Hematopoietic stem cells (HSC) form the basis of the adult hematopoietic system. Since the pioneering experiments performed by Till & McCulloch in 1961 (1) it has been known that HSC are cells capable of self renewal and – at the same time and under different influences – are able to give rise to all blood cell lineages. In the adult, HSC are maintained, divide and differentiate in the supportive microenvironment of the bone marrow, the HSC niche (2). Since the first reports on blood cells more than a century ago, a corpus of data has been collected on the spatiotemporal emergence and molecular control of developmental hematopoiesis. Pioneering studies addressing the emergence of the hematopoietic system were performed in the avian embryo whose large size, easy access and topology flat on the yolk make it highly suitable for observation and manipulation. The work was then extended to frogs and mammals i.e., mice and humans. Recently the bony fish Danio rerio has been increasingly used to decipher the molecular pathways involved in the development of the hematopoietic system (3–5). In parallel to these in vivo systems, valuable insights have been obtained from embryonic stem (ES) cell cultures (6–9). Taken together, these data have revealed that the developmental programs of blood and endothelium are closely related and remarkably conserved among species. However, several questions still remain regarding the origin(s) of definitive HSC, the cells at the origin of the adult haematopoietic system. For example, the nature of the cell(s) giving rise to the definitive hematopoietic cell types during development has not been fully elucidated. Most of the current data suggest that HSC emerge from a specific subset of endothelial cells that have the capacity to generate blood i.e., the hemogenic endothelium. Other data suggest that HSC emerge de novo from mesenchymal cells. In this review we aim to give a brief overview of the development of the hematopoietic system in vertebrates and we discuss recent data obtained from chick, mouse and human models on the origins of HSC.
The embryo harbors the source of definitive HSC
Based on observations performed a century ago on avian embryos, it was proposed that the yolk sac was pivotal in the production of hematopoietic cells (HC). Yolk sac-associated hematopoiesis mainly gives rise to erythroid cells easily visible due to the presence of hemoglobin. Several authors have, however, at the same time reported the presence of cells, interpreted as being of hematopoietic origin, associated with the ventral aspect of the embryonic aorta. Fifty years later, the role of the yolk sac in the production of HC was analyzed in a series of experiments conducted by Moore, Owen & Metcalf (10, 11) which constituted the foundation of the hematogeneous theory according to which HSC were all produced in the yolk sac at an early stage. HSC then migrate into the hematopoietic organ rudiments i.e., the fetal liver, spleen and bone marrow. Thus, these organs did not harbor their own hematopoietic cells but were subject to secondary colonization by yolk sac-derived cells, the bone marrow being the life-long reserve (12). Soon after this view was adopted, F. Dieterlen and co-workers initiated a series of experiments that challenged this notion of yolk sac origin. These experiments relied on the use of the quail/chick system (13) combined with the accessibility and topology of the avian embryo. They consisted in a homotopic grafting of a quail embryo on a chick yolk sac (14), an experimental scheme designated as “yolk sac chimera” (15). Contrary to the centripetal flow of HC postulated by Moore & Owen, hematopoietic organs were seeded exclusively by quail cells that even seeded the yolk sac by embryonic day (E)5 (16, 17). On E13, as many as 70% of circulating blood cells were of quail origin in some chimeras (17). Similar associations between congenic strains of chickens differing either by their sex chromosomes (18), their immunoglobulin haplotypes (19, 20) or major histocompatibility antigens (21) led to even more striking conclusions. Yolk sac cells decreased rapidly from E5 and were absent at hatching. Thus, yolk sac progenitors had proved incapable of long-term renewal and the “definitive” HSC colonizing the hematopoietic organs had to originate from the embryo proper.
These experiments also revealed that the yolk sac contributed two distinct generations of red cells: a primitive one that derives entirely from the yolk sac and a second one (fetal) that derives both from the embryo and the yolk sac, the latter being able to generate erythrocytes with a fetal globin make-up (17). Finally, using reversed yolk sac chimeras (chick embryos onto a quail yolk sac) combined with the use of the cell type- and species-specific monoclonal antibody QH1/MB1 (22, 23) that recognizes quail endothelial cells (EC) and HC, it was shown that a small HC population of yolk sac origin differentiated into primitive macrophages found in zones of apoptosis, notably in neural derivatives. They may represent the first microglial cells (24, 25). Such a yolk sac contribution to the early macrophage population was also demonstrated in the mouse embryo (26).
Aiming at identifying the intra-embryonic site(s) where adult-type HC were produced, Dieterlen-Lièvre and co-workers reported the presence of cell clusters, tightly associated with the ventral aspect of the aorta protruding into the aortic lumen. Their hematopoietic nature was confirmed by means of MB1/QH1 (23) and anti-CD45 (27) antibodies. The aortic region, also designated in mammals as the Psp/AGM (para-aortic splanchnopleura/aorta-gonad-mesonephros i.e., the aorta and the surrounding tissues at two successive stages of development), was shown to harbor cells giving rise to erythroid, monocytic, myeloid or multipotential colonies (28–30) as well as transplantable HSC (31).
Hemangioblast or hemogenic endothelium?
Hemangioblasts of the yolk sac. When yolk sac blood islands form, HC and EC emerge at the same time in close association. Proximity between these two cells types has prompted earlier embryologists to hypothesize the existence of an ancestral progenitor between EC and HC, called the hemangioblast (32). This hypothesis has been reinforced by the fact that EC and HC share many markers and that several gene mutations and deletions in the zebrafish and mouse embryos affect both cell types. Experimental data in the avian embryo, however, argue against its existence. The avian orthologue of VEGF-R2, Quek-1, (33) was shown to be expressed in the posterior primitive streak, a region previously shown to give rise to the yolk sac mesoderm that further differentiates into EC and HC (34). In order to identify hemangioblasts, Eichmannn and co-workers (35) sorted out Quek-1+ cells from the very early blastodisc and submitted the positive cell population to clonal differentiation assay. In the absence of growth factors, VEGF-R2+ cells gave rise to hematopoietic colonies. When VEGF was added, endothelial colonies differentiated while the number of hematopoietic colonies decreased substantially. Mixed colonies never developed, precluding the demonstration of a common precursor. Recent data, however, identified a Flk1+ population in the posterior primitive streak of E7–7.5 mouse embryos, supporting the existence of a clonal progenitor for HC and EC in vivo (36). The location and time of appearance of these cells are in agreement with a previous mapping experiment (37) and suggest that they are progenitors of the yolk sac. Intriguingly, lineage tracings have failed to identify a cell with EC and HC potential (37). The use of single cell tracing experiments will undoubtedly help to answer the question of the existence of this progenitor in vivo.
The concept of hemogenic endothelium. The proximity between EC and HC clusters in the aortic floor has suggested that the definitive HC, including HSC, are generated through an endothelial intermediate endowed with the potential to give rise to HC i.e., a hemogenic endothelium. Unlike what happens in the yolk sac where EC and HC appear simultaneously, the aorta is formed at least one day before the emergence of the hematopoietic clusters, suggesting the existence of an aorta-specific type of hemangioblast. An alternative hypothesis is that hematopoietic cells originate from a precursor, yet to be defined, at the level of the mesenchyme and migrate through the aorta wall and subsequently into the circulation. The origin of the hematopoietic clusters is a critical issue in developmental hematopoiesis since it brings up a fundamental question: are HSC obviously generated through an endothelial intermediate during early development or could they be generated de novo from additional cell sources?
Gene expressions in the hemogenic aorta of the chick embryo. With the aim of characterizing in detail the emergence of the hematopoietic clusters at the level of the aorta, we have first examined the expression of several markers from the stage of the paired aortae to the completion of aorta-associated hematopoiesis. When the aortae are still paired and no sign of hematopoiesis is visible, aortic EC express a set of genes; some of them are EC-specific (VEGF-R2 and VE-cadherin albeit at low levels) whereas the others are shared by EC and HC (SCL-Tal-1, Lmo2) and are thus referred to as hemangioblastic-specific genes. At the time aortae fuse, the transcription factor Runx1 becomes expressed. The Runx1 transcription factor was shown to be essential for definitive hematopoiesis to occur (38, 39). Runx1 associates with a co-factor termed CBFß to form the core binding factor. Inactivation of CBFß in the mouse embryo precludes the emergence of definitive hematopoiesis (40–42). Interestingly, expression of the avian orthologue for CBFß is also regionalized to the ventral aortic EC (Fig. 1C) (43), indicating that the core binding factor plays a role in aorta-associated hematopoiesis. The steps after fusion are characterized by thickening of the ventral EC, increased expression of hemangioblastic- (SCL-Tal-1, Lmo2) and HC-specific genes (Runx1, CBFß, GATA-3, CD45) and decreased expression of EC-specific genes (downgrading of VEGF-R2 and VE-cadherin). From that time, hematopoietic clusters become conspicuous (Fig. 1A) and expression of EC and HC-specific genes is dissociated. Runx1 remains expressed by the intra-aortic clusters and the adjacent hemogenic endothelium (Fig. 1B) (43). Using a knock-in approach in the mouse embryo, Runx1 was shown to be specifically expressed by the ventrolateral aortic EC, by the intra-aortic clusters (44) and was also reported to label the HSC fraction (45). A comparison of molecules expressed, respectively, by EC and HC clusters in the avian, mouse and human embryo is presented in Table 1.
|CD34||nd||nd||+||+||+||+||(60, 61, 76, 83, 84)|
|CD31 (PECAM)||nd||nd||+||+||+||+||(84, 85)|
|CD41 (αIIB)||+||−||+||−||nd||nd||(86, 87)|
|CD44 (HCAM)||nd||nd||nd||nd||+||−||(48, 88)|
|CD45||+||−||+||−||+||−||(46, 60, 63, 83)|
(αVß3, vitronectin receptor)
|CD117 (c-kit)||+#||−#||+||−||+||−||(60, 90, 91)#|
(L. Lecoin unpublished)
(VE-Cadherin, Cadherin 5)
|−||+||+/−§||+||+||+||(45, 60, 92–94)|
(MCAM, MUC18, S-endo)
|+||+||nd||nd||+||+||(Jaffredo, unpublished) (95)|
|Podocalyxin||+||+||nd||+||nd||nd||(96, 97) |
(Jaffredo and Bollerot, unpublished)
|Flk-1||+||−||+||+||+/−°||+||(46, 60, 90, 98, 99)|
|TIE-2 (CD202b, Tek)||+||+||+||+||nd||nd||(100, 101)|
|GATA-2||+||+||+||−||+||+||(60, 62, 90, 93)|
|GATA-3||+||+||+/−||−||nd||nd||(62, 63, 93, 102)|
|SCL-tal-1||+||+||+||+||+||+||(60, 90, 93, 103)|
|c-myb||+||−||nd||nd||+||−||(64, 90, 93)|
Dynamic tracing of the hemogenic endothelium in birds. In order to follow the fate of aortic EC, we have tagged the aortic endothelium with two vital tracers. One of these tracers is the acetylated low density lipoprotein (AcLDL) coupled to a fluorescent lipophilic marker (DiI) and the second is a non-replicative retroviral vector carrying a reporter gene. The first provides a short-term tracing whereas the second allows tagged cells to be followed for a longer time. Both were inoculated into the heart, a technique that distributes the tracers in direct contact with the endothelium lining the vessels. Inoculations were performed at the beginning of E2 i.e., one day before emergence of the aorta-associated HC. Cell identification was achieved using the anti-VEGF-R2 (35) and anti-CD45 antibodies (27) as, respectively, EC- and HC-specific probes.
As soon as 2 h after AcLDL-DiI inoculation, the whole vascular tree was labeled. The aortae, still paired at that time, were entirely lined by AcLDL+ EC while no CD45+ cell was present either at the level of the aorta or in the circulation. One day after inoculation, when the emerging hematopoietic clusters were examined, they were found to contain AcLDL-DiI+/CD45+ cells. Moreover, AcLDL-DiI+/CD45+ double-positive cells were found in the mesenchyme ventral to the aorta, suggesting that a subset of HC had ingressed into this region from the ventral endothelium (46, 47). Interestingly, the hematopoietic clusters were found to be attached by tight junctions to adjacent EC in the ventral side of the aorta (46), a situation similar to that seen in the mammalian embryo (44, 48). Thus, intra-aortic clusters derive from cells located in the aortic floor that previously had an endothelial phenotype.
The hemogenic endothelium in mammalian embryos. In mammalian embryos a body of evidence also suggests the existence of hemogenic endothelium. Similar to the situation in the avian embryo, electron microscopic studies show that hematopoietic clusters form tight junctions with aortic EC (44, 48, 49). Hematopoietic clusters do not appear to have junctions with the cell of the subaortic region that are themselves interconnected with each other (48). Immunohistological or in situ hybridization analyses show that EC and cluster-derived HC share many markers suggestive of a close developmental relationship (reviewed in (48, 50, 51)). The origins of HSC in the embryo have also been addressed by using an elegant transgenic system wherein the GFP reporter gene is driven by the Sca-1 (Ly-6A) regulatory elements. The Sca-1 protein is expressed on 99% of bone marrow HSC and progenitor cells and has been used to purify HSC from bone marrow, fetal liver or yolk sac (52–54). In these transgenic embryos, GFP was detected in individual cells inserted into the endothelial layer and also in hematopoietic clusters attached to the ventral aspect of the aorta (55, 56). When isolated at E11, this GFP+ cell fraction is able to reconstitute the hematopoietic system of irradiated hosts in an adoptive transfer system. Vascular AcLDL-DiI labeling of the vascular tree was also performed in the mouse embryo. AcLDL-DiI+/CD31+/CD34+/CD45− cells were sorted and put in semisolid culture. After 2 weeks, these cells give rise to definitive erythroid colonies as testified by their shape and hemoglobin chain make-up (57). Taken together these results are indicative of an endothelial origin of the aortic hematopoietic clusters and strongly support the notion of hemogenic endothelium.
Presence of groups of cells expressing hematopoietic markers in subaortic patches is another aspect of aorta-associated hematopoiesis in mouse and man (58–63). These sub-aortic patches were reported to harbor CD45−/lo/ c-kit/AA4.1/flk-1− cells and to be endowed with multilineage hematopoietic reconstitution in Rag1γc−/− immunodeficient mice (62). These data thus suggest the existence of HSC in the aortic mesenchyme in mammals. Whether these cells are developmentally related to the endothelium or emerge de novo from the aortic mesenchyme remains to be determined.
A close link between somite endothelial cells and production of hematopoietic cells from the aorta
In all the species examined so far, hematopoietic clusters are restricted to the ventral aspect of the vessel. Transplantation experiments in avian embryos have produced evidence that EC of the embryo belong to two different lineages depending on the mesodermal structures concerned i.e., the splanchnopleural mesoderm and the somite (64). These two lineages were found to differ dramatically in their intrinsic potential. EC originating from the splanchnic mesoderm arise in situ (65) and exhibit a dual hemangiopoietic potential i.e., it can give rise to both EC and HC. These EC give rise to the endothelial network of the visceral organs and were shown to form the endothelial floor of the aorta (64). EC originating from the somites are purely endothelial, colonize the somatic (somatopleural) mesoderm, and give rise to vasculature of the body wall (64, 66) as well as lymphatic vessels (67). In vivo lineage tracings with retroviral vectors have shown that somite EC and striated myogenic cells derived from a common precursor (68). Using Quek-1 (VEGF-R2) as a probe, Eichmann et al. (33) reported that the dorsolateral aspect of the somite shielded a population of EC. As somites mature, EC leave the structure to colonize the intermediate mesoderm and the lateral plate. At E5, somite EC contribution has increased considerably and involves the whole vascular network of the neural tube, nerve ganglia, dermis, kidney, body wall, including the large cardinal veins and limb bud. However, they were barred from entering ventrally into the visceral organ, revealing a tight, although as yet not understood, control of EC migration and homing (69). Interestingly, a transient contact with endoderm or treatment with VEGF, bFGF and TGFß growth factors can turn somite cells from being angiopoietic to hemangiopoietic (70). The mapping of territories colonized by EC originating either from somitic or splanchnopleural mesoderm has also suggested an unexpected regionalization of the aorta vascular endothelium. Grafts of either quail somites or segmental plate (the mesodermal stripe giving rise to the somites) into the chick embryo reveal a dual and complementary origin of the aorta: the roof and sides being contributed by somite-derived EC whereas the floor is contributed by splanchnopleura-derived EC (64, 69). Given this contribution, the restriction of the intra-aortic clusters to the aortic floor appears obvious. EC from the somite are purely endothelial and are thus unable to produce hematopoiesis. In contrast, splanchnopleura-derived EC that form the aortic floor are endowed with hemogenic capacities (64, 71); intra-aortic clusters are thus restricted to the ventral aspect of the aorta. As hematopoiesis proceeds, another interesting contribution of the somite is revealed; aortic hemangioblasts progressively disappear from the aortic floor to be replaced by EC from the somite. At the end of the intra-aortic cluster period, the aortic endothelium is found to be entirely of somite origin. Thus between the second and the fifth day of incubation, the aorta has been submitted to two successive remodelings: one, before emergence of the intra-aortic clusters contributes to replacing the original roof and sides with new roof and sides; another, occurring during intra-aortic cluster production, contributes to replacing the aortic floor. This second remodeling has major implications for hematopoietic production and provides an explanation for the time-restricted production of intra-aortic hematopoiesis reported in all the species described so far. Disappearance of the aortic hemangioblast indicates that they leave the aortic floor for elsewhere (69). Previous experiments have shown that aortic hemangioblasts undergo a switch from EC to HC (57, 72–76) and leave the aorta by either the blood flow or by entering into the mesenchyme where they form the para-aortic foci (46, 47). They disappear completely, meaning that the whole hemangioblast population undergoes phenotypic change. Since the hemogenic endothelium also ensures typical endothelial functions, the need for vessel integrity is obvious and is provided by somite EC. In addition, changing the floor from hemangioblasts to somite EC dramatically modifies its capacity to give rise to hematopoiesis since somite EC are restricted to angiopoiesis and are thus incapable of producing hematopoiesis; hence hematopoietic production ceased (69).
The allantois: another source of hemangioblasts?
In birds, the timing of intra-aortic clusters and para-aortic foci is compatible with the seeding of the thymus and bursa of Fabricius. Cells of the para-aortic foci have been shown to display B- and T-lymphoid potential as well as to harbor multipotent progenitors, probably HSC. However, seeding of the bone marrow begins by E10.5 in the chick embryo and lasts until birth. At this time, the activity of the para-aortic foci has ceased. The possibility that cells from the para-aortic foci seed the bone marrow during its very late phases of colonization appears unlikely. We have thus looked for another progenitor-producing site and identified the allantois. This appendage was shown to be involved in gas exchange, excretion, shell Ca++ resorption and bone formation. It also has the appropriate tissue make-up to produce hemopoiesis i.e., endoderm associated with mesoderm. By using India ink or AcLDL-DiI microangiographies, we first established that the allantois becomes vascularized at 75–80 h of incubation. It displays conspicuous red cells even before vascularization, indicating that hematopoiesis occurs in this site independent of the rest of the embryo (77, 78). The allantois undergoes a program characterized by the prominent expression of several “hemangioblastic” genes in the mesoderm and of other genes in the associated endoderm. VEGF-R2 is expressed in the mesoderm, at least from stage HH17 onward, and is shortly afterwards followed by the expression of several transcription factors (GATA-2, SCL/tal-1, and GATA-1). Blood island-like structures that contain both CD45+ cells and cells accumulating hemoglobin differentiate. These structures look exactly like blood islands in the yolk sac. This hematopoietic process takes place prior to the establishment of a vascular network connecting the allantois to the embryo. As far as the endoderm is concerned, GATA-3 mRNA is found in the region where allantois will differentiate before the posterior intestinal portal becomes anatomically distinct. Shortly before the bud grows out, GATA-2 is expressed in the endoderm and, at the same time, the hemangioblastic program is initiated in the mesoderm. GATA-3 is detected at least until E8 and GATA-2 until E3, the latest stages examined for these factors. Using in vitro cultures we showed that allantoic buds, dissected out before the establishment of circulation between the bud and the rest of the embryo, produced elliptic erythrocytes, typical of the definitive lineage. Moreover, using heterospecific grafts between chick and quail embryos, we demonstrated that the allantoic vascular network develops from intrinsic progenitors (78). Taken together, these results extend our earlier findings regarding the commitment of mesoderm to endothelial and hematopoietic lineages in allantois. The detection of a prominent GATA-3 expression restricted to the endoderm of the pre-allantoic region and allantoic bud, followed by that of GATA-2, is a novel and interesting piece of information in the context of organ formation and endoderm specification with respect to the emergence of hemopoietic cells.
It was suggested 25 years ago that the mouse chorioallantoic placenta could also be a source of HC (79). This appendage develops by fusion of the allantois to the ectoplacental cone. The work by Melchers reported on a peak in the number of B-lymphoid progenitors in the E12 placenta; the number of progenitors decreased thereafter. Recent work indicates that the mouse allantois also harbors a large number of erythromyeloid clonogenic progenitors. Cells with high proliferation potential, as well as the whole range of myeloid progenitors, were found to develop from allantois-isolated cells. This production begins at E8.5 (18 pairs of somites) and persists up to E17. Because embryos were obtained by mating GFP+ males with GFP- females, the respective contributions of the maternal vs embryonic compartment can easily be analyzed. The presence of early hematopoietic progenitors (able to give rise to large multilineage colonies) is 2–4 times more frequent in the allantois than the yolk sac or the fetal liver at the same embryonic stages (80). In addition, the placenta was shown to contain a large pool of HSC. The presence of long-term repopulating HSC is found to start at E10.5–E11.0 and to expand until E12.5–E13.5 to finally contain more than 15 times more HSC than the AGM at the same stage. This expansion that occurs prior to and during the expansion of HSC in the fetal liver suggests that the placenta harbors a strongly HSC-supportive microenvironment (55, 81). Cells expressing CD34 and CD31 as well as cells expressing c-kit, GATA-2, GATA-3 and Runx-1 are also found in the embryonic vasculature of the placenta (55). Taken together these data strongly suggest a previously unappreciated role of the placenta as a source of HSC. Further studies will be needed to determine the origins of placental HSC and the characteristics of the placental microenvironment as a support for HSC activity.
A new gene transfer system to tag endothelial cells and hemangioblasts during development: towards a genetic analysis of the development of the blood-forming system in the avian embryo?
The chick embryo has for decades been one of the major models used by developmental biologists. The recent availability of new technologies i.e., electroporation, ES cells combined with the availability of chick genome and EST databanks, has put the chick embryo back on stage as one of the most versatile experimental model available. Efficient gene transfer in the chick is mainly achieved via the use of viral vectors or electroporation. Viral-mediated gene transfer suffers several drawbacks, including the limited size of cDNA a virus can accommodate and the time-consuming preparation of viral stocks. Electroporation has been successfully applied to many epithelia but mesenchyme appears more difficult to target because the current tends to pass between cells rather than inside them. When applied to embryonic blood vessels, electroporation appears leaky probably because plasmids that move into the electric field can cross the thin endothelial barrier resulting in unwanted transfer to abluminal cells. To circumvent these problems we have developed a non-viral, lipofection-based test in which nucleic acids could be specifically transferred to EC in a simple way. Such a test should allow study of gene function in a short time window and should ideally be applicable during the whole embryonic life.
We have designed a technique for specific and efficient delivery of a nucleic acid molecule into vascular EC and hemangioblasts of the upper vertebrate embryo (82). Nucleic acids were compacted with cationic liposomes specifically designed for gene therapy applications. We delivered these lipoplexes to the embryonic circulation by intracardiac injection and observed very efficient transfection to EC. Indeed, most EC in the embryo were labeled (Fig. 2A & 2B). The hemogenic endothelium and intra-aortic clusters, known to generate adult-type HSC, were also strongly labeled (Fig. 2B & 2C). This delivery system was applied to avian and cultured mouse embryos with similar effect. This method offers new possibilities to rapidly analyze the function of any gene involved in the formation and maintenance of the vascular system. As the hemangioblast and its derivatives i.e., blood cells are frequently affected during pathologies, such an approach should also help in discovering new signaling pathways involved in normal or pathological situations and should provide opportunities for the identification of new therapeutic molecules. The simple design of the experiment combined with its low cost make it possible to screen for the function of a number of genes and gene variants in a short period of time. Moreover, since injections can be performed at any time during the first 6 days of chick development and possibly later, the present technology offers tight temporal control of the process of gene transfer and gene function analysis.