Inductive interactions mediated by interplay of asymmetric signalling underlie development of adult haematopoietic stem cells

During embryonic development, adult haematopoietic stem cells (HSCs) emerge preferentially in the ventral domain of the aorta in the aorta–gonad–mesonephros (AGM) region. Several signalling pathways such as Notch, Wnt, Shh and RA are implicated in this process, yet how these interact to regulate the emergence of HSCs has not previously been described in mammals. Using a combination of ex vivo and in vivo approaches, we report here that stage-specific reciprocal dorso–ventral inductive interactions and lateral input from the urogenital ridges are required to drive HSC development in the aorta. Our study strongly suggests that these inductive interactions in the AGM region are mediated by the interplay between spatially polarized signalling pathways. Specifically, Shh produced in the dorsal region of the AGM, stem cell factor in the ventral and lateral regions, and BMP inhibitory signals in the ventral tissue are integral parts of the regulatory system involved in the development of HSCs.

H aematopoietic stem cells (HSCs) lie at the foundation of the adult haematopoietic system, and give rise to cells of all blood lineages throughout the lifespan of an organism. An important property of adult (definitive) haematopoietic stem cells (dHSCs) is that they are capable of long-term reconstitution of the haematopoietic system upon transplantation into irradiated recipients. In the mouse, such cells develop by embryonic stages E10-E11 in the aorta-gonadmesonephros (AGM) region [1][2][3][4] . An ex vivo approach showed that the AGM region has a robust autonomous capacity to generate dHSCs 1 . The AGM region comprises the dorsal aorta flanked on both sides by the urogenital ridges (UGRs), which contain embryonic rudiments of kidney and mesonephros. HSCs develop in a polarized manner, predominantly in the ventral floor of the dorsal aorta (AoV), more rarely in the dorsal domain of the dorsal aorta (AoD), and are absent in the UGRs 2,5-7 . Localization of dHSCs to the AoV in mouse and human embryos was shown by long-term reconstitution experiments 5,6 .
Abundant evidence indicates that during development, a specialized embryonic endothelial compartment known as haematogenic (or haemogenic) endothelium gives rise to haematopoietic stem and progenitors cells [7][8][9][10] . The haematopoietic programme in various vertebrate models is executed predominantly in the AoV, and is recognized by the expression of essential haematopoietic transcription factors, for example, Runx1 and cKit, and the appearance of clusters of haematopoietic cells budding from the endothelium of the dorsal aorta 6,8,9,[11][12][13][14] .
It is broadly accepted that HSCs develop from the haematogenic endothelium within intra-aortic clusters. This transition involves several consecutive maturation steps of HSC precursors: pro-HSCs-pre-HSC type I-pre-HSC type II-dHSC [15][16][17] . All these precursors express endothelial markers, such as vascular-endothelial cadherin (VC) and CD31, and sequentially upregulate haematopoietic surface markers: CD41 (pro-HSCs), CD43 (pre-HSC type I) and finally CD45 (pre-HSC type II). This maturation process occurs in the dorsal aorta between E9 and E11. Specifically, pro-HSCs emerge at E9, pre-HSCs Type I appear at E10 and pre-HSCs type II predominantly at E11. Unlike dHSCs, pre-HSCs cannot reconstitute the adult haematopoietic system by direct transplantation and require prior maturation in an embryonic or neonatal environment [15][16][17][18][19] .
A number of signalling pathways (Notch, Wnt, retinoic acid, interleukin-3 and inflammatory) have been implicated in HSC development; however, a coherent picture is yet to be elucidated 15,17,[20][21][22][23][24][25][26][27][28][29][30][31] . HSC precursors (pro-HSCs, pre-HSCs type I and pre-HSCs type II) express cKit 17 from early developmental stages. A recent study has shown that the cKit ligand, known as stem cell factor (SCF), is a key regulator driving maturation of these HSC precursors into dHSCs in the AGM region 17 , which is in agreement with the marked decline of HSC activity in SCF mutant mice 32,33 . In the adult, SCF is critically important for HSC maintenance in the bone marrow niche, mainly in the endothelial compartment 32 . Sonic Hedgehog (Shh) and bone morphogenetic protein 4 (BMP4) pathways are also important mediators; in zebrafish, these two morphogenes are involved in arterial specification and haematopoietic patterning, respectively 34,35 . In the mouse, subaortic BMP4 and Shh/Indian Hedgehog derived from gut were also proposed to be responsible for HSC development 36,37 .
During development, interactions between spatially segregated compartments are essential for tissue patterning and specification, and are often mediated by gradients of secreted molecules [38][39][40] . Molecules secreted by distant tissues, such as somites, can influence HSC development in the AGM region [41][42][43][44][45] . Developing HSCs are embedded in the complex AGM microenvironment, suggesting that HSC development may require signals derived from different compartments of the AGM region. We sought to test this hypothesis. However, the analysis of HSC development in vivo is significantly hampered by low accessibility of embryos developing in utero, fast maturation of dHSCs, lack of uniquely specific markers for HSC precursors and their low numbers in the AGM region. Therefore, we employed here a robust ex vivo culture system that models HSC development in the embryo in combination with functional HSC analysis using in vivo long-term reconstitution assay [15][16][17] . Specifically, to study interactions between AGM subregions, we took advantage of the in vitro reaggregation system that enables close juxtaposition of cell types 15 .
We show that interactions between three compartments of the AGM, the AoV, the AoD and the UGRs, are necessary for efficient generation of dHSCs. First, we show that dHSC activity in the isolated E10.5 AoV is limited but can be significantly enhanced by co-culture with the AoD, and that this is mediated at least partly by Shh, secreted dorsally in vivo. Second, while HSC activity in isolated E11.5 AoD is limited, co-culture with a competent AoV microenvironment activates dHSC generation in the AoD. We found that this effect is mediated by SCF, which is secreted abundantly by the AoV stroma in vivo as shown here. Third, we show that downregulation of BMP4 signalling by BMP antagonist Noggin, which is present at high levels in the AoV and especially in intra-aortic clusters as revealed here by in vivo observations, is required for HSC development. Fourth, UGRs, which express high levels of SCF, also enhance HSC development in the dorsal aorta.
Our results based on in vivo observations and ex vivo modelling strongly suggest that juxtaposed, anatomically distinct domains within the AGM region create a complex landscape of interactive signals that underpins HSC development.

Results
Pre-HSCs localize preferentially to the AoV. As dHSCs mature from pre-HSCs, we investigated whether the emergence of dHSC predominantly in the AoV 6 is a result of asymmetric (ventralized) distribution of pre-HSCs. Dorsal aortae were separated from UGRs and bisected into AoV and AoD (including notochord) as described previously 6 ( Supplementary Fig. 1a). The different domains were then directly transplanted into irradiated mice to detect dHSCs. We first confirmed our previous observation that at E11.5 dHSCs appear almost exclusively in the AoV, although some dHSCs were in the AoD and engrafted few recipients at high level ( Supplementary Fig. 1b). Limiting dilution analysis showed that dHSCs are approximately four times more frequent in the AoV compared with AoD. UGRs did not contain HSCs in line with previous reports 2, 6 .
We then investigated the spatial distribution of pre-HSCs type I and pre-HSCs type II in E10.5-E11.5 embryos using the OP9 co-culture system supplemented with Il3 þ SCF þ Flt3 (termed 3GF), which allows pre-HSCs (which do not engraft by direct transplantation) to mature into dHSC that become detectable by long-term repopulation assay as described previously 16 . Doses of transplanted cells (expressed in embryo equivalents, e.e.) were chosen based on the requirements of individual experiments (explained in Methods section). In these experiments (Fig. 1), the dose injected was high (1-2 e.e.) to detect potentially low dHSC numbers in AoD and UGRs.
We have shown previously that E10.5 AGM region mainly contains type I pre-HSCs, whereas at E11.5, type I and type II pre-HSCs co-exist 16  AoV, 20 showed high levels (470%) of donor-derived long-term haematopoietic chimerism (Fig. 1a). In contrast, only 7 out of 16 recipients of cultured AoD were repopulated at high levels (470%), while the remaining recipients showed lower or no repopulation (7 and 2, respectively). Cultured UGRs did not produce dHSCs (Fig. 1a). Thus, we conclude that the E10.5 AoD does contain pre-HSCs but at significantly lower numbers than the AoV.
We then investigated whether pre-HSCs localization changes in E11.5 embryos and found that pre-HSCs were still exclusively localized to the dorsal aorta; UGRs carefully separated from the lateral mesenchyme adjacent to the dorsal aorta did not give any repopulation after culture (Fig. 1b). To establish the location of pre-HSCs within the E11.5 dorsal aorta, cell populations enriched for pre-HSCs type I (VC þ CD45 -) and pre-HSCs type II (VC þ CD45 þ ) were sorted from AoV and AoD, and co-cultured with OP9 stromal cells in the presence of 3GF as described previously 16 . We again were able to detect pre-HSC activity in AoD although at lower levels than in AoV. After maturation ex vivo, pre-HSCs type I from AoV and AoD repopulated 7 of 11 and 2 of 8 recipients, respectively (Fig. 1c). Similarly, cultured pre-HSCs type II from AoV and AoD repopulated 11 out of 12 and 4 out of 10 recipients, respectively (Fig. 1d). In all cases, multilineage engraftment was confirmed ( Supplementary Fig. 2). These data show that pre-HSCs are significantly enriched in AoV.
Reciprocal inductive interactions between AoD and AoV. To explore hypothetical interactions between AoD and AoV, we made use of a dissociation-reaggregation system that recapitulates HSC development ex vivo 15 . This system allowed us to integrate AGM domains in a three-dimensional tissue-like organoid 15 and study their interactions in HSC development. To track the origin of dHSCs, AoV and AoD from wild-type (WT) and green fluorescent protein (GFP) embryos with constitutive expression of GFP 46 were co-aggregated (termed AoV//AoD co-aggregates) and cultured for 5 days in the presence of 3GF before transplantation (Fig. 2a). Mice transplanted with AoV//AoD co-aggregates can be reconstituted by dHSCs coming from AoD and AoV. The presence of GFP allowed the individual contributions of AoV and AoD to the total repopulation level within the same mouse to be assessed (Fig. 2b,c). This is presented in two separate columns in the graph. Namely, while columns 1 and 3 represent the same recipient mice, the former shows exclusively the contribution of the AoD and the latter shows exclusively the contribution of the AoV into each recipient. To assess the influence of AoD and AoV interaction on HSC development, the repopulation by co-aggregated AoD (column 1) or AoV (column 3) can then be compared with repopulation by independently cultured AoD (column 2) or AoV (column 4). All experiments included reciprocal use of WT and GFP tissues in AoV//AoD co-aggregates, and we observed no difference in repopulation properties between WT and GFP embryos. Homotypic AoV//AoV and AoD//AoD co-aggregates were always used as controls. Note that in these experiments, only 0.2 e.e. were injected per recipient, to ensure that the repopulation levels were not saturated and to allow any inductive effects to be revealed.
Using this approach, we found that the E10.5 AoV generates more dHSCs when combined with AoD than on its own (Fig. 2b, compare two rightmost columns). One day later, E11.5 AoD had no positive influence on dHSC generation by AoV (Fig. 2c, compare two rightmost columns). Conversely, the E11.5 AoD produced more HSCs when reaggregated with the AoV than on its own (Fig. 2c, compare two leftmost columns). This inductive effect of AoV on AoD was not observed at E10.5 (Fig. 2b, compare two leftmost columns). These ex vivo modelling experiments revealed reciprocal stage-specific effects of AoV and AoD on HSC development, which could be explained by the differential release of factors by the two regions and/or by differences in the competency of the target cells to respond to signals.
UGRs enhance HSC development in the dorsal aorta. UGRs do not develop HSCs within their own microenvironment (Fig. 1a,b). We tested whether the competent environment of the dorsal aorta can induce dHSCs in UGRs. Chimeric WT//GFP co-aggregates of E11.5 aortas and UGRs were cultured in the presence of 3GF, and then transplanted into irradiated mice (Fig. 2a). Mice injected with these reaggregates could be repopulated with dHSCs coming from the aortas and/or UGRs. The presence of GFP allowed the contributions of aortas and UGRs to the total repopulation level in the recipient mice to be assessed and these are presented in the two separate columns (Fig. 2d). Long-term haematopoietic repopulation originated exclusively from the dorsal aorta and not from UGRs (Fig. 2d), demonstrating that the dorsal aorta cannot induce dHSC generation in the UGRs. We then investigated whether UGRs could induce dHSCs in the dorsal aorta. When comparing aorta alone to aorta cultured with UGRs, we found that the presence of UGRs significantly enhanced HSC development in the dorsal aorta (Fig. 2e). This positive effect of UGRs was observed for both AoD and AoV at E10.5 and E11.5 ( Supplementary Fig. 3).
SCF expression is involved in polarized HSC development. The results described above could be explained by the differential production of factors by the AoD, AoV and UGRs. We have recently shown that SCF is a potent inducer of early HSC precursors in an ex vivo system 17 . Quantitative reverse transcription-PCR (qRT-PCR) showed that SCF expression in the AGM region is polarized, exhibiting higher levels in AoV and UGRs compared with the AoD (Fig. 3a). Clear segregation between the dorsal and ventral compartments of the AGM region can be readily visualized by enhanced expression of SCF in the AoV and UGRs using SCF-GFP knock-in mice 32 ( Fig. 3b) or direct detection of SCF protein ( Supplementary Fig. 4). Both non-endothelial stroma and endothelial cells expressed SCF at high levels, but the former showed clear dorso-ventral polarization in SCF expression (Fig. 3c). Importantly, we found that the pre-HSC type I population also expressed SCF, suggesting that a positive-autocrine loop may be involved in their development (Fig. 3c). We then tested whether SCF can induce dHSC formation in the AoD. Addition of soluble SCF efficiently induced dHSC generation from E10.5 and E11.5 AoD explants (Fig. 3d,e). SCF could also enhance dHSC generation by AoV (Fig. 3f). Conversely, recombinant soluble SCF receptor (SCF-Rh), which binds SCF with high affinity and blocks activation of the cKit receptor, significantly reduced the number of repopulated mice transplanted with AoD cultures compared with SCF-treated or untreated control explants (Fig. 3d).
Shh signalling enhances dHSC generation. Since dissected AoD tissue includes the notochord, it is likely to be a rich source of Shh 47 . We confirmed this by qRT-PCR ( Fig. 4a) and by analysis of GBS (Gli-binding site)-GFP 48 reporter mice for Hh pathway activity ( Supplementary Fig. 5). A significantly higher level of Hh activity was observed in the AoD compared with the AoV tissue ( Supplementary Fig. 5). We therefore tested whether Shh could be responsible for the positive impact of AoD on AoV-derived dHSCs. Indeed, addition of recombinant Shh protein to cultures lacking exogenous cytokines significantly increased HSC development in E10.5 AoV (Fig. 4c). Similarly, a Shh doxycycline-inducible OP9 cell line (doxOP9-Shh; Supplementary Fig. 6) co-aggregated with E10.5 AoV cells, also increased HSC development from AoV after doxycycline induction and this was abrogated in the presence of an Hh antagonist (Fig. 4d). The explant and co-aggregation analyses above suggest that the inductive effect of AoD on AoV observed at E10.5 is at least partly mediated by Shh.
At E11.5, addition of Shh could no longer induce HSC development in AoV (Fig. 4f), which is in line with the more subtle effect of AoD on AoV at this stage (Fig. 2c). In keeping with this result, addition of the Hh antagonist inhibited HSC development in the E10.5, but not in the E11.5 AGM region ( Fig. 4e,g). Notably, qRT-PCR analysis revealed significant expression levels of Patched1 (Hh receptor) and Gli1 (Hh target) in pre-HSCs type I, but almost no expression of these genes in pre-HSC type II (Fig. 4b). These data suggest that Shh signalling acts specifically on pre-HSCs type I and, as E10.5 AGM contains Left column: to test interactions between AoV and AoD, chimeric AoV//AoD reaggregates were generated and transplanted into irradiated recipients after 4-5 days of culture (b,c). Right column: to test interactions between Ao and UGRs, chimeric Ao//UGR reaggregates were generated and transplanted into irradiated recipients after 4-5 days culture (d). GFP þ and/or GFP À donorderived long-term repopulation allowed us to conclude whether dHSCs originated from AoV, AoD or UGRs. Accordingly, the tissue of origin of donor dHSCs is indicated below each graph. (b) E10.5 aortas from WT and GFP embryos were used to generate chimeric reaggregates as depicted schematically above plots. The reciprocal combination of WT and GFP tissues was used to generate AoV//AoD reaggregates. The tissue source of dHSCs is shown separately in the leftmost (AoD) and rightmost (AoV) columns as indicated below the plot (0.2 e.e. per recipient; two independent experiments).
(c) E11.5 aortas from WT and GFP embryos were used to generate chimeric reaggregates. The tissue source of dHSCs is shown separately in the leftmost (AoD) and rightmost (AoV) columns as indicated below the plot (0.2 e.e. per recipient; two independent experiments). (d) E11.5 aortas (Ao) and UGRs from WT and GFP embryos were used to generate Ao//UGR chimeric reaggregates. As depicted schematically above the plot, the reciprocal combination of WTand GFP tissues was used to generate Ao//UGR reaggregates. The tissue source of dHSCs is shown separately in left (Ao) and right (UGRs) columns as indicated below the plot (0.01 e.e. per recipient; six independent experiments).
(e) Reaggregation of WT Ao with UGRs generate more dHSCs than Ao alone (0.05 e.e. per recipient; two independent experiments). (b-e) In all these experiments, tissues were cultured with three growth factors.
almost exclusively pre-HSCs type I (ref. 16), they provide an explanation for the stage-specific effect of Shh. Why are dHSCs absent in the Shh-rich AoD in vivo? We found that expression of Patched1 and Gli1 is considerably lower in pre-HSC type I derived from the AoD compared with AoV ( Fig. 4b), suggesting their reduced sensitivity to Shh. Accordingly, the stimulatory effect of Shh on AoD explants was insignificant compared with its effect on AoV explants (Fig. 4c). Low sensitivity of dorsally localized pre-HSC type I to Shh may be responsible for attenuated maturation of HSCs in AoD in vivo.
BMP signalling is downregulated in the dHSC lineage. The emergence of dHSCs in the AoV has been previously attributed to ventrally localized BMP signalling 34,36,49 . As expected, qRT-PCR showed that BMP4 is expressed at higher levels in the AoV compared with the AoD (Fig. 5a). Accordingly, BMP4 and phosphorylated forms of Smad1,5,8 (P-Smad), indicative of active Bmp signalling 50 , were abundant in the ventral subaortic mesenchyme (Fig. 5b,c,e-g; Supplementary Fig. 7).
However, when we analysed BMP targets genes (Id1, Id2 and Id3) in cell populations directly isolated from the embryo, we found that Id genes are highly expressed in endothelial cells and pre-HSCs type I, but their expression is significantly lower in pre-HSC type II (Fig. 5d), suggesting a decrease of BMP activity between pre-HSC type I and type II stage. Accordingly, at the morphological level, P-Smad is highly expressed in some luminal aortic endothelial (CD31 þ CD45 À ) cells and at the base of intra-aortic haematopoietic clusters ( Fig. 5e-g; Supplementary  Fig. 7). Within the clusters, P-Smad levels were lower in CD31 þ CD45 À /low cells (Fig. 5e,g, green arrows) and further reduced in cells localized at the periphery of the cluster that acquired CD45, which, together with CD31, defines the pre-HSC type II phenotype ( Fig. 5e-g, white arrowheads; Supplementary Fig. 8). Taken together, our data suggest that BMP activity is downregulated during progression of the HSC lineage in vivo (Fig. 6e).
How can high BMP4 levels in the AoV be reconciled with the downregulation of BMP activity in the HSC lineage? qRT-PCR showed that BMP antagonists such as Noggin, Chordin and Chordin-like2 are expressed at high levels in the AoV in vivo (Fig. 6a). Indicatively, inhibitory Smad6 and Smad7 are also expressed at higher levels in AoV, compared with AoD (Fig. 6a). Furthermore, immunostaining experiments confirm the preferential expression of Noggin in the ventral domain of the AGM (Fig. 6b). Thus, BMP antagonists may be responsible for downregulation of BMP signalling in developing HSCs.
Confocal analysis showed that BMP4 is mainly expressed in subendothelial mesenchyme and is barely detectable in intra-aortic clusters in vivo (Figs 5b and 6c; Supplementary  Fig. 9a). By contrast, although Noggin is also expressed in subendothelial mesenchyme (Fig. 6b), its expression is readily observed in haematopoietic clusters ( Fig. 6c; Supplementary  Fig. 9b). Accordingly, qRT-PCR analysis showed that Noggin is highly expressed in the ventral mesenchyme and at lower but significant levels in pre-HSCs type I (Fig. 6d). Thus, BMP signalling inhibits HSC development. To assess the impact of BMP signalling on HSC generation, we used AGM cultures supplemented with BMP4 and assessed dHSC production by the long-term transplantation assay. As the fetal calf serum contains BMP4 (ref. 51), we used a serum-free culture system and demonstrated a consistent dose-dependent inhibitory effect of BMP4 on HSC development in E10.5-E11.5 AGM cultures with concomitant increase in the expression of BMP target genes (Fig. 7a-c; Supplementary Fig. 10a).
By contrast, addition of Noggin (Z100 ng ml À 1 ), in the presence or absence of serum, considerably enhanced production of dHSCs in AoV explants and decreased the expression of BMP target genes ( Fig. 7d; Supplementary Figs 10b and 11). Notably, Noggin could also induce HSC development in AoD explants in the presence and particularly in the absence of serum ( Fig. 7d;  Supplementary Fig. 11). These results suggest that efficient HSC development requires that BMP signalling in the AGM region is kept at relatively low levels.
Interactions between SCF, Shh and BMP signalling pathways. The above studies indicate that HSC development in the AGM region is orchestrated in a complex signalling environment consisting of overlapping gradients of SCF, Shh and BMP.
We tested whether these signalling pathways can influence each other. We obtained no conclusive data to suggest that addition of Shh/BMP4 modulates SCF expression and vice versa ( Supplementary Fig. 12). However, supplementation of AGM cultures with Shh resulted in consistent upregulation of Noggin and Smad6 (Fig. 8a). Noggin in turn enhanced Shh expression, thus creating a positive-feedback loop supporting maturation of HSCs (Fig. 8b). Conversely, BMP4 reduced Shh and Ptch1 expression (Fig. 8b). These results obtained using ex vivo modelling system indicate that spatially polarized Shh, SCF and BMP signalling in the AGM region are integral parts of an epistatic regulatory system that drives HSC development in the AGM region in vivo (Fig. 8c).

Discussion
We have shown previously that during murine embryo development definitive HSCs emerge predominantly in the ventral domain of the dorsal aorta (AoV) 6 . This spatially polarized production of HSCs might be explained by different origins of dorsal and ventral endothelium and/or by asymmetric production of key factors involved in HSC development 37,52,53 and we reasoned that directional inductive interactions between AGM compartments could be involved. Great insight into inductive interactions in various organs has previously been obtained through in vitro modelling 39 . Here we modelled interactions between AGM domains in a co-culture system, which supports HSC development 15 . Using this ex vivo system, we demonstrate that at early stages (E10.5) HSC maturation in the AoV region is enhanced by the presence of the AoD. One day later (E11.5), the AoV microenvironment is able to induce HSC development in the AoD, previously thought to be mostly devoid of HSC activity 6 . We also found that UGRs can enhance HSC production from the dorsal aorta, but cannot generate dHSCs themselves, even under influence of the dorsal aorta. Thus, our data strongly suggest that reciprocal stage-specific inductive AoD//AoV interactions and involvement of UGRs are required for execution of the robust development of HSCs in vivo.
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms10784 ARTICLE (both type I and type II), their numbers are lower than in AoV, in line with lower intra-aortic cluster formation previously described in mouse AoD 6,13 . Second, as shown here, dHSCs can be induced in the AoD by the AoV, and therefore the dHSCs deficiency in AoD cannot be explained solely by asymmetric pre-HSC distribution, but may also be influenced by differences in the microenvironment.
To study this, we focused on SCF, Shh and BMP4, whose expressions are dorso-ventrally polarized in the AGM region 36,47,49 (and current data). We found that SCF is an inductive signal that is expressed at high levels in the AoV and UGRs, and can stimulate HSC development in isolated AoD, a region which had previously been considered to be mostly devoid of HSC activity. This is in agreement with a key role of SCF in HSC maturation 17 . We found that the aortic endothelial compartment expresses high levels of SCF, suggesting its important role in HSC development comparable to the bone marrow microenvironment of adult HSCs 32 . Importantly, we found that the pre-HSC type I population expresses SCF suggesting a positive-autocrine loop, which could promote HSC development.
Shh signalling in zebrafish is required for aortic angioblast migration and subsequent arterial specification of the dorsal aorta 34,54 . We found that in mouse Shh stimulates and a Hh antagonist inhibits the development of HSCs at E10.5 but not at E11.5, in keeping with a previous study 37 . The induction of dHSCs in AoV by AoD is also limited to the E10.5 stage. Since Shh is secreted by the notochord (which is included in AoD-dissected tissue), this stage specificity is likely defined by the predominant presence of pre-HSCs type I at E10.5, which express higher levels of Shh signalling components (Ptch1 and Gli1) compared with pre-HSCs type II. By E11.5, the pre-HSC population is mainly represented by type II cells 15 . Stage-specific loss of sensitivity to Hh signalling was also described in the developing neural tube 55 . Notably, the poor ability of AoD to develop HSCs despite abundant presence of Shh can also be explained by lower levels of Ptch1 and Gli1 detected in AoD-compared with AoV-derived pre-HSC type I. Our ex vivo modelling data indicate that AoD-derived Shh is an active inducer of HSC development in the AGM region. This conclusion does not exclude the possibility that Shh secreted by the gut could also reach the dorsal aorta 37 , although by E10.5 these sites are separated by an extended mesentery.
BMP4 signalling is a key factor involved during differentiation of ventral mesoderm and its further specification into haematopoietic cells. In zebrafish, BMP signalling is clearly required during the patterning of the dorsal aorta and for the emergence of dHSCs in the ventral wall 34 . Its role in mouse is less clear due to the early lethality of BMP mutants 56 . Several lines of evidence point to BMP4 as a good candidate regulating HSC development. Indeed, BMP4 is highly expressed in the ventral mesenchyme underneath the dorsal aorta 34,36,49 ; some reports suggested its role in controlling dHSC emergence 36,57,58 . However, the in vitro systems used likely assayed the maintenance of dHSCs, rather than their maturation. It was also reported that BMP4 signalling can define their differentiation potential 59 . BMP4 is also involved in the regulation of essential haematopoietic transcription factors such as Scl/Gata2/Fli1 and Runx1 (refs 60,61). Here we analysed BMP signalling activity in the dHSC lineage in the AGM region. We show that in vivo the pre-HSC type I to type II transition is accompanied by a downregulation of BMP targets (Id genes).
This correlates with our data demonstrating that BMP activity is downregulated in intra-aortic clusters and the observations of others that Runx1 expression is attenuating in the developing HSC lineage 60,62,63 . How is this decrease of BMP activity achieved in vivo, despite the presence of BMP4 in AoV? It has previously been noted that in amphibian embryos several BMP inhibitors are also expressed in AoV 34 . Similarly, our analysis of the embryo showed high expression of a number of BMP antagonists as well as inhibitory Smad6 and Smad7 in mouse AoV that may counteract BMP4 action in HSC lineage. Furthermore, we found that in the AGM region BMP4 and Noggin are spatially segregated: Noggin being present in haematopoietic clusters and BMP4 being mainly expressed underneath the aortic endothelium. Therefore, maturing HSCs in clusters are exposed to low BMP4 concentration and high concentration of the BMP antagonist Noggin. Furthermore, our qRT-PCR analysis shows that the pre-HSC type I population expresses Noggin, which possibly creates a very effective shield that protects them from BMP4. Accordingly, our ex vivo analysis strongly suggests that downregulation of BMP signalling is functionally important for HSC development in the embryo. Indeed, forced BMP signalling activation by the addition of BMP4, strongly inhibits HSC development, and conversely the addition of Noggin stimulates HSC development in E10.5-E11.5 AGM cultures. These results are in line with recent observation that deletion of Smad4, a common transducer for BMP4/TGFb signalling, markedly augments the formation of intra-aortic clusters 64 . Our data do not exclude the possibility that BMP4 is essential for specification of mouse dHSCs at earlier stages, as described in the zebrafish model, where BMP signalling is required for HSC development at stages closer to mouse E8.5 (ref. 34). Our analysis indicates that all three signalling pathways studied can cooperate for HSC development (Fig. 8c). Notably, the interplay of Shh and BMP pathways is broadly involved in development. For example, counter gradients of polarized Shh and BMP signalling in the developing spinal cord specify neuronal subsets along the dorso-ventral axis 65 , and the dorsal aorta resembles the neural tube with inverse orientation of Shh-and BMP-secreting centres 34 . However, we detected an antagonistic relationship between Shh and BMP pathways. At the molecular level, Shh can induce Noggin and Smad6 expression, thus inhibiting BMP4 signalling. In turn, BMP4 suppresses and, accordingly, Noggin enhances Shh signalling. Cooperation between Shh and Noggin has been previously described as critically important for developmental specification of somitic, neural and hair follicle cells [66][67][68] . Our in vitro data suggest that the feed-forward loop Shh-Noggin/Noggin-Shh is also involved in HSC development in vivo.
We propose a model where the polarized secreted factors form complex fields of gradients in vivo, which define an effector zone for optimal HSC development in the dorsal aorta and lead to the ventrally shifted appearance of dHSCs (Fig. 8c). Of interest, intra-aortic clusters are abundant in ventro-lateral positions 69 , which may reflect the position of this zone. The dissection close to such a zone could lead to accidental inclusion of powerful dHSCs in AoD samples observed here. Furthermore, it is possible that spatial segregation of co-operating and spatial overlap of antagonizing factors may also be important for adjustment of HSC development in vivo. Indeed, although the pool of pre-HSCs in the AGM region markedly expands during E9.5-11.5 (Rybtsov et al., submitted), complete maturation of the HSC pool is limited: while the majority of cells reach the pre-HSC type II stage, only one or two dHSCs are generated by the end of E11. Such controlled dynamics of HSC development may be needed to prevent a burst of active haematopoiesis in the AGM region. How exactly HSC maturation dynamics depend on overlapping concentrations of factors requires further analysis. Although ex vivo modelling is a powerful tool to dissect mechanisms of HSC development in vivo, there will likely be some variation in details. For example, spatial polarization in the developing HSC niche may define kinetics of HSC development in vivo. While we have demonstrated spatial polarization in vivo of the factors driving HSC development in our model system, it is currently unclear whether any factors become expressed in a polarized manner within the reaggregates and as such, whether polarization is also a pre-requisite for HSC maturation. Alternatively, if polarization is not required, the entire reaggregate may replicate the optimal zone for HSC development, resulting in massive generation of dHSCs. The distinction between these two scenarios will require further investigation.
In summary, our ex vivo modelling experiments suggest that HSC development in the embryo involves stage-dependent interactions between dorsal, ventral and lateral domains of the AGM region, mediated at least partly by the interplay of SCF, Shh, BMP4 and Noggin. Further detailed analysis will be required to better understand the complexity of the AGM signalling landscape in which HSC development takes place. Such knowledge may lead to development of novel protocols for the generation of definitive HSCs in vitro for clinical applications.  Explant and reaggregate cultures. Isolated E10.5 and E11.5 AGM regions were subdissected into AoV, AoD and UGRs. The notochord was included in the AoD. AGM regions were either cultured as explants or dissociated by collagenase/dispase and then either self-reaggregated or co-aggregated with OP9 stromal cells. For self-reaggregation, AGM cell suspensions were centrifuged at 430g for 12 mn in 200-ml pipette tips sealed with parafilm to form a pullet. For co-aggregation with OP9, cell suspension of 1 e.e. of AGM cells or 1 e.e. of sorted cells were mixed with 10 5 OP9 cells prior centrifugation. Cell aggregates or explants were cultured at the liquid-gas interface on 0.8-mm nitrocellulose filters (Millipore) for 5 days (37°C, 5% CO 2 ) in 5 ml of IMDM (Invitrogen), 20% fetal calf serum supplemented with L-glutamine (4 mM), penicillin/streptomycin (50 U ml À 1 ), and 100 ng ml À 1 SCF, 100 ng ml À 1 IL3 and 100 ng ml À 1 Flt3l (termed 3GF), unless otherwise indicated (all purchased from Peprotech). For generation of chimeric reaggregate cultures, AGM regions isolated from age-matched BL/6 WT and BL/6 GFP embryos were subdissected into AoV, AoD and UGRs, and after cell dissociation combined in reaggreates.
The different culture systems used in our study are listed in Supplementary  Table 1, as well as indications as to their efficiency in terms of HSC development.
Mouse recombinant proteins Bmp4, Noggin, Shh and human SCF receptor were purchased from R&D Systems and the Hh antagonist was a gift from Genentech (Dr Stephen Gould).
Transplantation assay. AGM tissues from C57BL/6 CD45.2/2 embryos were pooled (unless otherwise stated) and cell suspensions obtained after dissociation with collagenase/dispase (Roche) for 40 min at 37°C. Donor cells were injected intravenously into C57BL/6 CD45.1/2 sublethally irradiated (1,150 rad) mice along with 20,000 C57BL/6 CD45.1/1 bone marrow carrier cells. The amount of transplanted cells is expressed in embryo equivalents, defined as a unit of cells equivalent to the number present in one organ. (for example, 0.2 e.e. corresponds to 20% of cells present in one AGM region). The number of embryo equivalents injected for each experiment was chosen based on the expected outcome of dHSC numbers, which can vary for a given tissue depending on culture conditions. For example, the use of different culture systems (co-aggregation with OP9 cells, self-reaggregation or explants) in the presence or absence of cytokines gives different HSC outcomes. Depending on the experimental question, we had to adjust the number of embryo equivalents injected to detect a small number of dHSCs (high dose of cells injected) or to avoid saturation of repopulation to allow assessment of inductive factors/tissues (low dose of cells injected).
Quantitative reverse transcription-PCR. Total RNA was isolated from AoV, AoD, UGRs (using at least 4 e.e. of tissue for each sample) and OP9 with Qiagen RNeasy microkit or minikit (Qiagen), and complementary (cDNA) was prepared using SuperScriptVILO cDNA Synthesis (Invitrogen). The qRT-PCR was performed using the LightCycler 480 SYBR Green I MasterMix (Roche). For gene expression on a small number of cells, total RNA was isolated from 200 cells directly FACS-sorted into a PCR tube containing 10 ml of Â 2 Reaction Mix (CellsDirect, Invitrogen) and 0.2 ml RNase inhibitor (SUPERase-In Ambion AM2694). cDNA preamplification of the genes of interest was achieved by adding directly Superscript III/Taq mix (CellsDirect) and primers (10 mM each) to the cell lysate (PCR programme: 50°C for 15 min; 95°C for 2 min; 18 cycles of: 95°C for 15 s; 60°C for 4 min). A universal probe library strategy with Lightcycler 480 probes mastermix kit (Roche) was used to perform qRT-PCR on these samples. Duplicate reactions were set-up with diluted preamplified cDNA. PCR primers were designed using the Roche software. All experiments were performed in biological duplicates or more. Expression values were normalized against the TATA-binding protein, and standard error of the mean (s.e.m.) were calculated and plotted using the Prism6 software (GraphPad). Primer sequences can be found in Supplementary Table 2. The expression data are expressed as mean±s.e.m. Statistical analysis were performed using Student's t-tests. Significance is indicated in the figures with the following convention: (*Po0.05; **Po0.01; ***Po0.005).
Shh doxycycline-inducible OP9 cell line (doxOP9-Shh). cDNA coding for Shh was cloned into pPBhCMV1-cHA-IRESVenuspA vector, a doxycycline-inducible bicistronic expression vector-gift from H. Niwa 70 -allowing both Shh and Venus to be expressed upon induction. OP9 cells (n ¼ 100,000) were electroporated with this plasmid using NEON transfecting system (Invitrogen). The next day, the electroporated cells were cultured in the presence of 1 mg ml À 1 doxycycline (Clontech). After 24 h of treatment, doxOP9-Shh cells expressing Venus were sorted using FACSAriaII and plated in culture without doxycycline. After 1 week in culture, a second sort was performed to remove any cells constitutively expressing Venus. Before co-aggregation with AGM cells, doxOP9-Shh cells were cultured for 24 h in regular growth medium supplemented with 1 mg ml À 1 doxycycline. Co-aggregates were cultured in the presence of doxycycline for 5 days. OP9 cells were grown at 37°C in a humidified atmosphere (5% CO 2 ). The growth medium was IMDM (Invitrogen) supplemented with 20% fetal calf serum and penicillin/ streptomycin (50 U ml À 1 ).