The embryonic origins of human haematopoiesis


Caroline Marshall, Molecular Immunology Unit, Institute of Child Health, 30 Guilford Street, London WC1N 1EH, UK. E-mail:

Mammalian haematopoiesis can be subdivided into two main systems, the embryonic primitive system and the definitive system. During primitive haematopoiesis, nucleated erythroid cells arise in the extraembryonic yolk sac and circulate through the embryo supplying oxygen and nutrients to the developing tissues. As the embryo develops, this early circulation is superseded by the definitive multilineage blood system that seeds the liver, bone marrow and other haematopoietic tissues. Definitive haematopoiesis is sustained by pluripotent haematopoietic stem cells (HSCs) which have the capacity to self-renew and are capable of long-term repopulation in myeloablated recipients. However, HSCs exist only in small numbers in human bone marrow and peripheral blood. Consequently, their usefulness for clinical application is dependent upon recreating the appropriate microenvironment which will allow expansion of HSC in vitro while retaining their pluripotent phenotype and repopulating capacity. An understanding of the embryonic origin of long-term repopulating (LTR)-HSCs and the factors that regulate their development in vivo will provide invaluable information to facilitate their expansion in culture. As a consequence, the study of human embryonic haematopoiesis has stimulated a considerable amount of interest and research in recent years.

In this article, we will describe what is currently known about the origin of definitive HSCs and the questions these findings have raised concerning their development. We will examine evidence from animal models which has provided many clues as to the regulatory factors which may be involved and relate these to recent findings in the human embryo. Finally, we will discuss possible mechanisms for HSC ontogenesis suggested by these data.

Evidence for an intraembryonic source of definitive HSCs

Previously, it was thought that the embryonic liver became colonized by stem cells derived from the extraembryonic yolk sac blood islands and that these cells seeded the definitive haematopoietic system. However, experiments in which quail embryos were grafted onto chick yolk sac revealed that the major source of definitive HSCs was contained within the embryo itself (Dieterlen-Lievre, 1975). The subsequent detection of a territory with haematopoietic potential in the wall of the chick dorsal aorta suggested that definitive HSCs might therefore arise in this region (Cormier et al, 1986). In the mouse, an analogous intraembryonic region, termed the aorta–gonad–mesonephros (AGM), is derived from the para-aortic splanchnopleural (P-Sp) mesoderm and contains the dorsal aorta, putative gonadal ridge and mesonephros. At 7·5 d post-coitum (DPC), before circulation is established between the embryo and yolk sac, multipotent stem cells capable of lymphoid and myeloid differentiation are found in the P-Sp region surrounding the dorsal aorta but not in the yolk sac. Furthermore, at 10 DPC this region contains spleen colony-forming unit (CFU-S) activity at a frequency greater than yolk sac or liver (Medvinsky et al, 1993, 1996; Dzierzak & Medvinsky, 1995). Together, these findings suggest that HSC within the P-Sp-derived AGM at 10 DPC are not derived from circulating yolk sac progenitors, but are generated in situ (Cumano et al, 1996). Isolated AGM cultured in vitro also reveals this region to be a rich source of HSCs which arise autonomously and independently from the yolk sac and liver and which are capable of reconstituting irradiated recipient mice before the onset of liver haematopoiesis (Muller et al, 1994; Medvinsky & Dzierzak, 1996; Sanchez et al, 1996). It is thought that it is these AGM-derived HSCs which migrate to the embryonic liver and seed the definitive blood system. The contribution of yolk sac cells to the mammalian definitive system is still under discussion but, from the studies described above, appears to be minimal. Long-term multilineage repopulation by cells isolated from murine yolk sac at 9 DPC has been reported in newborn mice (Yoder et al, 1997); however, by this time circulation is established and it is possible that these LTR-HSCs are in fact AGM derived. In the human embryo, pluripotent haematopoietic progenitors with high-proliferative capacity also arise almost exclusively from non-hepatic intraembryonic tissue between 25 and 35 d gestation (Huyhn et al, 1995).

Cells with haematopoietic potential within the AGM were first identified as clusters within the mesoderm underlying and adhering to the ventral floor of the chick dorsal aorta (Cormier et al, 1986). Subsequently, examination of tissue sections through human (Tavian et al, 1996; Marshall et al, 1999) and murine (Wood et al, 1997) embryos has also revealed clusters of round cells adhering to the ventral floor of the dorsal aorta within the AGM, although no clusters have thus far been detected in the underlying mesoderm. Similar adherent clusters of cells are also found in the proximal umbilical and vitteline arteries, which interconnect with the dorsal aorta (Wood et al, 1997; Tavian et al, 1999). The clear similarities of distribution between species suggest that intraembryonic haematopoiesis is a common developmental process. In common with endothelial cells lining the dorsal aorta, cells within the clusters express the membrane glycoprotein CD34 (Fig 1) (Tavian et al, 1996; Wood et al, 1997; Garcia-Porrero et al, 1998; Marshall et al, 1999). They also express the haematopoietic-specific marker CD45, which distinguishes them from adjacent CD34+/CD45 endothelial cells. Their appearance is highly restricted, both spatially and temporally, during embryogenesis. In the human, an accumulation of around 800 cells develops in the preumbilical dorsal aorta, often around the level of the upper limb bud between 28 and 38 d gestation (comparable with 9·5–11·5 DPC in the mouse). It is now widely accepted that the first definitive multipotent HSCs that seed the adult blood system are generated within the embryonic AGM region and, most likely, these stem cells are contained within these intravascular clusters from where they migrate to and colonize the fetal liver.

Figure 1.

(A) A 34-d human embryo showing heart (h), liver (l) and anterior limb bud (al). The aorta–gonad–mesonephros (AGM) region extends from the umbilicus to the anterior limb bud as indicated (agm) and contains the dorsal aorta (da). (B) Transverse section through the AGM region outlined in (A) showing the dorsal aorta (da) and surrounding mesenchyme, stained with a monoclonal antibody to the membrane glycoprotein CD34 (brown cells). A cluster of CD34+ haematopoietic cells (h) are associated with the ventral floor of the aorta within this region, in contact with endothelial cells (e) lining the wall of the aorta which also express CD34. (For immunohistochemistry method, see Marshall et al, 1999). Original magnification 39×.

Markers of haematopoiesis in agm clusters

Analysis of the human and murine embryonic AGM has revealed that cells within the intra-aortic clusters express many markers known to be associated with haematopoiesis, including the stem cell factor receptor c-Kit and the transcriptional regulators GATA-2, acute myeloid leukaemia (AML)-1 and stem cell leukaemia (SCL) (Table I). Gene disruption studies have shown that GATA-2 and AML-1/Cbfa2 are essential for definitive haematopoiesis (Table II). GATA-2-null embryos survive up to E11·5, the onset of liver haematopoiesis, complicit with a critical role in definitive but not primitive haematopoiesis (Tsai et al, 1994). Disruption of AML-1/Cbfa2 expression also results in a block in all definitive lineages with failure of fetal liver haematopoiesis and embryonic death by E12·5 (Castilla et al, 1996; Okuda et al, 1996; Wang et al, 1996). In contrast, the T-cell leukaemia oncoprotein tal-1/SCL is essential for both primitive (yolk sac) and definitive (intraembryonic) blood formation. Disruption of the SCL gene in transgenic mice results in a complete lack of blood formation in SCL–/– embryos and is lethal between E8·5 and E10·5 at the commencement of circulation (Shivdasani et al, 1995). Moreover, the failure of SCL–/– murine embryonic stem (ES) cells to give rise to precursor or mature haematopoietic cells of any lineage suggests that SCL may be involved in blood specification at a prehaematopoietic cell stage (Porcher et al, 1996).

Table I.  Comparison of molecules expressed by endothelial cells (EC) lining the dorsal aorta and by cells within intra-aortic haematopoietic clusters (HCs) in the murine and human embryonic AGM.
  1. Shared expression patterns suggest a common haemangioblast precursor. nd, not done.

CD34++++Garcia-Porrero et al (1998); Wood et al (1997);
Tavian et al (1996); Marshall et al (1999)
CD31(PECAM)++/–++Garcia-Porrero et al (1998); Tavian et al (1996)
CD45ndnd+Tavian et al (1996); Marshall et al (1999)
CD44(HCAM)ndnd+Watt et al (2000); C. J. Marshall (unpublished)
c-kit++Bernex et al (1996); Labastie et al (1998); Marshall et al (1999)
Flk-1/KDRndnd++Labastie et al (1998); Marshall et al (1999)
TIE-1++  C. J. Marshall (unpublished)
VE-cadherinndnd++C. J. Marshall (unpublished)
VCAM-1ndnd++C. J. Marshall (unpublished)
SCLndnd++Labastie et al (1998); Marshall et al (1999)
GATA-2ndnd++Labastie et al (1998); Marshall et al (1999)
c-mybndnd+Labastie et al (1998)
  North et al (1999)
WASPndnd+C. J. Marshall (unpublished)
Ulex europaeus
ndnd+Tavian et al (1996)
Table II.   Phenotypic effects of gene knock-outs for molecules expressed by aorta-associated haematopoietic cells and endothelial cells within the mammalian embryonic AGM, arranged in order of reducing severity.
Knock-outEndothelial cells/vesselsPrimitive bloodDefinitive bloodLethal
TIE-2 E9·5–10·5

Using single-cell reverse transcriptase polymerase chain reaction (PCR), the expression of early haematopoietic and lineage-specific genes has been determined in individually sorted CD34+/c-kit+ AGM cells. The majority of this population isolated from day 11 AGM coexpress the stem-cell and early haematopoiesis-associated transcription factors GATA-2, AML-1, PU.1 and Lmo-2 (Delassus et al, 1999). In the 4- to 6-week-old human embryo, cells within the CD34+ intra-aortic clusters express SCL/tal-1, GATA-2, GATA-3, c-myb and c-Kit (Bernex et al, 1996; Labastie et al, 1998; Marshall et al, 1999). They also express a number of molecules involved in homing and adhesion including CD44/HCAM, Wiskott Aldrich Syndrome Protein (WASP), CD106/VCAM-1, VE-cadherin (unpublished observations; Watt et al, 2000) and CD31/PECAM (Garcia-Porrero et al, 1998).

Evidence for a common endothelial haematopoietic precursor

An interesting feature emerging from this extensive analysis is the number of molecules expressed both by haematopoietic clusters within the clusters and by endothelial cells lining the aorta wall with which they are closely associated (Table I). For some time, it has been proposed that both haematopoietic and endothelial cells may derive directly from a common precursor, the haemangioblast. Individual blast-like bipotential cells, derived from ES cells, are capable of giving rise to both lineages in vitro (Choi et al, 1998); however, the identification of the putative haemangioblast in vivo has proved difficult.

One candidate marker for the haemangioblast is the receptor for vascular endothelial growth factor (VEGFR2), also known as Flk-1 (mouse) and KDR (human). Flk-1/KDR is expressed by endothelial cells and is essential for vessel formation. In homozygous flk-1-deficient mouse embryos, blood islands fail to form in the yolk sac, organized vessels do not develop and haematopoietic progenitors occur only very rarely (Shalaby et al, 1995, 1997) (Table II). However, reduced haematopoiesis in the absence of Flk-1 could be a secondary effect resulting from the lack of a suitable vascular environment rather than due to the ablation of a precursor cell type. In culture, Flk-1+ murine ES cells possess both haematopoietic and endothelial potential (Nishikawa et al, 1998), and individual VEGFR2 cells sorted from early chick mesoderm, before vascularization, are able to generate both haematopoietic and endothelial cells (Eichmann et al, 1997). In this latter system, endothelial development requires the presence of VEGF, whereas HC specification is VEGF independent although it does require alternative VEGFR2 activation. In agreement with this, within the chick embryonic AGM, cells in the floor of the aorta appear to undergo a phenotypic transition from VEGFR2+ to CD45+ as they become haematopoietic (Jaffredo et al, 1998).

KDR, the human homologue of Flk-1, is expressed in 0·1–0·5% of CD34+ blood cells post-natally. Reconstitution experiments in NOD/SCID mice have demonstrated that all pluripotent LTR-HSCs are contained within the CD34+/KDR+ blood fraction (Ziegler et al, 1999). The remaining CD34+/KDR population includes more mature, lineage-committed progenitor cells, suggesting that KDR expression is an early event in haematopoietic specification. Circulating CD34+/KDR+ human peripheral blood cells can also adopt an adherent endothelial phenotype in culture (Asahara et al, 1997), and the equivalent embryonic population may represent cells with haemangiogenic potential. In the 5-week human embryonic AGM, KDR is expressed both on endothelial cells lining the wall of the dorsal aorta and on haematopoietic cells within the associated intra-aortic clusters (Labastie et al, 1998; Marshall et al, 1999). This finding supports the suggestion that haematopoietic clusters within the AGM are indeed pluripotent stem cells that have recently diverged from an endothelial-like precursor.

Sites of precursor development – a closer look at the aortic floor

The mechanism by which such haematopoietic precursors might arise has been a subject of considerable interest and discussion. It has been suggested that haemangioblast precursors migrate into the wall of the dorsal aorta from the underlying mesenchyme, only committing to one or other lineage on arrival. Alternatively, bipotential precursor cells already located within the floor of the aorta could preferentially commit to a haematopoietic fate in response to a localized and transitory signal. A third possibility is that endothelial cells, under certain specific conditions, are capable of de-differentiation and can switch to a haematopoietic phenotype. The dorsal aorta is formed by the process of vasculogenesis from the organization of pre-existing cells into an endothelial cell-lined vessel, suggesting that the endothelium is established before intra-aortic definitive HCs appear. It therefore seems likely that this bipotential cell is located within the endothelial wall at the time of HC emergence. However, it cannot be ruled out that cells with haematopoietic potential move from the subaortic mesoderm and insert into the endothelium at some point between vessel formation and the appearance of HCs.

It has also been suggested that apparent interruptions observed in the ventral endothelial layer underlying intra-aortic HCs represent gaps through which haematopoietic cells could migrate from the subaortic mesoderm or, alternatively, mark areas of endothelial cell transdifferentiation (Tavian et al, 1999). However, ultrastructural examination of tissue sections through the 10·5 DPC murine AGM indicates that the basal lamina of the endothelium, where it is associated with the HC, is intact (Fig 2A). Above the basal lamina, adjacent endothelial cells are connected to each other and to haematopoietic cells within the adjoining cluster by tight junctions (Fig 2A–C). Likewise, mesodermal cells in the subaortic region, beneath the basal lamina, are also interconnected by tight junctions. However, there appear to be no structural connections between these two layers. The sides and dorsal roof of the aorta are lined by a monolayer of flattened endothelial cells (Fig 2D). In striking contrast, within the ventral floor of the aorta adjacent to the clusters and continuous with the endothelium, large rounded cells are frequently observed extending into the lumen. Many of these cells contain large cystic formations close to where they attach to the basal lamina or to adjacent cells and appear to be budding out from the aortic wall (Fig 2A–C). Indeed, the HC itself has the appearance of these budding cells piling up. Discontinuities in the floor of the aorta may therefore be due to this budding activity rather than to the disappearance of the endothelial wall.

Figure 2.

Electron micrographs of the dorsal aorta in the murine AGM region at 10·5 DPC showing the endothelial layer and underlying mesenchyme. (A) Ventral floor of aorta adjacent to the haematopoietic cluster. Rounded cells within the endothelial layer (E) are connected via tight junctions (short arrows) to each other but not to cells in the underlying mesenchyme (M). Between the endothelium and mesenchyme, a basal lamina of extracellular matrix is visible (long arrows). Insert detail of boxed area at higher magnification showing tight junction. (B) Rounded haematopoietic cells (h) within the cluster extending into the lumen of the dorsal aorta (L) remain interconnected by tight junctions. (C) Cells are clearly visible ‘budding’ out of the ventral endothelial layer above an intact basal lamina. Many budding cells contain large cystic formations (asterisk) and appear to be partially detaching from the aortic floor. (D) Dorsally, the endothelial layer is composed of a monolayer of flattened cells separated from a loosely arranged mesenchyme by basal lamina. Large haem-packed, nucleated primitive erythroid cells (P) are found circulating in the lumen of the aorta. Notably, mesenchymal cells underlying the ventral floor of the aorta are more densely packed (A) than those around the lateral and dorsal aspects (D). Original magnifications 4000×.

There is some evidence that the transcription factor AML-1 may play a role in the budding of haematopoietic cells out of the endothelium, and it has been suggested that it may also act as a molecular switch specifying the conversion from an endothelial to a haematopoietic fate (North et al, 1999). Targeted insertion of β-galactosidase into one allele of the murine Cbfa2 gene reveals that AML-1 is expressed by a subset of endothelial cells restricted to sites of haematopoietic emergence in the yolk sac, vitelline and umbilical arteries and in the ventral floor of the dorsal aorta (North et al, 1999). In the aorta, AML-1 expression is transient and commences at 8·5 DPC before the emergence of haematopoietic cells. Moreover, in embryos lacking Cbfa2 intra-aortic clusters fail to form and, although extraembryonic primitive erythroid cells develop normally, budding of haematopoietic cells from the endothelium of the yolk sac capillaries is impaired. AML-1 expression may therefore identify an intermediate stage between the haemangioblast and final commitment to either the endothelial or haematopoietic lineage.

Further support for the generation of intra-aortic HCs from existing precursors or intermediates within the endothelium is provided by the expression patterns of the endothelial-specific receptor tyrosine kinases TIE-1 and TIE-2/TEK, relatives of the Flk-1/KDR receptor. Like Flk-1, TIE-2 is expressed on endothelial cells at the level of the putative bipotential haemangioblast. The majority of TIE-2+ cells within the AGM fall within the CD34+/c-Kit+ fraction and can give rise to clones containing both endothelial and multipotent HSCs in culture (Takakura et al, 1998; Hamaguchi et al, 1999). Its role in haematopoiesis is thought to be the aggregation and adherence of HSCs to endothelial layers, thus facilitating their proliferation. TIE-2-deficient embryos have reduced definitive haematopoiesis, probably due to a failure of stem cell aggregation and expansion within the dorsal aorta. Endothelial cells develop but death occurs around 9·5–10·5 DPC as a result of defects in angiogenesis, vascular remodelling and vessel integrity (Table II). TIE-1 is also expressed on vascular endothelial cells and is required for cell integrity and maintenance but not, seemingly, for early endothelial differentiation or vessel formation. Mice lacking TIE-1 expression develop a normal vascular network but die just before or after birth from haemorrhaging and oedema due to leakage of blood cells through the vascular endothelium (Puri et al, 1995; Sato et al, 1995) (Table II). Analysis of both human and mouse bone marrow indicates that TIE-1 is also expressed on early multipotent CD34+ haematopoietic stem cells including long-term culture-initiating cells (LTC-ICs) and LTR-HSCs (Hashiyama et al, 1996; Yano et al, 1997). This expression is generally lost as the cells mature. Because Flk-1 and TIE-2 are essential for normal vessel formation, the reduction in haematopoiesis seen in deficient embryos could be due to the removal of a permissive environment rather than ablation of a haemangiogenic precursor. TIE-1, on the other hand, is not required for vessel formation nor does it appear to play a role in haematopoiesis. The site of AGM haematopoietic development is therefore intact in TIE-1-deficient embryos, and TIE-1 expression in CD34+ haematopoietic cells could therefore truly reflect a common haemangioblast origin. A murine model in which TIE-1 expression is coupled to β-galactosidase activity has allowed the relationship between endothelial and haematopoietic cells within the AGM to be visualized. At 10·5 DPC, expression is continuous between cells in the floor of the aorta and the associated haematopoietic cells. Notably, no TIE-1+ cells are detected in the underlying mesoderm (Fig 3).

Figure 3.

Expression of the endothelial-specific tyrosine kinase receptor TIE-1 in the murine embryonic AGM region at 10·5 DPC. Coupling of the TIE-1 gene to β-galactosidase results in blue staining of positive cells. (A) TIE-1 expression is restricted to endothelial cells lining vessels including the dorsal aorta (da) and cardinal veins (cv). (B) Higher magnification of (A). Expression of TIE-1 is continuous between the endothelium (e) and cells within the adherent haematopoietic cluster associated with the ventral floor of the aorta (h) supporting a common origin. Original magnifications: (A) 15×; (B) 39×.

In the chick embryo, pockets of GATA-3-expressing cells lying within the ventral mesoderm underlying the aorta and associated with the appearance of HCs have been put forward as putative sites of haemangioblast generation which, presumably, cross the endothelium to enter the vessel lumen (Manaia et al, 2000). Recently, a detailed analysis of the RNA expression patterns of Lmo-2 and another GATA family member, GATA-3, in E7-12 murine embryonic tissue has led to the suggestion that these two factors may interact within the murine AGM and be involved in the specification of intraembryonic haematopoietic precursors (Manaia et al, 2000). GATA-3 mRNA is also found in cells scattered beneath the aortic floor in the human AGM at 5 weeks (Labastie et al, 1998). However, the precise relationship between GATA-3 expression and haematopoietic specification is unclear and, in the human, expression of known early haematopoietic markers, such as KDR and SCL, does not appear to co-localize with GATA-3 in this subaortic region.

At present, it seems most likely that intra-aortic haematopoietic clusters are composed of cells budding out from the floor of the aorta rather than cells migrating through the aortic wall from the underlying mesoderm. Although mechanisms remain undetermined, this process may involve the de-differentiation of existing endothelial cells or, alternatively, the induction of bipotential haemangioblasts located within the wall of the aorta.

Morphological polarity within the human agm

The clusters of haematopoietic cells observed in the dorsal aorta are always associated with the ventral wall, suggesting an asymmetry across the dorsoventral axis. A close examination of the surrounding mesenchymal tissue in fixed tissue sections does indeed reveal morphological polarity. A discreet region of densely packed cells, of between five and seven layers thick in the human and of three or four layers in the mouse, underlies the ventral aortic wall (Marshall et al, 1999). In contrast, cells surrounding the lateral and dorsal walls of the aorta are more loosely arranged. This observed polarity is conducive to the reported assembly of vascular smooth muscle cells (VSMCs) around the quail embryonic aorta (Hungerford et al, 1996). Recruitment of mesoderm-derived VSMCs around nascent endothelium is a critical event in building the vessel wall and proceeds radially from ventral to dorsal, resulting in a condensation of VSMCs beneath the ventral endothelium.

The significance of this region in AGM haematopoiesis is revealed by the localized expression patterns of growth factors and other molecules associated with the appearance of HCs. Smooth muscle α-actin (SMα-A) is expressed by haematopoiesis-supporting stromal cells in human bone marrow (Galmiche et al, 1993). In the human embryonic AGM at 27 d gestation, preceding the emergence of HCs, a few SMα-A+ cells are associated with the ventral wall of the dorsal aorta (Tavian et al, 1999). At 33 d, a multilayered condensation of SMα-A-expressing cells underlies the ventral wall associated with intra-aortic haematopoietic clusters compared with only a single layer at the dorsal wall. The suggestion that this dense ventral region could represent a stromal layer facilitating haematopoietic activity in the AGM is supported by the restricted expression patterns of adhesion and extracellular matrix (ECM) molecules. HCA, a homologue of the murine activated leucocyte cell adhesion molecule (ALCAM), is expressed on CD34+ HSCs and on stromal cells in primary haematopoietic sites and is believed to mediate HSC–stroma interactions. In the 4- to 6-week-old human AGM, a small number of HCA+ cells underlie the aortic wall and are mainly restricted to the ventral region (Cortes et al, 1999).

An even more strikingly polarized pattern of expression is exhibited by the ECM molecule tenascin-C. Around the human embryonic dorsal aorta at 34 d, high-level expression of tenascin-C is concentrated in the ventral stromal region immediately underneath the endothelial wall and specifically associated with the presence of intra-aortic haematopoietic clusters (Marshall et al, 1999). At more caudal levels, expression in this stroma appears lower than in the surrounding mesenchyme. During development, the tenascins mediate cell transformations, by influencing cell shape and motility, at the epithelial–mesenchymal interface in tooth, lung and kidney morphogenesis (Crossin et al, 1986; Prieto et al, 1990). Tenascin-C is expressed in murine bone marrow-derived stromal cells (Seiffert et al, 1998), and although tenascin-C-deficient mice appear to have normal haematopoiesis post-natally the capacity of bone marrow stromal cells derived from these mice to support haematopoiesis is significantly reduced (Ohta et al, 1998). The distinctive expression pattern of tenascin-C in the ventral wall of the dorsal aorta suggests that it may be involved in haematopoietic specification within the endothelium, possibly in the mediation of cell-growth factor interaction.

Factors influencing hsc development: evidence from other animal models

What an embryonic cell becomes is largely determined by environmental signals in the form of growth factors or morphogens, which, if received via surface receptors, can trigger cells to divide, differentiate or even die. These extracellular signals are mediated by a cascade of signalling molecules and, finally, by transcription factors which control gene expression in the nucleus. Transgenic models have shown that a number of factors are absolutely required for normal haematopoiesis, but their precise function is unclear. As human embryonic tissue is comparatively scarce and cannot be manipulated in vivo, much of the information we currently have regarding haematopoietic development and regulation comes from studies in other animals. Although obvious differences exist between mammals and amphibians, many systems and proteins at the molecular level are remarkably conserved across species. The South African clawed toad, Xenopus laevis, and the zebrafish, Danio rerio, produce large numbers of embryos which develop outside the uterus and have proved invaluable models in which to study the factors that influence blood cell formation.

In Xenopus embryos, primitive blood cells are produced in ventral blood islands, equivalent to the mammalian yolk sac, and multilineage definitive blood cells are generated in dorsal lateral plate, analogous to the AGM region. The transcription factor SCL is expressed in these tissues early in embryogenesis. Ectopic SCL induces globin expression in Xenopus animal cap (AC) cells, which do not normally contribute to blood, suggesting that SCL is sufficient to specify a haematopoietic fate (Zhang & Evans, 1996; Mead et al, 1998). In zebrafish, ectopic expression of SCL mRNA results in a dramatic increase in the number of Flk-1+ haematopoietic and endothelial precursors (Gering et al, 1998). Murine ES cells lacking SCL expression exhibit endothelial markers, including Flk-1, GATA-2 and CD34, but fail to express haematopoietic-restricted genes (Elefanty et al, 1997; Faloon et al, 2000). These data suggest that SCL is involved in the commitment of precursors to a haematopoietic fate and possibly also in the specification of haemangioblasts from mesodermal cells.

In Xenopus, SCL expression is induced by bone morphogenetic protein (BMP)-4 (Mead et al, 1998), a member of the transforming growth factor (TGF)-β superfamily of secreted polypeptide growth factors which collectively have a marked influence on pattern formation and cell fate determination throughout embryogenesis (Hogan, 1996). In the mouse, between 8·5 and 9·5 DPC, BMP-4 is expressed in the splanchnopleuric (endoderm derived) and somatopleuric (ectoderm derived) mesoderms and is essential at a number of different stages and in a variety of processes during development. In early stages, it is also required for mesoderm induction, and mice lacking BMP-4 expression are normally arrested at the egg cylinder stage with little or no mesoderm (Winnier et al, 1995). Mutant embryos developing beyond this stage have retarded, truncated and disorganized posterior structures, revealing an essential role for BMP-4 in dorsoventral patterning, and deficient haematopoietic tissue. However, because of the major defects resulting from gene ablation, the involvement of BMP-4 in the induction of intraembryonic haematopoiesis has been difficult to study in these mice.

BMP-4 influences pattern formation along concentration gradients which are detected by responding cells. The establishment of these gradients appears to be due to long-range inhibition or inactivation of BMP-4 signalling by antagonists, such as noggin and chordin, rather than by extensive BMP-4 diffusion (Jones & Smith, 1998). The downstream effects of BMP-4 signalling are graded, depending on concentration, and can either inhibit or activate specific gene expression. In agreement with this mode of action, BMP-4 has been shown to induce globin expression in cultured murine ES cells in a concentration-dependent manner (Johansson & Wiles, 1995). Interestingly, in this system, BMP-4 will also induce expression of mesodermal markers, but at a lower concentration than that required to induce blood cell formation. A similar graded response is found in established haematopoiesis. At low concentrations, BMP-4 induces cord blood-derived CD34+ stem cells to proliferate and differentiate, whereas at high concentrations the long-term-repopulating capacity and stem cell phenotype is prolonged (Bhatia et al, 1999).

Because of the difficulties in evaluating effects in vivo in the mouse embryo, information about the downstream effectors of BMP-4 signalling has largely come from studies in Xenopus. The transcriptional regulators GATA-2 and SCL have been shown to be essential for definitive haematopoiesis. Xenopus ectodermal AC cells transiently activate GATA-1 and 2 but do not normally commit to the blood lineage. Injection of BMP-4 RNA into AC cells results in increased GATA-2 expression and concomitant stimulation of globin synthesis. Conversely, expression of a dominant negative BMP-4 receptor in ventral mesoderm suppresses GATA-2 (Maeno et al, 1996). In culture, BMP-4 induces AC cells to commit fully to the blood programme, after mesoderm has formed, via induction of SCL (Zhang & Evans, 1996). These data imply that, in Xenopus, BMP-4 receptor signalling is required for haematopoietic induction from mesodermal cells.

It has been suggested that the role of BMP-4 in haematopoietic induction may be to pattern the mesoderm, by influencing gene expression, in co-operation with other mesoderm-inducing factors, such as fibroblast growth factor (FGF) and activin (Huber et al, 1998). Although BMP-4 induces ventral blood island (VBI) formation and GATA-2 expression, injection of embryonic FGF into ventral Xenopus tissue appears to have the opposite effect, suppressing these events (Xu et al, 1999). Moreover, expression of a dominant negative FGF receptor in lateral mesoderm, which does not normally form blood tissue, results in a dramatic expansion of the VBI. FGF induces expression of a transcription factor, PV.1, which inhibits blood development. One interpretation of these observations is that FGF induction of PV.1 may negatively regulate BMP-4 induction of GATA-2 to allow expansion of the haemangioblast pool. This role for FGF is supported by studies in murine ES cells. As previously mentioned, cultured ES cells can give rise to blast colony-forming units (BL-CFCs) which are thought to represent the haemangioblast and express Flk-1. ES cells in which the FGF receptor 1 gene fgfr1–/– has been knocked out, blocking FGF signalling, differentiate poorly and produce fewer haematopoietic colonies compared with wild type (Faloon et al, 2000). Embryoid bodies (EBs) generated from fgfr1–/–cells express normal levels of BMP-4 but have reduced SCL, Flk-1, c-Kit and globin expression. Addition of basic (b)FGF during EB differentiation increases the number of BL-CFCs/haemangioblasts and Flk-1+ cells produced, an effect which is enhanced by the addition of activin A. On the other hand, scl–/– EBs contain endothelial-type Flk-1+ cells but do not produce BL-CFCs or haematopoietic cells. As ES cell differentiation is analogous to early embryonic development, it would appear that FGF-mediated signalling is required for the proliferation, and possibly the development, of haemangioblasts and that SCL expression is subsequently required for the progression from haemangioblast to haematopoietic cell. It should be noted that FGF and activin do not appear to increase haematopoietic activity in EBs in the absence of BMP-4, again implicating BMP-4 in the haematopoietic pathway (Johansson & Wiles, 1995).

Another member of the TGF-β family, TGF-β1, is a potent negative regulator of blood cell proliferation in vitro. In culture, TGF-β1 directly inhibits the initial cell divisions of murine-derived LTR-HSCs (Sitnicka et al, 1996). However, the effects of TGF-β1 in vivo are more difficult to interpret. Targeted mutations in the murine TGF-β1 or TGF-β receptor type II genes frequently result in embryonic death at 10·5 DPC from severe anaemia as a result of reduced yolk sac vasculogenesis and haematopoiesis, not endothelial and haematopoietic cell hyperplasia as might be expected (Dickson et al, 1995; Oshima et al, 1996). During embryogenesis, therefore, TGF-β1 appears to regulate positively extraembryonic endothelial differentiation and primitive blood formation. Its role in intraembryonic haematopoiesis has not been fully investigated. Interestingly, in the avian system, it has been shown that the roof and sides of the dorsal aorta are derived from the somatopleuric mesoderm and contain endothelial cells which lack haematopoietic potential, whereas the ventral floor consists of splanchnopleuric mesoderm-derived cells with haemangioblastic bipotential (Pardanaud et al, 1996). Transplantation experiments involving the grafting of quail mesoderm onto chick hosts have shown that haemangioblastic potential can be conferred on somatopleuric mesoderm on transient contact with endoderm or treatment with VEGF (the ligand for Flk-1), bFGF or TGF-β1 (Pardanaud & Dieterlen-Lievre, 1999). Conversely, transient contact with ectoderm abolishes haemangiopoietic potential in splanchnopleuric mesoderm, suggesting that a positive and a negative gradient modulates the haemangioblastic and endothelial potential of mesodermal cells. In the murine AGM, transient expression of AML-1 in cells within the ventral floor of the aorta but not in the roof and sides would also appear to distinguish between two populations of endothelial cells with different potentials (North et al, 1999).

Expression patterns of candidate factors in the human embryo

In agreement with a role in ventral patterning, immunohistochemical analysis of the 34-d human embryo reveals a distinct gradient of BMP-4 expression across the dorsoventral axis within the AGM region. BMP-4 is expressed throughout the ventral mesoderm, which contains the dorsal aorta, but is noticeably restricted within the dorsal mesoderm (Marshall et al, 2000). Around the aorta, expression is polarized to the ventral floor within the stromal layer underlying the intra-aortic haematopoietic clusters. At 28 d, when clusters are just beginning to appear, this polarization to the aortic floor is even more striking as the dorsoventral gradient has not yet been established. By 38 d, after haematopoietic clusters have disappeared, BMP-4 expression is uniform around the entire aorta. This expression pattern is highly suggestive of a role for BMP-4 in the induction of haematopoiesis within the embryonic AGM. If BMP-4 does indeed influence haematopoietic development in a concentration-dependent manner, as is suggested by ES cell studies, a higher concentration of BMP-4 in the ventral floor of the aorta could explain why haematopoietic activity is restricted to this region and not detected around the roof and sides. BMP-4 is also expressed in cells surrounding blood islands in the human yolk sac and may also be involved in the generation of primitive blood cells.

In contrast to BMP-4, TGF-β1 expression is uniformly low around the dorsal aorta in the 34-d human AGM with no polarization to the ventral stroma. However, haematopoietic cells within the intra-aortic cluster expressed TGF-β1 at noticeably higher levels than the adjacent endothelium (Marshall et al, 2000). It has yet to be determined whether TGF-β1 is involved in the regulation of newly generated haematopoietic cells within the embryo, a role which it appears to play in the extraembryonic mesoderm, or whether it inhibits proliferation of these cells within the aorta, maintaining a primitive phenotype, before their migration to the fetal liver.

Morphogens and transcription factors: possible mechanisms for hsc ontogenesis

During embryogenesis, similar factors, or related family members, are involved in the regulation of a variety of systems. For example, members of the FGF and BMP families interact in a number of processes including lung bud and tooth morphogenesis. In the developing murine lung, FGF10 expression in splanchnic mesodermal mesenchyme marks prospective sites of lung bud formation and induces proliferation and chemotaxis of the overlying endodermal epithelium (Weaver et al, 2000). FGF10 induces a gradient of BMP-4 expression in the endoderm, which becomes localized to the tip of the extending lung bud, and inhibits FGF-induced proliferation. In this way, FGF and BMP-4 regulate lung bud branching in a localized and specific way to attain the correct pattern. In a parallel situation, BMP-4 and FGF8 combine to regulate the correct positioning of teeth in the developing mandibular arch. Ectodermal FGF signalling induces expression of the transcription factor Pax9 which marks the sites of tooth formation and may play a role in specifying primitive odontoblasts (Neubuser et al, 1997). FGF induction of Pax9 is inhibited by BMP-2 and BMP-4, restricting the domains of Pax9 expression and therefore tooth development. Interestingly, after these initial steps, BMP-4 has an opposite, inductive effect on odontogenesis, mediating tissue interactions in the developing tooth. These examples illustrate how FGFs and BMPs combine to mediate mesenchymal/epithelial interactions and to regulate cell specification across gradients, resulting in local and specific effects.

Morphogenesis generally can be divided into three main stages: (1) epithelial–mesenchymal interactions which determine sites of activity, (2) condensation (aggregation of similar cells) involving the determination and proliferation of precursors and their commitment to a particular lineage and (3) terminal differentiation. This last stage involves cell growth factor–ECM interactions. During tooth formation, condensed cells express the cell-surface proteoglycan receptor syndecan which binds the ECM molecule tenascin with high avidity and, in so doing, modulates the transition of cells from proliferation to differentiation (Hall & Miyake, 1995). Members of the TGF-β superfamily, including BMPs, are thought to be involved in the terminal differentiation of odontoblasts. Tenascin is upregulated during this process and it may function by immobilizing growth factors, such as BMP and TGF-β, favouring receptor binding (Ruch et al, 1995). BMPs and tenascin are also linked during chondrogenesis (cartilage formation). BMP-2 expression in combination with low-level tenascin is associated with epithelial–mesenchymal interactions, whereas higher concentrations of tenascin and BMP-4 expression during the condensation phase initiate chondroblast differentiation (Hall & Miyake, 1995).

Evidence from studies in Xenopus suggests that downstream effectors of BMP-4 signalling include the haematopoiesis regulators SCL and GATA-2. Members of the AML/Cbfa family of transcription factors have also been linked to the BMP pathway. AML3/Cbfa1 is required for osteoblast differentiation during bone formation (Otto et al, 1997). In vitro addition of a BMP4/7 heterodimer increases expression of AML3 in both osteoblastic and non-osteoblastic cell lines (Tsuji et al, 1998). Overexpression of a dominant negative truncated BMP receptor (type 1B) blocks AML3 expression and BMP-2-induced osteoblast differentiation in cultured murine calvariae cells (Chen et al, 1998). Exogenous BMP-2 can also induce novel and transient expression of AML3 in myoblastic cells, which would normally differentiate to muscle, and concomitant expression of early osteo-related genes, suggesting that BMP-induced AML3 expression may divert a committed myoblast to an osteogenic fate (Lee et al, 1999). In the light of these observations, the transient expression of AML-1 in the floor of the dorsal aorta in conjunction with the polarization of BMP-4 to this region is highly suggestive of a connection between BMP-4 and AML-1 in haematopoietic specification and budding in the embryonic AGM (Fig 4A).

Figure 4.

A proposed model, based on reported expression patterns and gene disruption studies described in the text, for the generation of haematopoietic cells from embryonic mesoderm within the mammalian AGM region. (A) BMPs and the AML family of transcriptional regulators have been implicated in other systems, including bone morphogenesis, in the commitment of precursor cells to a specific lineage. BMP-4 and AML-1 are expressed in distinctive patterns immediately before and during the formation of intra-aortic haematopoietic clusters and may play a critical role in embryonic haematopoietic specification. (B) BMP-4 and FGF may interact to convert splanchnopleuric mesoderm (P-Sp) to haemangioblast and to expand the haemangioblast pool before lineage commitment. BMP-4 may subsequently act on P-Sp-derived haemangioblasts or endothelial intermediates located in the floor of the dorsal aorta inducing a programme of haematopoietic gene expression including AML-1. The relative expression of Flk-1 (KDR) and SCL, high or low, may define or determine an endothelial or haematopoietic fate.

Based on expression patterns in the AGM and the observations described above, we have proposed a model for BMP-4, FGF and TGF-β1 signalling in the progression from mesoderm, through haemangioblast to committed haematopoietic progenitor in the human intraembryonic AGM (Fig 4B). BMP-4 is involved early in the specification of mesoderm and may subsequently interact with FGF to convert mesenchymal cells to haemangioblasts and to expand the haemangioblast pool before lineage commitment. A gradient of BMP-4, alone or with other factors, then induces a programme of haematopoietic gene expression, including AML-1, in haemangioblasts or endothelial intermediates located in the floor of the dorsal aorta. The acquisition or loss of receptors and transcriptional regulators defines each stage from bipotential precursor to lineage-committed/stem cell (Fig 4B). In the case of Flk-1 (KDR) and SCL, the relative expression of each may define or determine an endothelial or haematopoietic fate. Endothelial cells in the roof and sides of the dorsal aorta may be derived from somatopleural rather than splanchnopleural mesoderm and may lack haemangioblast potential from the outset or receive different developmental signals.

Future directions

The role of intraembryonic haematopoiesis in the establishment of a definitive blood system has become progressively more apparent and now appears to be a common developmental stage across many species. However, the microenvironmental determinants of haematopoietic specification from haemangioblastic precursors remain poorly understood. Fate determination ultimately depends on the induction and maintenance of a programme of gene expression within a cell. In individual cells, the response may be graded and may be dependent on concentration levels of signalling molecules such as FGF and BMP-4. Too little or too much of a particular signal may inhibit one programme of genes and permit or induce activation of an alternative programme. The successful generation of intraembryonic HSCs is therefore likely to be a balancing act involving the interplay of a number of factors on responding cells. The discovery of new molecules that regulate haematopoietic cell fate and the generation of inducible mutant systems will help to decipher the molecular pathways involved in intraembryonic blood formation. An understanding of these pathways will advance the development of techniques for ex vivo expansion of human HSCs and, consequently, their usefulness in clinical applications.


The authors would like to thank the Wellcome trust for support and to acknowledge and thank the following persons: Professor Christine Kinnon, Molecular Immunology Unit, ICH, for assistance in preparation of this manuscript. Rachel Moore, formerly in the Developmental Biology Unit, ICH, for preparation of human embryonic tissue through the MRC Human Embryo Bank. Professor Landon and Brian Moore at the Institute of Neurology, London, for preparation of electron micrographs. Professor Adrian Woolf, Nephrourology Unit, ICH, for use of the TIE-1/lacZ mouse model.