Perspective: RUNX1 and the endothelial origin of blood

Abstract

The transcription factor RUNX1 is required in the embryo for formation of the adult hematopoietic system. Here we describe the seminal findings that led to the discovery of RUNX1 and of its critical role in blood cell formation in the embryo from hemogenic endothelium. We also present RNA-Seq data demonstrating that hemogenic endothelial cells in different anatomic sites, which produce hematopoietic progenitors with dissimilar differentiation potentials, are molecularly distinct. Hemogenic and non-hemogenic endothelial cells in the yolk sac are more closely related to each other than either are to hemogenic or non-hemogenic endothelial cells in the major arteries. Thus, a major driver of the different lineage potentials of the committed erythro-myeloid progenitors that emerge in the yolk sac, versus hematopoietic stem cells that originate in the major arteries, is likely to be the distinct molecular properties of the hemogenic endothelial cells from which they are derived. We use bioinformatics analyses to predict signaling pathways active in arterial hemogenic endothelium, which include the functionally validated pathways Notch, Wnt, and Hedgehog. We also use a novel bioinformatics approach to assemble transcriptional regulatory networks and predict transcription factors that may be specifically involved in hematopoietic cell formation from arterial hemogenic endothelium, which is the origin of the adult hematopoietic system.

History

The discovery of RUNX1

An account of the developmental origin of blood cannot be told without the major protagonist of the story, RUNX1. RUNX was first discovered in the fly, in the classic screen conducted by Nusslein-Volhard and Wieschaus to identify mutations that affect development [1]. The mutation runt was named for the fact that it resulted in runted embryos - a defect caused by pre-segmentation patterning defects. The runt gene was cloned by Gergen and colleagues, and the protein it encoded shown to be nuclear, but its identity as a transcription factor was not discovered at that time [2].

The human RUNX1 gene was the next to be cloned by Ohki and colleagues, as one of a pair of genes disrupted by the (8;21)(q22;q22) translocation in acute myeloid leukemia (AML) [3]. Homology to the Drosophila runt gene was noted [4], but since the biochemical function of the Drosophila runt protein was not known, the function of the human homologue remained a mystery. Not long after, however, three labs using two different approaches uncovered RUNX1’s function. Our group (Speck) purified RUNX1 as a sequence-specific DNA binding protein that regulated the disease specificity of a mouse retrovirus [5]. The team of Ito and Shigesada purified RUNX1’s homologue RUNX2 based on its role in murine polyomavirus replication and transcription [6]. Both groups showed that the purified transcription factors consisted of two subunits, one that bound DNA directly (RUNX1 or RUNX2), and a common non-DNA binding subunit called core binding factor β (CBFβ) that increased the affinity of RUNX1 and RUNX2 for DNA [6–8]. The name “core binding factor” (CBF) derives from the DNA motif in the polyomavirus and retrovirus enhancers to which the RUNX proteins bound, which had previously been named “Core” [9]. At around the same time RUNX1 and CBFβ were purified and cloned, the Hiebert lab used a “selection and amplification binding” technique to determine whether the human RUNX1 protein bound DNA and, if so, which DNA sequence it recognized [10]. The Hiebert lab identified the same DNA sequence that was used by our lab and the Ito/Shigesada labs to purify the proteins. Closing the circle, Liu et al. showed that the inv(16)(p13.1;q22) associated with AML created a chimeric protein that fused the non DNA binding CBFβ subunit to the coiled-coil tail region of a smooth muscle myosin heavy chain [11]. Hence multiple lines of investigation converged, linking RUNX1 to CBFβ, the t(8;21) to the inv(16), and human leukemia to mouse leukemia. These discoveries are a great example of the major contribution the study of viruses and model organisms made to our understanding of human disease.

A role for RUNX1 in the embryonic origin of blood

Of all of the paths that led to the discovery of RUNX1 and CBFβ, only the chromosomal translocations hinted at an essential role at the earliest stages of blood cell formation. As background, hematopoiesis in the embryo unfolds in three waves, and both RUNX1 and CBFβ are required in the last two waves. The first, primitive wave produces primitive erythrocytes, diploid megakaryocytes, and primitive macrophages, all of which differentiate from mesoderm in the yolk sac blood islands beginning at embryonic day (E) 7.25 in the mouse [12–14]. Wave 2 consists of the first “definitive” progenitors, which include erythro-myeloid progenitors (EMPs) that emerge in the yolk sac beginning at E8.25 [12, 15], and lymphoid progenitors that appear at E9.5 in the yolk sac, and in the caudal part of the embryo in the dorsal aorta, vitelline and umbilical arteries [16–21]. Wave 3, the final wave of blood formation, includes pre-hematopoietic stem cells (pre-HSCs) that are unable to engraft adult mice directly, but colonize the fetal liver where they mature into adult-repopulating HSCs [22–27]. Wave 3 also includes a small number of adult-repopulating HSCs in the dorsal aorta, vitelline and umbilical arteries, and in the placenta, which presumably have matured in situ from pre-HSCs [24, 26, 28–34].

Knockouts of RUNX1 and CBFβ resulted in the absence of all wave 2 and 3 progenitors, including adult-repopulating HSCs [35–40]. To gain insight into the nature of this hematopoietic block, our lab introduced a lacZ reporter gene into the Runx1 locus to learn where RUNX1 was expressed in the blood lineage [41]. We found RUNX1 only in the relatively rare wave 2 and 3 progenitors in the embryo, which are far outnumbered by primitive erythrocytes that are initially RUNX1⁺ but quickly lose RUNX1 expression. This restriction of RUNX1 expression to wave 2 and 3 progenitors (and also to wave 1 macrophages) was actually a key feature that enabled us to pinpoint their origin. Some RUNX1⁺ wave 2 and 3 progenitors could be found in the circulation, and in the fetal liver, but we also found clusters of round RUNX1⁺ hematopoietic cells attached to RUNX1⁺ endothelial cells in arteries, specifically in anatomic sites where the first functional wave 2 and 3 progenitors, and histological evidence of blood cell formation had been described [17, 24, 32, 42–51]. Most importantly, we showed that RUNX1 loss blocked the formation of wave 2 and 3 progenitors from this rare population of “hemogenic” endothelial cells.

The concept that blood formed from hemogenic endothelium had been proposed in the early 20^th century [44, 52, 53], and just prior to our work several groups showed that blood could differentiate from endothelial cells [54, 55]. Despite these compelling data, the notion that hemogenic endothelium was the immediate precursor of blood was not widely accepted at the time of our discovery. Our identification of RUNX1 as the first specific marker of hemogenic endothelium, and demonstration that RUNX1 was required for blood cell formation from hemogenic endothelium [41, 56, 57], lent considerable support to the notion that hemogenic endothelial cells were the immediate precursor of the adult hematopoietic organ. Later, live imaging studies performed by multiple laboratories showed blood cells forming from endothelial cells in real time, leading to the consensus that endothelial cells are the immediate precursor of blood cells [58–61].

Heterogeneity of hemogenic endothelium

An interesting question is how different populations of hemogenic endothelial (HE) cells produce embryonic blood progenitors with distinct properties. For example, HE cells in the yolk sac primarily give rise to EMPs, whilst in the major arteries HE cells give rise to pre-HSCs and HSCs. We hypothesize that differences in the lineage potential of hematopoietic stem and progenitor cells (HSPCs) produced in the yolk sac and major arteries likely originates, at least in part, from differences in the intrinsic properties of the HE cells from which they arise. Several lines of evidence support this hypothesis. For instance, HE cells that give rise to EMPs are both arterial and venous in origin, whereas pre-HSCs differentiate from arterial HE [62–64]. Second, different signaling pathways are involved in EMP and pre-HSC formation. For example, EMP formation from HE in the yolk sac vascular plexus does not require Notch signaling, while HSPC production from arterial endothelium is strictly dependent on Notch [65–68]. Similarly, inflammatory signaling regulates the numbers of HSPCs produced in the major arteries, but appears to have no influence on EMP numbers in the yolk sac [69–72].

To determine whether HE cells in different anatomic sites are indeed molecularly distinct, we purified HE cells and non-hemogenic endothelial (E) cells from the yolk sac and major arteries at two different embryonic stages, embryonic day 9.5 (E9.5) and E10.5, and determined their transcriptomes by RNA-Seq. We isolated major arteries (dorsal aorta, umbilical, vitelline) by dissecting the caudal region of the mouse embryo from which the head, heart, pulmonary regions, liver, gut tube, tail and limb buds were removed. We used green fluorescent protein (GFP) expression from a Runx1:GFP knockin allele [73] as a tool to separate HE from E. We purified HE from the arteries and yolk sac as Runx1:GFP⁺ CD31⁺ Kit^lo/- CD45⁻ CD41⁻ cells, and E as Runx1:GFP⁻ CD31⁺ Kit⁻ CD45⁻ CD41⁻ cells (Fig. 1A,B). We confirmed the functional purity of the cells in ex vivo assays. Both HE and E cells could form endothelial tubes in culture (Fig. 1C). On the other hand, endothelial cells with the ability to form CD45⁺ blood cells in culture were highly enriched in the purified Runx1:GFP⁺ HE population (the frequency was 500x higher in HE versus E from the arteries, and 100X higher in HE versus E from the yolk sac, Fig. 1D). Purified HE and E cells produced very few hematopoietic colonies in methylcellulose cultures (0.5–1.5 colony forming units per embryo equivalent) (Fig. 1E), indicating that the HE cells were not significantly contaminated by HSPCs.

Figure 1. Isolation and functional characterization of hemogenic endothelial (HE) and endothelial (E) cells from the arteries and yolk sac.

A) FACS gating strategy to purify HE and E cells from the arteries. Shown are profiles from E10.5 embryos.

B) Purification of HE and E from yolk sacs (YS), E10.5.

C) Frequency of cells that produced endothelial tubes on OP9 cells cultured with vascular endothelial growth factor (VEGF). Data from E10.5 HE and E are shown. Average frequencies are indicated above the bars. Significance determined by ANOVA and Tukey’s multiple comparisons test (mean ± 95% CI, *P < 0.05, **P < 0.01, n=3 experiments).

D) Frequency of HE and E cells that produced CD45⁺ hematopoietic cells following 8–10 days of culture on OP9 stromal cells in the presence of stem cell factor (SCF), interleukin 3 (IL-3), FLT3 ligand (Flt3L) and IL-7. Data from E10.5 HE and E are shown. Significance determined by ANOVA and Tukey’s multiple comparison tests (mean ± 95% CI, **P < 0.01, n=3 experiments).

E) Methylcellulose assays to enumerate colony forming units - culture (CFU-C) in sorted populations of endothelial cells, and unfractionated yolk sac cells as a control, represented as the number of CFU-Cs per embryo equivalent. Data from E10.5 samples shown. Significance determined by ANOVA and Tukey’s multiple comparison tests (mean ± SD, P < 0.0001, n = 4–8 samples, data combined from 3 experiments).

We bulk sequenced approximately 20,000 cells, obtaining ~74.7 million uniquely mapped reads per sample (Accession: GSE103813). The RNA-seq data is highly reproducible, with an average transcriptome-wide correlation of biological replicates of 0.95 (not shown). Principle component analysis (PCA) of the HE and E RNA-seq data, along with data from E14.5 fetal liver (FL) HSCs (Lin⁻Sca1⁺Kit⁺CD48⁻CD150⁺) and adult bone marrow (BM) HSCs (Lin⁻Sca1⁺Kit⁺CD34⁻CD48⁻CD150+) clustered HE and E far away from FL and BM HSCs (Figure 2A). The tissue of origin was the strongest driver of clustering, with HE and E cells from yolk sacs clustered closely with each other, and less closely with their equivalent populations in the arteries. Similarly, HE and E from the major arteries clustered more closely to each other than to the corresponding cells in the yolk sac. Therefore, HE and E from the major arteries are molecularly more closely related to each other than either is to HE or E from the yolk sac. We used EBSeq [74] to identify differentially expressed genes between pairs of endothelial populations. Consistent with the result of the PCA analysis, there were more differentially expressed genes between HE in the yolk sac and arteries, than between HE and E from a single anatomic site (Fig. 2B, Supplemental Table 1). The close relationship of HE and E cells from the major arteries was also reported by Baron et al. in their recently published single cell RNA-seq study [75].

Figure 2. Global comparison of the transcriptomes of HE and E.

A) Principle component analysis of the transcriptome data. Also shown are fetal liver (FL) HSCs and adult bone marrow (BM) HSCs for comparison. The t-distributed stochastic neighbor embedding (t-SNE) algorithm [98] yielded the same result (not shown).

B) Number of differentially expressed genes based on pairwise comparisons (false discovery rate of 0.05 and fold change of 1.5).

C) Enriched pathways among differentially expressed genes from each pairwise comparison of E9.5 HE and E (false discovery rate of 0.05).

D) A novel computational framework for condition-specific transcriptional regulatory network (TRN) construction and identification of key transcription factors. Nodes in the TRN represent genes, and edges represent regulatory relationship between TFs and target genes. The framework first constructs a consensus TRN for each cell type by using five methods, CLR (the context likelihood of relatedness) [99], a method based on Pearson correlation [99], GENIE3 (gene network inference with ensemble of trees) [100], Inferelator [101], and TIGRESS (trustful inference of gene regulation using stability selection) [102]. Next, the framework prioritizes key TFs based on their potentials of regulating the entire set of differentially expressed genes between the two TRNs compared. Details are provided in the Supplemental Methods.

E) Prioritized key TFs for E9.5 HE (artery HE plus YS HE) versus E (artery plus YS E). Shown in blue are the four TFs used by Sandler et al. and Lis et al. to convert endothelial cells into blood cells [82, 83]. Analysis of comparable E10.5 samples also prioritized Pitx1, Spi1, Runx1, Tead3, Rhox6, Tbx2, Hox8, Hbp1, Mxd1, Znf512B, and Hoxc8 (not shown).

F) Prioritized key TFs for E9.5 artery versus YS HE. TFs prioritized in both E9.5 and E10.5 artery versus YS HE include Tead2, Zfp423, Mecom, Sox7, Msx1, Etv4, Rcor2, Mycl, Tbx3, Twist1, Rhox6, Hoxa4, Tcf3, and Tcf7 (not shown).

G–I) Expression levels of prioritized TFs in corresponding comparisons of E9.5 HE and E. FPKM, Fragments Per Kilobase of transcript per Million mapped reads.

We performed pathway enrichment analysis among differentially expressed genes for each comparison of E9.5 endothelial cell populations. For simplicity sake we focused on the earlier time point, E9.5, which is the peak of EMP formation in the yolk sac, and is the first time RUNX1⁺ HE cells are abundant in the arteries [12, 63] (Fig. 2C). The Notch pathway was significantly enriched in the arterial HE versus E comparison but not in yolk sac HE versus E. Additional signaling pathways known to be involved in HSPC formation from the arteries such as Wnt, hedgehog, retinoid acid, and hypoxia [76–80] were predicted to be more active in arterial HE than in yolk sac HE.

Cell-type-specific transcriptional regulatory networks and key transcriptional regulators

To identify novel transcription factors (TFs) regulating HE formation, we developed a computational method for constructing condition-specific transcriptional regulatory networks (TRNs) and identifying key TFs using the constructed TRNs. By comparing a pair of cell-type-specific TRNs, we could then predict key TFs that mediate differential expression between the two TRNs (Fig. 2D). We identified 38 TFs in the comparison of E9.5 HE to E (combining YS and artery samples for each cell type), and 40 TFs in the artery HE versus yolk sac HE comparison (Fig. 2E–I, Supplemental Methods). GFI1 and RUNX1, both of which are essential for HSPC formation from HE [41, 81], are among the top ranked TFs in the HE versus E comparison, supporting our bioinformatics approach (Fig. 2E). SPI1 and FOSB, which together with RUNX1 and GFI1 were shown to be sufficient to reprogram endothelial cells into hematopoietic cells [82, 83] were also highly ranked. Other significantly ranked TFs in the HE to E comparison include the TEAD factors, known to regulate hematopoietic specification via the Hippo pathway [84]. TFs with known roles in hematopoiesis, but as of yet no reported function in embryonic HSPC formation, including Arid3b, Ikzf4, and Bcl11a [85–87] were also highly ranked. A similar comparison of E10.5 HE to E resulted in a list of 34 TFs, 10 of which overlapped with TFs predicted to be active at E9.5 (see Fig. 2E, legend). Several top-ranked TFs in the E9.5 artery versus yolk sac HE comparison (Fig. 2F), including Tcf7, Gata3, Hey1, and Hey2, are direct Notch targets [88–90], consistent with the important role for Notch signaling in arterial specification and hematopoiesis. The E9.5 artery HE versus yolk sac HE comparison also predicts that multiple Hox family genes (Hoxa1, Hoxa4, Hoxa10 and Hoxc10) may be in arterial HE. Meis1, shown to play an important role in embryonic HSPC formation [91, 92], was also highly ranked in the comparison of arterial versus yolk sac HE. Meis1 was recently shown to be critical for the transition of HE into progenitor cells in human induced pluripotent stem cell cultures [93].

In conclusion, the study of RUNX1 has provided critical insights into the embryonic origins of blood, and continues to be an extensively used marker of HE. Reporters based on RUNX1 expression have been useful for isolating and studying the generation of HE, and for analyzing the differentiation of HSPCs from HE [41, 73, 94–97]. Here we used expression of GFP from the endogenous Runx1 locus to isolate and compare HE from different anatomic sites. We provide molecular data confirming that HE cells from different sites of HSPC formation are molecularly distinct. These data will be a useful resource for investigators aiming to generate specific subsets of HE in culture, including arterial HE that is the source of pre-HSCs and HSCs.

Highlights.

• Discovery of RUNX1 and its role in hemogenic endothelium.

• Hemogenic endothelium from different anatomic sites is molecularly distinct.

• Novel bioinformatics approach predicts transcription factors in hemogenic endothelium