcεR1α⁺ mast cells were detected. To overcome this obstacle, we tested the ability of GMP and pre-GM progenitors to engraft after intraperitoneal injection, and found that total engraftment of the peritoneum by both pre-GMs and GMPs was 1-2 orders of magnitude higher compared to that observed in bone marrow and spleen after intrafemoral transplant (Supplementary Fig. 6d), and also mast cell development was readily detected (Fig. 8b). Following intraperitoneal transplantation, GE⁺ GMPs generated exclusively peritoneal mast cells and eosinophils, whereas only monocytes and neutrophils were derived from GE⁻ GMPs (Fig. 8c). Similarly, GE⁺ pre-GMs produced predominantly mast cells and eosinophils, whereas GE⁻Flt3⁺ pre-GMs and LMPPs generated almost exclusively neutrophils and monocytes (Fig. 8d,e). The cellular phenotypes defined by flow cytometry were further validated by morphology (Supplementary Fig. 6e-h) and selective GE⁺ expression in eosinophils and mast cells (Supplementary Fig. 6i). Low numbers of neutrophils (but no monocytes) were produced by GE⁺ pre-GMs, and LMPPs and GE⁻Flt3⁺ pre-GMs generated limited numbers of eosinophils (but no mast cells), indicating that the more upstream myeloid progenitors (LMPPs and pre-GMs) retain some lineage plasticity, restricted to the eosinophil and neutrophil lineages in vivo, which is lost in both GE⁺ GMP and GE⁻ GMPs (Fig. 8c-e). These results confirmed the same lineage potentials in vivo as observed in vitro for both GE⁺ myeloid progenitors (GE⁺ preGMs, GE⁺ GMPs) and GE⁻ progenitors (LMPPs, GE⁻ preGMs, GE⁻ GMPs). We therefore propose to designate GE⁺ preGMs as candidate erythroid-megakaryocyte-primed multi-potent progenitors (EMkMPPs), and GE⁺ GMPs as eosinophil-mast cell progenitors (EoMPs). In contrast, since GE⁻ preGMs and GE⁻ GMPs generate predominantly neutrophils and monocytes, we propose to designate them pre-neutrophil-monocyte (preNM) and neutrophil-monocyte progenitors (NMPs), respectively. In different current hematopoietic models all monocyte-macrophage and granuolcyte lineage potentials are proposed to reside in one common GMP^{2, 8, 9, 10, 12, 39, 40} (Supplementary Fig. 7a,b). Instead our findings implicate the organization of GE⁺ progenitors (EMkMPP, EoMP) and GE⁻ progenitors (LMPP, preNM and NMP) into largely separate hierarchies that generate lymphoid cells, neutrophils and monocytes-macrophages, and eosinophils, mast cells, platelets and erythroid cells, respectively (Supplementary Fig. 7c,d).

Mouse lines

The Gata1-EGFP transgene was generated by bacterial artificial chromosome recombinant engineering in bacteria at the EMBL Genome Engineering Core Facility (http://www.embl.it/services/genome-engineering-service/index.html). In this transgene an EGFP-polyA cassette replaced the coding part of the second exon of the mouse Gata1 gene. Gata1-EGFP mice were generated by intracytoplasmic sperm injection⁵¹ at the EMBL Monterotondo Transgenic Facility (http://www.embl.it/services/transgenic_facility/index.html). Mice were backcrossed for more than 6 generations onto a C57Bl/6J background. Mice were bred and maintained at the animal facilities of Edinburgh University and Oxford University, UK. Experimental protocols were approved by the Edinburgh University School of Biological Sciences Ethical Review Committee, and the Oxford University Clinical Medicine Ethical Review Committee. Flt3l knockout⁵², Kit W⁴¹/W⁴¹ (https://www.jax.org/strain/000119), Gata1 conditional knockout mice⁵³ and Mx1-Cre transgenic mice⁵⁴ were previously generated. All experiments were performed under projects licenses from the UK Home Office. Mice were sacrificed for harvesting hematopoietic cell populations at an age of 6-17 weeks. Both male and female mice were used.

FACS and flow cytometry

BD LSRFortessa, LSRII , a FACS Aria II, AriaIII and FACSAria Fusion (BD Biosciences) were used for flow cytometry and cell sorting. Cells were prepared for cell sorting by either CD117 enrichment by autoMACS Pro Separator (Miltenyi Biotec)²⁶ or lineage depletion using immunomagnetic beads¹². In stainings where anti-FcγRII/III antibody was not included, cells were incubated with Fc-block. FlowJo analysis software (TreeStar) was used for subsequent data analysis. Antibodies used are listed below. Populations were defined as follows: HSCs: Lin⁻Sca-1⁺c-Kit⁺Flt3⁻CD150⁺Gata1-EGFP⁻, LMPP: Lin⁻Sca-1⁺c-Kit⁺Flt3^hiGata1-EGFP⁻, CLP: Lin⁻B220⁻Sca-1^loc-Kit^loFlt3⁺IL-7Rα⁺, MkP: Lin⁻Sca-1⁻c-Kit⁺CD150⁺CD41⁺, pre-GM: Lin⁻Sca-1⁻c-Kit⁺CD41⁻FcγRII/III⁻CD105⁻CD150⁻, GMP: Lin⁻Sca1⁻c-Kit⁺CD41⁻CD150⁻FcγRII/III⁺; preMegE: Lin⁻Sca-1⁻c-Kit⁺CD41⁻FcγRII/III⁻CD150⁺CD105⁻, preCFU-E: Lin⁻Sca-1⁻c-Kit⁺CD41⁻FcγRII/III⁻CD150⁺CD105⁺, CFU-E: Lin⁻Sca-1⁻c-Kit⁺CD41⁻FcγRII/III⁻CD150⁻CD105⁺. Eosinophils from the peritoneal cavity: Lin(CD4, CD5, CD8, Ter119, B220)⁻ CD11b^loLy6G⁻SiglecF⁺FcεR1α⁻F4/80⁻Gata1-EGFP⁺. Macrophages from the peritoneal cavity: Lin(CD4, CD5, CD8, Ter119, B220)⁻CD11b⁺Ly6G⁻SiglecF⁻FcεR1α⁻F4/80⁺Gata1-EGFP⁻. Neutrophils from peripheral blood: Lin(CD4, CD5, CD8, Ter119, B220)⁻CD11b^loLy6G⁺SiglecF^-FcεR1α⁻F4/80⁻. See also Transplantation studies.

Transplantation studies

In order to directly compare the in vivo potential of pre-GM and GMP progenitors subfractionated based on Gata1 expression, equal numbers of GE⁺ and GE⁻ cells were co-transplanted competitively into W⁴¹/W⁴¹ recipient mice (5-10 weeks of age) on a C57Bl/6 and CD45.1 background. By utilizing Gata1-EGFP^tg/+ mice on a CD45.2 or a mixed CD45.1/2 background, the contribution of the two competitor populations to the myeloid lineages could be separated and distinguished from each other endogenous cells based on CD45 allotype, and directly compared in a single recipient. Briefly, GE⁺ GMPs or GE⁺Flt3⁻ pre-GMs were sorted from a CD45.2 donor and mixed with equal cell numbers of GE⁻ GMPs or GE⁻Flt3⁺ pre-GMs, or vice versa. LMPP cells were isolated from either CD45.1/2 or CD45.2 Gata1-EGFP mice. Purified cells were either injected intraperitoneally (I.P.; 1000-4000 cells from each donor) or intra-femorally (I.F.; 2000-11000 cells from each donor) into W⁴¹/W⁴¹ CD45.1 mice. For I.P. transplantation, recipients were I.P. injected 12-24 hours before transplantation with 1.5ml of distilled water to deplete endogenous mast cells⁵⁵.

Contribution to the myeloid lineages was assessed by peritoneal lavage 4-20 days post-injections for I.P. injected mice and by analysis of injected femur and spleen 7 and 10 days post-injection for I.F. injected mice. The contribution of CD45.2 and CD45.1/2 donor cells to mast cells, eosinophils, monocytes and neutrophils was assessed by flow cytometry using the following markers: mast cells: Lin⁻Ly6G⁻c-Kit⁺Fcer1α⁺; eosinophils: Lin⁻CD11b^loLy6G⁻c-Kit⁻Fcer1α⁻SiglecF⁺; neutrophils: Lin⁻Cd11b^loLy6G^hi; monocytes: Lin⁻CD11b^hiLy6G⁻c-Kit⁻Fcer1α⁻SiglecF⁻. Lin: CD4, CD5, CD8a, CD19, Ter119. The contribution of the individual donors (CD45.2 or CD45.1/2) to each lineage was adjusted to the number of cells injected and number of total mononuclear cells analyzed, and expressed as cells per million mononuclear cells per 1000 transplanted cells analyzed. Average values below 2 events per 1 million cells analyzed per 1000 cells injected were not above the non-specific background, and therefore not used to evaluate lineage-contribution of donor-derived cells.

Gata1 knock out studies

Chimeric conditional Gata1^cKO mice were established by co-transplanting 0.5x10⁶ Gata1^cKO (Gata1^fl/yMx1-Cre^tg/+Gata1-EGFP^tg/+, CD45.2) and 0.5x10⁶ wild type bone marrow cells (CD45.1) into lethally irradiated CD45.1/2 recipients. For the non-deletion control, Gata1^fl/yMx1-Cre^+/+Gata1-EGFP^tg/+, CD45.2 were used. After 6 weeks, engraftment was established by analysis of CD45 allotypes in peripheral blood, and the CD45.2 cells contributed between 30-60%. To induce deletion of the Gata1 gene by CRE, mice were treated with 225µg poly(I-C) 8-10 weeks after transplantation by 3 intra-peritoneal injections every other day. Since Gata1 is essential for differentiation to erythroid progenitors and erythrocytes²⁹, Gata1 deletion was confirmed 8-9 weeks after poly(I-C) treatment by absence of CD45.2 CFU-E cells or EGFP signal in erythroid cells.

In vitro cultures

To test myeloid potential, cells were cultured in IMDM with L-glutamine (Invitrogen), 20% heat inactivated fetal calf serum (Gibco), penicillin/streptomycin (Invitrogen) and 10^-4M β-mercaptoethanol (Sigma), supplemented with 20ng/ml mSCF (Peprotech), 20ng/ml mIL-3 (Peprotech), 10ng/ml mGM-CSF (Peprotech), 50ng/ml mIL-5 (R&D Systems) and 50ng/ml mIL-9 (R&D Systems) at 37°C, 5% CO₂. Bulk cultures of LMPP and pre-GM were started with 100-20.000 cells per culture. Myeloid potential of single cells was tested by sorting single cells directly into Terasaki plates, unless stated differently in the text, each well containing 20μl medium. GMP cells were derived in vitro by culturing progenitors (pre-GM or LMPP) for 3 days with 100ng/ml mSCF and 10ng/ml mIl-3 in round bottom 96 well plates and FcγRII/III⁺ cells were in this experiment considered as GMPs. In vitro derived GMPs were manually plated at an average of 5 cells/well into 60 well Terasaki plates. For eosinophil favouring culture conditions, GMP cells (200 cells/culture) were grown with 20ng/ml mSCF, 10ng/ml mGM-CSF and 50 ng/ml Il-5 in round bottom 96 well plates. After 5 days, the cells were washed and put back in culture for another 2 days without mSCF and mGM-CSF. All cytospins were prepared with a Shandon cytospin at 1000 RPM with low acceleration, followed by May-Grünwald-Giemsa stain (VWR).

Immunocytochemistry

Air dried cytospins were fixed with 4% PFA in PBS for 5 minutes. Slides were first blocked with PBS/5%BSA containing Fc-block for 30 min at RT and then with 3% H₂O₂ in water for 5 minutes. Slides were incubated with chicken polyclonal anti-GFP (Abcam, ab13970) in PBS/5% BSA for 4 hours at room temperature and next with peroxidase donkey anti-chicken antibody (Stratech, 703-035-155-JIR) for 1 hour at room temperature. Peroxidase was detected with DAB substrate kit (Vector Laboratories, SK-4100) according to manufacturer’s instructions. Cells were counterstained with Harris Modified Hematoxylin Solution (Sigma-Aldrich, HHS32)) for 10 minutes.

Colony forming assays

For evaluation of erythroid potential from progenitor populations, cells were seeded in Methocult M3436 and myeloid potential was evaluated with Methocult M3534. Colonies were scored after 8 days of culture. CFU-Mix colonies were scored in Methocult M3434 based on colonies containing red cells after 8 days of culture. Megakaryocytic potential was evaluated using the Megacult collagen-based assay. All these media were from StemCell Technologies, Vancouver, Canada. Megakaryocyte cultures were supplemented with 10ng/ml mIL-3, 20ng/ml hIL-6 (Peprotech), 50ng/ml hIL-11 (Peprotech), 50ng/ml hThpo (Peprotech), and megakaryocytes detected using acetylthiocholiniodide staining (Sigma) according to manufacturer’s instructions after 7-8 days culture. For evaluation of Mk potential from single GE⁺ and GE⁻ Lin⁻Sca-1⁺c-Kit⁺Flt3^hi cells, these cells were sorted into X-vivo15 medium containing glutamax and gentamycin (BioWhittaker) supplemented with 10% fetal calf serum (Hyclone), 10^-4M β-mercaptoethanol, 50ng/mL mSCF, 50ng/mL hFLT3 ligand (hFL; Immunex), 50ng/mL hThpo, 20ng/mL mIL-3 and 5U/ml hEPO (Roche). For each experiment, 240 cells were manually plated at 1 cell per well into 60 well Terasaki plates. Mk potential was evaluated 8 or 12 days after culture using an inverted microscope.

Measurement of lymphoid lineage potentials

B and T lymphocyte potentials of purified progenitor populations were evaluated following co-culture on OP9 and OP9-DL1 stroma cells, respectively (kindly provided by Dr. A. Cumano, Pasteur Institute, Paris, France). LMPPs, GE⁺ and GE⁻ pre-GMs, and GE⁺ and GE⁻ GMPs were seeded onto stromal monolayers. OP9 cultures were supplemented with 25ng/ml mSCF, 25ng/ml hFL (Peprotech) and 10ng/ml hIL-7 (Peprotech), while OP9-DL1 cultures were supplemented with 25ng/ml mSCF and 25ng/ml hFlt3L. SCF was removed after 7 days of culture. Readout was by FACS at 14-21 days of culture. B cells were defined as CD19⁺ and T cells were defined as CD4⁺CD8⁺ and/or Thy1.2⁺CD25⁺ (Supplementary Fig. 4e,f).

Quantitative PCR

For gene expression analysis of bulk cultures and cultured cell populations, RNA was purified from 2000-6000 cells per replicate with Trizol (Invitrogen, Carlsbad, CA, USA) and reverse-transcribed with SuperScript VILO cDNA synthesis kit according to the manufacturer’s instructions (Invitrogen, Carlsbad, CA, USA). For single cells and progenitor populations (200 cells per replicate), CellsDirect™ One-Step qRT-PCR kit (Life technologies, 11753-100) was used according to manufacturer’s protocol for preparation and amplification of cDNA. The BioMark 96.96, 48.48 or 192.24 Dynamic Array platform (Fluidigm) and either Taqman® assays (for progenitor populations and single cells) or the Universal probe library system (Roche) (for bulk cultures and cultured cell populations) were used according to the manufacturers’ instructions. UPL primers and probes were chosen by the Assay Design Center web service for mouse genes. Default parameters were used and highest-ranking primers-probe combination was chosen for each gene.

Analysis of quantitative PCR of single cells

ΔCt values relative to Hprt (Ct(Hprt) - Ct(Gene)) were zero centered for each gene by subtraction of the mean value of all positive cells for the gene. These normalized values were used to generate heat maps using GENE-E software.

Microarray gene profiling and bioinformatics analysis

Global gene expression was analyzed in FACS purified HSCs, LMPPs, CLPs, EGFP⁺ and EGFP⁻ pre-GM, GE⁺ and GE⁻ GMPs, and pre-MegEs. For all cell populations, 3-4 individually sorted samples from different pools of mice were prepared. For each sample, 5000 cells were sorted directly into RLT buffer (Qiagen) and RNA was extracted according to the manufacturer's instructions (Qiagen). Microarray analysis was performed by the Paterson Institute Microarray Service, Manchester UK, using MoGene 1.0ST arrays. Data were RMA normalized⁵⁶. Custom R scripts were used to perform principal component analysis (PCA) on 2,305 genes that showed high variation of gene expression across populations (ANOVA with adjusted P values<0.05 and coefficient of variation (CV) equal to or larger than 0.4). GSEA²⁷ was performed using CLP and preMegE gene sets previously described⁶.

Single cell RNA sequencing; cell capture and preparation of cDNA

5000 pre-GM cells from wild type C57BL/6 male mice (CD45.1) were sorted in 2 μl PBS/50%FCS. Single cells were captured on a small-sized (5-10μm cell diameter) C₁™ Single-Cell Auto Prep IFC for mRNA Sequencing (Fluidigm) using the Fluidigm C1 system. Cells were loaded onto the chip at a concentration of ~300 cells/μl and imaged by phase-contrast microscopy to check single cell per capture site. 64 cells were captured and used for RNA-seq. Cells were lysed and cDNA prepared on the C1 fluidigm chip according to manufacturer’s protocol, using SMARTer Ultra Low RNA kit for Illumina (Clontech).

Single cell RNA sequencing; library construction

Single-cell Library size distribution was checked on a High-Sensitivity DNA chip (Agilent Bioanalyzer) (average concentration 239pg/μl, SD 149pg/μl). 1ng of cDNA was used for tagmentation with Nextera XT DNA Sample Preparation kit (Illumina). Purification was done with a 1:1 ratio of AMPure XP beads with a final elution in 25μl of Resuspension Buffer (Nextera). Samples were loaded on a High-Sensitivity DNA chip to check the size and quality of the indexed library (average size 612bp, SD 104), while quantification was done with Qubit High-Sensitivity DNA kit (Invitrogen) (average concentration 347ng/ml SD 95). Libraries were pooled to a final concentration of 10nM and were sequenced with Illumina HiSeq 2000 (single end 50 basepair reads) at The Wellcome Trust Centre for Human Genetics at the University of Oxford.

Single cell RNA sequencing; data analysis and visualization

51 bp reads were aligned to the murine genome (mm9) using STAR⁵⁷ and default parameters. See (Supplementary Fig. 8) for the information about the number of uniquely mapped reads, percentage of mapping, and the number of detected genes per each single cell). Non-uniquely mapped reads were discarded. Transcript expression values were quantified using Cufflinks⁵⁸ as fragments per kilobase of transcript per million mapped reads (FPKM). Overlapping transcripts were collapsed giving the highest one expression value per gene loci. One cell was removed from the analysis due to the low detection number of genes (sample81, Supplementary Fig. 8). Custom R scripts were used to perform PCA, hierarchical clustering, and visualization on the log2 of FPKM scale. In brief, PCA analysis was performed on all cells using all genes, which have the average expression values >=1. We selected top 100 genes with the highest absolute correlation coefficient (PCA component loadings) in one of the first three components for the hierarchical clustering analysis. Genes differentially expressed between the two main cellular were identified using ANOVA (P<0.001; FDR<0.05) and used for the final clustering of single cells.

Statistical analysis

Student’s t-test was used to determine statistical significance for normally distributed data. For single cell expression levels the Mann-Whitney U-test was used, and Fisher’s exact test was used to compare expression frequencies at the single cell level between populations. No statistical method was used to predetermine sample size, and experiments were not randomized. No data were excluded from analysis. The Investigators were not blinded to allocation during experiments and outcome assessment.