Hemogenic endothelium generates mesoangioblasts that contribute to several mesodermal lineages in vivo

The vascular and the hematopoietic systems are deeply entwined throughout embryonic development. The first endothelial cells (ECs) in the gastrulating embryo originate from the lateral and posterior mesoderm. Later, they migrate towards the yolk sac (YS), where they differentiate into ECs and hematopoietic cells (HCs) of the blood islands (Medvinsky et al., 2011). This close anatomical and temporal relationship between HCs and ECs has suggested a shared common mesodermal ancestor: the hemangioblast. This progenitor was initially identified in the embryonic stem cell (ESC) model (Yamashita et al., 2000). Other studies associate the first hematopoietic stem cells (HSCs) with phenotypically differentiated ECs endowed with hematopoietic potential: the hemogenic endothelium. Indeed, fate-mapping and in vivo time-lapse imaging in mouse and zebrafish revealed that HCs originate from VE-Cadherin⁺ (VE-Cad, Cdh5) ECs in the dorsal aorta (Bertrand et al., 2010; Boisset et al., 2010; Kissa and Herbomel, 2010; Zovein et al., 2008) and insights have also been given on the different potential of hemogenic endothelial progenitors (Chen et al., 2011). By linking these two hypotheses, it has been demonstrated in the ESC model that the hemangioblast generates HCs through the formation of a hemogenic endothelium intermediate. This cell population is found in gastrulating mouse embryos and generates HCs on further culture (Lancrin et al., 2009).

The embryonic dorsal aorta is the source of another population of progenitor cells, named mesoangioblasts (MABs) (Minasi et al., 2002). These culture-defined cells express hemangioblastic, hematopoietic, endothelial and mesodermal markers, and possess self-renewal capacity and mesodermal differentiation potency in vitro and in vivo (Brunelli et al., 2004; Minasi et al., 2002); in vitro only, they display hematopoietic potential. MABs have been successfully used in cell transplantation protocols that led to a significant recovery of the structure and function of skeletal muscle in dystrophic animals (Brunelli et al., 2007; Sampaolesi et al., 2003). Although MAB myogenic potency is certainly useful for cell transplantation protocols, no insights have been obtained on their origin, role during normal development and tissue remodeling, or their relationship with other postnatal vessel-associated cells with therapeutic potential (Dellavalle et al., 2011; Mitchell et al., 2010).

To follow the fate of VE-Cad expressing ECs in the early and late embryo, we performed a genetic lineage-tracing analysis, combining pulses of induction of Cre recombinase with a time course evaluation of the reporter expression. We show that extra-embryonic VE-Cad⁺ endothelium generates the first wave of HCs that colonize the embryo mesenchyme. Moreover, we identify hemogenic endothelium as the source of progenitor cells that resemble embryonic MABs and physiologically contribute to several mesodermal lineages in the embryo, including skeletal muscle.

To genetically label endothelial-derived cell populations, we used the Cdh5-CreER^T2 transgenic mouse (Wang et al., 2010), expressing a tamoxifen (TAM)-inducible form of Cre recombinase (CRE-ER^T2) under the control of VE-Cad regulatory sequences. All experiments were performed in Cdh5-CreER^T2;R26R-EYFP (Srinivas et al., 2001) double transgenics.

We induced Cre recombination with one single intraperitoneal (I/P) TAM injection in the pregnant mother at E8.5 (Fig. 1A) to avoid labeling the YS mesoderm, which transiently expresses VE-Cad at E7.5 (Drake and Fleming, 2000). FACS analysis of cells dissociated from E9.5 embryos showed a higher percentage of EYFP⁺ cells in the YS, followed by placenta and embryo proper (Fig. 1B). The efficiency of Cre recombination was first measured by FACS as the percentage of EYFP⁺ cells in the CD31⁺ population, and was found to be 65.1±4.7% in the embryo proper 24 h after Cre induction (Fig. 1C). Immunofluorescence (IF) analysis of E9.5 embryos revealed co-localization of EYFP with endothelial-specific markers (VE-Cad and CD31/Pecam1) (Fig. 1D). Cre recombination efficiency (estimated as the number of EYFP⁺ cells on total CD31⁺ or VE-Cadherin⁺ cells) was approximately 80% (Fig. 1D). At this stage, EYFP⁺CD31⁻ or EYFP⁺VE-Cad⁻ cells were undetectable in the embryo proper. As a control of Cre recombination leakage, we performed IF and FACS analysis on embryos without TAM induction. At all stages analyzed (E10.5-E15.5), we occasionally detected very rare EYFP⁺ cells (in one-quarter of analyzed samples, 0.01% of the total cells in the embryo). These were invariably CD31⁺ and/or VE-Cadherin⁺ (supplementary material Fig. S1A-C). This minimal amount of Cre leakage was accounted for; nevertheless, it did not affect the conclusions of our study.

Strict tissue specificity is a crucial issue for all lineage-tracing approaches. To verify this, TAM injection was performed at E8.5, and EYFP⁺ and EYFP⁻ cells were separately sorted from the embryo proper and the YS at E9.5 (gating as in Fig. 1B). qRT-PCR analysis showed that Pax3 (paraxial mesoderm, dermamyotome) and paraxis (paraxial mesoderm) mRNAs were confined in the EYFP⁻ population (Fig. 1E). VE-Cad and CD31-expressing cells were found almost entirely in the EYFP⁺ fraction in both embryo proper and YS, further confirming the recombination efficiency. Runx1 (hemogenic ECs and HCs) was mainly expressed in the YS, and also in EYFP⁻ cells, as were CD41 (Itga2b) and CD45 (Ptprc) (other markers of HCs). Sox7, a gene found to regulate VE-Cad expression in the hemogenic endothelium (Costa et al., 2012) was exclusively expressed in EYFP⁺ cells in the embryo, while it was detected in both EYFP⁺ and EYFP⁻ cells in the YS. These results confirmed that induction of Cre recombination at E8.5 resulted in labeling of cells that express VE-Cad (eVE-Cad⁺ for embryonic VE-Cadherin-expressing cells), have mainly an endothelial phenotype in the embryo and also a hematopoietic phenotype in the YS, while completely lacking paraxial mesoderm characteristics.

The presence of abundant hematopoietic and hemogenic endothelium specific transcripts 24 h after induction in eVE-Cad⁺-derived cells prompted us to investigate their hematopoietic nature and fate. The endothelium in the embryo proper contributes to definitive hematopoiesis (Boisset et al., 2010; Zovein et al., 2008), starting at E10-E10.5 in the dorsal aorta, with rare clusters appearing in the umbilical and vitelline arteries at E9.5 (Yokomizo et al., 2011). Indeed, at E10.5-E11.5, we detected EYFP⁺ intra-aortic clusters originating from EYFP⁺CD31⁺ hemogenic endothelium (supplementary material Fig. S2). All EYFP⁺ intra-aortic cluster cells expressed Kit. Most clusters were CD31⁺ and also expressed CD41 or CD45 (supplementary material Fig. S2).

Conversely, at E9.5 only rare EYFP⁺ cells in the embryo proper expressed hematopoietic markers (Fig. 2A,D,E), whereas EYFP⁺CD45⁺ and EYFP⁺CD41⁺ cells were found in the YS (Fig. 2B,D-E) and less abundantly in the placenta (Fig. 2C-E). We detected EYFP⁺Kit⁺ cells in the YS as early as E9 (supplementary material Fig. S3i-ii). This implies that the YS endothelium, and to a lesser extent the placental endothelium, are already hemogenic at E8.5.

Further IF analysis in the placenta revealed the presence of EYFP⁺CD45⁺ and EYFP⁺CD41⁺ cells at E10.5-E12.5 (supplementary material Fig. S4A). FACS analysis showed that among CD45⁺ cells, EYFP⁺ cells accounted for a lower percentage (1-2.5%) compared with YS and embryo (20-30%) (supplementary material Fig. S4B). Considering that many placental CD45⁺ cells are of maternal origin, and thus EYFP⁻, these data suggest that a small, but still significant, number of hematopoietic cells in the placenta are of endothelial origin.

We then focused on the fate of EYFP⁺ cells in the embryo proper, from E10.5 (Fig. 3A). In addition to endothelial and intra-aortic hematopoietic clusters, EYFP⁺CD45⁺ cells could be seen on the abluminal side of vessels and in the surrounding mesenchyme (Fig. 3B-Ci). This was evident around the dorsal aorta (arrows in Fig. 3B-Ci; supplementary material Movie 1) and around vessels in the limb bud (arrow in supplementary material Fig. S5Ai). Some EYFP⁺CD41⁺ cells were also observed in the mesenchyme (supplementary material Fig. S5Aii). At E10.5, the number of EYFP⁺CD41⁺ and EYFP⁺CD45⁺ cells in the mesenchyme of the embryo proper was almost equivalent (Fig. 3D); some of them expressed Kit⁺ (supplementary material Fig. S3iii-iv). CD45⁺ cells in the mesenchyme were not natively labeled as they did not express endothelial markers (Fig. 3B-Ci; supplementary material Fig. S5Ai), which means labeling must have occurred at an earlier time. Moreover, at E10.5 these were the only EYFP⁺VE-Cad⁻ cells in the embryo. At E11.5 a higher number of EYFP⁺CD45⁺CD31⁻ cells were present in the mesenchyme, in particular in the subaortic mesenchyme (Fig. 3Cii,D; supplementary material Movie 2 and Fig. S5Aiii). EYFP⁺CD41⁺ cells and EYFP⁺Kit⁺ cells were also detected in the embryo proper, outside EYFP⁺CD31⁺ vessels (supplementary material Fig. S3v-vi and Fig. S5Aiv). At E12.5, EYFP⁺CD45⁺ cells in the mesenchyme were even more abundant and could be detected throughout the whole embryo (Fig. 3Ciii,D; supplementary material Fig. S5B). At this stage, EYFP⁺CD41⁺ cells were rare in the mesenchyme (Fig. 3D). These data indicate that eVE-Cad⁺ cells give rise to HCs that migrate into and through mesenchymal tissues.

Remarkably, at E10.5-E12.5 many isolated EYFP⁺CD45⁺ cells in the embryo mesenchyme also expressed α-SMA, which at these early stages is expressed in a range of cells of mesodermal origin (smooth, skeletal, cardiac muscle cells, pericytes) (Crawford et al., 2002) but not in HCs (Fig. 3Ei-iii,G). Primitive embryonic macrophages (MΦs) arise in the yolk sac as early as E7.5, before the appearance of definitive HSCs. Although they differ from adult MΦs, they are characterized by the unambiguous expression of markers such as F4/80 and CD11b (Bertrand et al., 2005; Ovchinnikov, 2008). A substantial percentage of EYFP⁺CD45⁺ cells in the mesenchyme were indeed F4/80⁺ (Fig. 3Fi-ii,H). Still, from E10.5 to E12.5 about 35% of EYFP⁺CD45⁺ cells in the mesenchyme were not macrophages (F4/80⁻) (Fig. 3Fii,H).

EYFP⁺CD45⁺α-SMA⁺ and EYFP⁺α-SMA⁺F4/80⁻ cells in the mesenchyme were detected in various anatomical locations, mostly associated perivascularly with large vessels (Fig. 3E,Fi) and in the developing limbs. These cells increased over time, their ratio was similar (Fig. 3G) and they were invariably CD41⁻ (supplementary material Fig. S5Aii,iv). At E12.5, about 52% of CD45⁺α-SMA⁺ F4/80⁻ cells were EYFP⁺ (supplementary material Fig. S5D; Table 1).

To further characterize eVE-Cad⁺-derived cells in the embryo, we performed several analyses on freshly isolated cells. In particular, we focused on three different EYFP⁺ cell subsets: CD45⁻CD11b⁻F4/80⁻, CD45⁺CD11b⁺/F4/80⁺ and CD45⁺CD11b⁻F4/80⁻ (Fig. 4A,B). Sorting CD11b⁺ and/or F4/80⁺ cells ensures a high likelihood of inclusion of all macrophages (Pucci et al., 2009; Schulz et al., 2012); therefore, we will from now on refer to CD45⁺CD11b⁺/F4/80⁺ cells as CD45⁺MΦ, and to CD45⁺CD11b⁻F4/80⁻ as CD45⁺non-MΦ.

We characterized the above-mentioned populations by FACS and qRT-PCR on sorted fractions. At E12.5, CD45⁺non-MΦ cells expressed high levels of markers for both HCs (Kit, CD41) and mesenchymal cells (CD73, CD71, α-SMA, PDGFR-β, endoglin/CD105) (Fig. 4C-E). CD45⁻ cells expressed high levels of endothelial markers such as CD31, VE-Cadherin, Flk1, Tie2 and CD105. CD45⁺MΦ were the only cells to express macrophage/myeloid-specific genes such as Cx3Cr1 and PU.1 (Spi1). The latter also expressed other mesenchymal markers, but to a lesser extent compared with the CD45⁺non-MΦ population, in particular at the protein level. Some of the CD45⁺MΦ-expressing mesenchymal markers could be MΦs that have phagocytized mesenchymal cells and transiently acquired or exposed their antigens (Russo et al., 2000). However, as CD45⁺MΦ also transcribe these genes, this is the first indication that embryonic MΦs have a very heterogeneous gene signature. These results also suggest that a percentage of EYFP⁺CD45⁺ cells, which appear to emerge from vessels in the mesenchyme, are distinct from MΦs.

Further analysis of the CD45⁺non-MΦ population (Fig. 4E) revealed that it is composed of four fractions: Kit⁺CD105⁻/CD71⁻ cells, which display a HC phenotype; Kit⁻CD105⁺/CD71⁺ cells, with a more-enhanced mesenchymal phenotype; Kit⁺CD105⁺/CD71⁺ cells that could be a transitional population; and Kit⁻CD105⁻/CD71⁻, which are possibly more differentiated CD45⁺ cells.

The presence of transitional populations co-expressing HC and mesenchymal markers is supported by the fact that 1 day later (E13.5; supplementary material Fig. S6) the CD45⁻ population expressed mesenchymal markers at higher levels and the CD45⁺ non-MΦ increased the expression of Kit, while still expressing mesenchymal markers at a higher level than at E12.5. This suggests that cells that co-expressed HC and mesenchymal genes are now downregulating some lineage-specific genes and transitioning to a more univocal phenotype. Accordingly, CD45⁺MΦ cells significantly downregulated mesenchymal genes at E13.5.

Two possible origins of mesenchymal EYFP⁺CD45⁺non-MΦ cells can be postulated. We may observe endothelial-to-hematopoietic transition of hemogenic ECs in a direction that, in the mouse, is opposite to that of classical hematopoietic clusters, as reported during vascular remodeling of the vitelline artery (Zovein et al., 2010) but not further confirmed (Yokomizo et al., 2011). Conversely, these cells may originate from other VE-Cad-expressing cells in the embryo or extra-embryonic tissue, that reach the embryo mesenchyme by extravasation, where some maintain a hematopoietic phenotype, while others start to co-express mesenchymal markers. To elucidate this phenomenon, we performed Cre activation at different time points. Analysis of EYFP⁺ cells was carried out at 12 h and 24 h after TAM injection to immediately follow the fate of labeled cells (Fig. 5A).

Labeling at E9.5 with analysis at E10-E10.5 revealed a lower number of EYFP⁺CD41⁺ cells in the YS (Fig. 5Bi-ii,F) compared with those detected when labeling at E8.5, whereas the number of EYFP⁺CD45⁺ was similar in both cases (Fig. 5Biii,F). Rare EYFP⁺CD41⁺ and EYFP⁺CD45⁺ cells were seen in the placenta at this time point (Fig. 5F; supplementary material Fig. S4C). Interestingly, EYFP⁺CD41⁺ and EYFP⁺CD45⁺ cells were hardly detected in either YS or placenta when injection was performed at E10.5 and analysis at E11-E11.5 (Fig. 5D,F; supplementary material Fig. S4C). This suggests that the hemogenic potency of extra-embryonic endothelium is restricted to a short time-window, starting at E8.5 or before, and progressively decreasing until E10.5, possibly ending shortly thereafter.

When induction was performed at E9.5 and E10.5, no EYFP⁺ CD41⁺ or EYFP⁺CD45⁺ were observed in the mesenchyme of the embryo 24 h later (Fig. 5C,E). These EYFP⁺ cells were seen in the embryo proper starting from E10.5 only when injection was performed at E8.5 (Fig. 3B,C; supplementary material Fig. S5), showing that at least 48 h are needed to allow EYFP⁺ labeled cells at E8.5 to be found in the embryonic mesenchyme. With our induction protocol, expression of reporter genes is completed by 24 h (Park et al., 2007; Zovein et al., 2008). If EYFP⁺CD45⁺ and EYFP⁺CD41⁺ cells in the embryonic mesenchyme originated from hemogenic endothelium in the embryo proper we would have been able to detect them at E10.5-E11.5 by labeling VE-Cad⁺ cells in the E9.5-E10.5 time-window, when embryonic endothelium starts to be hemogenic. This is not the case; thus, EYFP⁺CD45⁺ and EYFP⁺CD41⁺ cells in the embryonic mesenchyme likely originate from extra-embryonic endothelium (predominantly YS), and become labeled in the E8.5-E9.5 time-window, as YS hemogenic activity decreases later on, as described above. These results also imply that CD45⁺α-SMA⁺ cells in the mesenchyme have the same origin.

We have previously reported the isolation of multipotent vessel-associated stem cells in the embryo, MABs (Minasi et al., 2002). Our results suggest that MABs resemble the EYFP⁺CD45⁺α-SMA⁺F4/80⁻ cells observed in the embryo mesenchyme and may originate from an endothelial or hemogenic endothelial precursor, the EYFP⁺CD45⁺non-MΦ cells, which gives rise to both hematopoietic and mesenchymal lineages. To further investigate this hypothesis, we evaluated the contribution of eVE-Cad⁺-derived cells to other mesodermal lineages (Fig. 6A).

At E12.5-E15.5 in the smooth muscle layer of large vessels we detected EYFP⁺CD31⁻ cells co-expressing α-SMA⁺ or NG2⁺, markers of the pericyte-smooth muscle lineage at those stages (1.1-2.3% of the total α-SMA⁺ or NG2⁺ peri-endothelial smooth muscle cells) (Fig. 6B, Table 1). EYFP⁺NG2⁺ cells found earlier at E10.5 were mostly CD31⁺; at E11.5 they were mainly CD31⁻ and in a perivascular location (supplementary material Fig. S5C).

We next examined the contribution of eVE-Cad⁺-derived progenitors to the skeletal muscle lineage. At E9.5-E11.5, no EYFP⁺ cells were found to colocalize with markers of somitic populations, such as Pax3, MyoD and Myf5 (supplementary material Fig. S7A), suggesting that eVE-Cad⁺-derived progenitors do not give rise to myogenic cells within somites. Similar results were obtained when Cre recombinase was induced at later stages (supplementary material Fig. S7B). This was confirmed by qRT-PCR on sorted cells (supplementary material Fig. S7C). From E12.5 we observed EYFP⁺CD31⁻ mononucleated cells expressing skeletal myosin heavy chain (MyHC) (Fig. 6Cii-iv) or α-SMA (Fig. 6Ci), which at this stage is transiently expressed by skeletal myoblasts. In E12.5 embryos EYFP⁺CD31⁻MyHC⁺/α-SMA⁺ cells were still rare (Fig. 6Ci-ii,G). Importantly, at E12.5 and E13.5 we observed the presence of EYFP⁺ myoblasts expressing myogenin (Fig. 6D), indicating that eVE-Cad⁺-derived cells can undergo at least some of the steps of canonical skeletal myogenesis. EYFP⁺MyHC⁺ CD31⁻ cells or EYFP⁺Desmin⁺CD31⁻ were detected throughout the embryo until P0, the last time point of our analysis (Fig. 6C-F; supplementary material Fig. S8A and Movies 3, 4; Table 1). Fig. 6G summarizes the quantification of eVE-Cad⁺-derived myogenic cells during development. qRT-PCR analysis on sorted EYFP⁺ cells at different stages confirmed the expression of MyHC, desmin, MyoD, Pax3 and myogenin from E12.5 (Fig. 6H; supplementary material Fig. S8B). Interestingly, TAM injection at E9.5 resulted in a lower expression of all myogenic transcripts (Fig. 6H; supplementary material Fig. S8B). We concluded that some cells expressing VE-Cad at E8.5 and/or their progeny enter the embryonic myogenic lineage without going through somitic intermediates as part of their normal, unperturbed fate.

At E15.5 the heart displayed a high degree of EYFP labeling in the endocardium. However, no EYFP⁺ cardiomyocytes were detected (supplementary material Fig. S9A). Almost all fibroblast-like endocardial cushion cells were EYFP⁺CD31⁻ (inset of supplementary material Fig. S9A), confirming that endocardial cushions arise from a subset of ECs that undergo endothelial-to-mesenchymal transition (Liebner et al., 2004).

At E17.5 the dermal layer surrounding the developing fingers contained a high number of EYFP⁺ cells (supplementary material Fig. S9B). EYFP⁺CD31⁻Collagen I⁺ cells and EYFP⁺Collagen I⁺F4/80⁻ cells (supplementary material Fig. S9Biii,iv) were detected (3.7% of total Collagen I⁺ cells, Table 1), indicating that eVE-Cad⁺-derived cells, distinct from embryonic MΦs, had entered the dermal lineage. As expected, and partially described before by FACS (Fig. 4B), eVE-Cad⁺ derived cells colonized fetal secondary hematopoietic organs (fetal liver, thymus and spleen) (supplementary material Fig. S9C). Taken together, these results confirm that eVE-Cad⁺-derived cells contribute to several mesodermal lineages with variable, but significant, frequency.

To further verify whether eVE-Cad⁺-derived cells resembled embryonic MABs after isolation and culture, we prepared aorta explant cultures from E9.5-E11.5 embryos 24 h after TAM induction, following the MAB isolation protocol (Fig. 7A; supplementary material Fig. S10A) (Tonlorenzi et al., 2007). After 1 week we observed the emergence of round, weakly adhering refractile cells, distinct from the populations of fibroblasts and floating HCs, and displaying MAB morphology. The majority of these cells were EYFP⁺ and CD45⁺ (Fig. 7B; supplementary material Fig. S10B). These data suggest that MABs originate from hemogenic endothelium, and, at least at early stages of ex vivo culture, display a hematopoietic phenotype.

We FACS isolated EYFP⁺ cells from E12.5 embryos, a stage when CD45⁺ cells in the mesenchyme are abundant (Fig. 7C). Isolated EYFP⁺ cells were cultured in conditions suited for MABs and grew as a mixed population of spindle-shaped cells and small round-shaped cells (Fig. 7D).

eVE-Cad⁺-derived cells were exposed to several differentiating stimuli. When treated with VEGF, they formed endothelial-like networks (Fig. 7Ei). TGFβ increased the number of α-SMA⁺ cells with smooth muscle phenotype (80% versus 10% of untreated cells) (Fig. 7Eii). EYFP⁺ cells expressed alkaline phosphatase (AP, a marker of pericytes and initial osteblastic differentiation) when treated with BMP2 (60-70% of AP⁺ cells versus 5% of untreated) (Fig. 7Eiii). Culturing EYFP⁺ cells in adipogenic medium induced the appearance of adipocytes containing Oil Red O-positive fat droplets (Fig. 7Eiv).

When co-cultured with C2C12 murine myoblasts (after infection with a lentivirus directing expression of nLacZ), EYFP⁺ cells fused into multi-nucleated MyHC⁺ myotubes (Fig. 7Ev,vi). Single mononucleated MyHC⁺β-Gal⁺ cells were also detectable (arrow in Fig. 7Ev), indicating that eVE-Cad⁺-derived cells were capable of direct myogenic differentiation. nLacZ-eVE-Cad⁺-derived cells were injected into cardiotoxin-injured tibialis anterior (TA) muscles of wild-type mice. Donor-derived X-Gal⁺ nuclei were found inside myofibers (Fig. 7Fi,ii), preferentially localized in a central position, typical of myonuclei inside regenerating myofibers (white arrow in Fig. 7Fii) but also peripherally and underneath the basal lamina (purple arrow in Fig. 7Fii). When injected into TA of immunosuppressed α-sarcoglycan (α-SG) KO mice, a model of limb-girdle muscular dystrophy, we observed the appearance of clusters of α-SG⁺-regenerating myofibers (Fig. 7G,H). We conclude that eVE-Cad⁺-derived cells are able to contribute to muscle regeneration in vivo.

The previous in vivo and ex vivo analyses showed that the EYFP⁺ population in the embryo became more heterogeneous with time. To evaluate this heterogeneity in vitro and to discriminate which population carried MAB characteristics, we sorted the CD45⁺MΦ (CD45⁺F4/80⁺CD11b⁺) and the CD45⁺nonMΦ (CD45⁺F4/80⁻CD11b⁻) subsets (Fig. 8A,B).

In MAB culture conditions, only the CD45⁺non-MΦ fraction gave rise to cells closely resembling embryonic MABs and co-expressing CD45 and α-SMA (Fig. 8C). CD45⁺MΦ and CD45⁻ cells grew poorly and the latter were generally bigger and heterogeneous in size and shape.

To identify the progeny of sorted cells at the single cell level, we performed a colony-forming cell (CFC) assay. Colonies started to form after 4 days, displaying different morphology (Fig. 8Div-vi). Only CD45⁺ non-MΦ cells gave rise to mesenchymal-like colonies with high frequency (Fig. 8D, graph).

Single colony qRT-PCR (Fig. 8E) showed that mesenchymal markers such as α-SMA, PDGFRβ and CD90, CD105, CD73, CD71, NG2 were exclusively or significantly more expressed by CD45⁺ non-MΦ mesenchymal colonies. Hematopoietic CD45⁺ non-MΦ colonies, likely containing HCs, specifically expressed Tie2 and Kit. Hematopoietic CD45⁺MΦ colonies, containing embryonic MΦs, expressed Ang1, but not Tie2, and did not express mesenchymal markers (Fig. 8E). The presence of CD45⁺ non-MΦ-derived mixed hematopoietic and mesenchymal colonies is in agreement with the observation of a transitional population co-expressing HC and mesenchymal markers (Fig. 4E).

Taken together, these data suggest that cells bearing characteristics of cultured embryonic MABs originate from eVE-Cad⁺-derived CD45⁺ cells distinct from MΦs and can give rise to both hematopoietic cells and mesenchymal progenitor cells.

Endothelium has been shown to be a crossroad of several cell lineages during development. Recent work has shed light on the existence of hemogenic endothelium (Boisset et al., 2010; Kissa and Herbomel, 2010; Zovein et al., 2008), a specialized embryonic endothelial population that gives rise to the precursors of HSCs and therefore virtually to all the adult blood. Here, we show that at early developmental stages a population of progenitors with mesodermal potency, including HCs, originates from VE-Cad⁺ ECs in the YS and in the placenta, in a very narrow temporal window (starting at E8.5 or before, progressively decreasing until E10.5 or shortly thereafter) (Fig. 9). Importantly, these extra-embryonic ECs are the only ones at this stage to give rise to HCs that are later found within the embryo mesenchyme. By activating Cre at different time points, we confirm that hemogenic endothelium in the embryo proper does not display this feature at early stages. Without a live imaging approach of whole intact mouse embryos, which is not currently possible to our knowledge, we can only speculate that they reach the mesenchyme through the circulation.

Clear evidence of the emergence of HSCs in the YS has not been provided (Wang and Wagers, 2011). E9-10 YS-derived cells can provide long-term contribution to adult hematopoiesis (Yoder et al., 1997), and by E11.5 YS contains true multilineage dHCs (Kumaravelu et al., 2002), thought to derive from hemangioblasts that migrate from the primitive streak (Huber et al., 2004). Yet, it has been suggested that from E8.25, CD41⁺ definitive hematopoietic progenitors originate in the YS from cells bearing endothelial features (Li et al., 2005; Wang and Wagers, 2011).

Likewise, the placenta is a hematopoietic organ that can generate HSCs de novo and support their expansion without promoting differentiation, and it contains a large population of HSCs that arise slightly later than in the YS, concomitant with their appearance in the AGM region (Dzierzak and Robin, 2010; Rhodes et al., 2008). Several pieces of evidence suggest that some placental hematopoietic stem and progenitor cells are probably generated in situ by hemogenic endothelial cells (Zovein et al., 2008). We confirm the hematopoietic potency of extra-embryonic hemogenic endothelium and provide the first evidence that extra-embryonic and embryonic hemogenic endothelia have different timings of activity, and possibly distinct biological characteristics.

The YS is also the source of embryonic MΦs, which first appear around E7.5 and infiltrate the embryo via the vasculature before E10.5, where they play important roles in tissue remodeling and organ development (Bertrand et al., 2005; Ovchinnikov, 2008; Schulz et al., 2012). We suggest that at least a subset of F4/80⁺/CD11b⁺ embryonic MΦs are generated from hemogenic endothelium. We also show the presence of another population, CD45⁺non-MΦ eVE-Cad⁺-derived cells in the mesenchyme that express α-SMA (Crawford et al., 2002) and other mesenchymal markers, such as CD71 (Tfrc), CD73 (Nt5e), CD105 (Eng) and PDGFRβ. Interestingly, CD105 expression has been correlated with YS bipotent progenitors that give rise to hematopoietic and endothelial progeny (Borges et al., 2013). In our case, CD45⁻ cells express higher levels of CD105, whereas HCs are associated with a lower expression.

Several studies hinted at the possibility of the existence of a common precursor that may give rise to both endothelium and blood cells or pericytes/mesenchymal cells (DeRuiter et al., 1997; Yamashita et al., 2000). The embryonic dorsal aorta contains MABs, culture-defined progenitors that express hemo-angioblastic markers and are able, in vitro and in transplantation assays, to differentiate into most mesodermal tissues, including skeletal myofibers (Minasi et al., 2002). This is in agreement with the identification of mesenchymal stem/progenitor cells in the major hematopoietic sites during mouse development (Mendes, 2005). The distribution/tissue contribution of CD45⁺non-MΦs eVE-Cad⁺-derived cells overlaps with the one observed with transplantation of MAB cell lines, although we did not analyze clonal populations. Of importance, VE-Cad⁺-derived cells express skeletal myogenic markers at RNA and protein levels. Although in many cases the frequency of labeling in mesodermal cells (smooth and skeletal muscle, dermis) was low, it was still significant and way above (minimum 10-fold) the level of any leakage observed at any stage, which only occurred in CD31⁺/VE-Cadherin⁺ cells (Table 1). In addition, we may underestimate the extent of the actual contribution of eVE-Cad⁺ cells to the various lineages, because of the incomplete efficiency of Cre recombination.

Experiments on aorta explants and selected sorted populations showed that, at midgestation, eVE-Cad⁺-derived cells give rise to endothelium, blood and mesenchymal cells in the embryo, and that a fraction of these cells maintained hematopoietic (CD45 and Kit) and mesenchymal markers. Furthermore, only the non-MΦ HCs subset of eVE-Cad⁺-derived cells was able to give rise to cells closely resembling embryonic MABs and to generate with high efficiency colonies expressing mesenchymal genes. These data correlate with the in vivo presence of a transitional population of CD45⁺non-MΦ eVE-Cad⁺-derived cells co-expressing HC and mesenchymal markers that, during development, specify to either fate. These data therefore suggest that the CD45⁺ non-MΦ eVE-Cad⁺-derived cells bearing mesenchymal features are the in vivo counterparts of MABs, and indicate that the contribution of CD45⁺MΦ eVE-Cad⁺-derived cells to multilineage differentiation in vivo and in vitro, by direct differentiation or fusion, is marginal. This is in agreement with reports showing that CD11b⁺ cells are not able to integrate into regenerating muscle fibers (Doyonnas et al., 2004; Ojima et al., 2004), whereas some myelomonocytic progenitors Kit⁺ cells are. This is suggestive of a relationship between myelomonocytic cells and pericytes, MABs and other mesodermal precursors that will be extremely interesting to investigate.

In conclusion, we have demonstrated that extra-embryonic hemogenic endothelium generates, in a narrow developmental time window, the first wave of CD41⁺ and CD45⁺ HCs that disseminate the mesenchyme in the embryo proper. Most importantly, we suggest for the first time the developmental origin of embryonic MABs as a small subpopoulation of hemogenic endothelium-derived cells. Moreover, our data support the idea that the in vitro and in vivo myogenic and multi-lineage differentiation capabilities of MABs are not a cell culture or transplantation artefact, but rather an expression of a differentiation potential that also exists during normal development.

Details of transgenic mice used in this work have been published previously: Cdh5-CREER^T2 (Wang et al., 2010) and R26R-EYFP (Srinivas et al., 2001). All experiments conformed to Italian law and were performed under internal regulations for animal care and handling (IACUC 355 and 489).

Cre activity was induced by one single intra-peritoneal injection of pregnant females with 2 mg/25 g body weight of tamoxifen (TAM) (Sigma; 10 mg/ml in corn oil). Embryos were collected after natural overnight matings. Staging of E9-E11.5 embryos was performed by counting somite pairs (E9, 18-25 sp; E9.5, 26-30 sp; E10, 31-35 sp; E10.5, 36-40 sp; E11, 41-45 sp; E11.5, 46-50 sp). For E12.5 and later stages, fertilization was considered to take place at 6 a.m.

Immunofluorescence on frozen section of embryos and muscles was carried out as described previously (Brunelli et al., 2007). Details of the antibodies are provided in supplementary material Table S1. At least six embryos were analyzed for each condition (stage and TAM induction). Histological quantifications were carried out by counting at least 20 fields (20× and/or 40×) for each data point.

EYFP⁺ cells sorted from embryos or picked from single colonies were processed using RNeasy Micro kit (Qiagen). Reverse transcription was performed using High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems). Quantitative real-time PCR (qRT-PCR) analysis was performed using Mx3000P (Stratagene) or 7900HT FAST (Applied Biosystems) Real-Time PCR detection systems. cDNA samples were amplified using GoTaq qPCR Master Mix and Hot Start Polymerase (Promega). Primer sequences are provided in supplementary material Table S3. Data points were calculated in triplicate. CT values greater than 35 were considered negative signals. Quantification was carried out using the comparative C_T method. 28S or cyclophilin A were used as an internal control.

eVE-Cad⁺-derived cells were isolated and processed as described previously (Biressi et al., 2007). For intracellular plus membrane FACS analysis, cells were first stained with membrane antibodies, then fixed and permeabilized with 0.1% saponin and 1% BSA in PBS.

Cell sorting was performed using the MoFLo system (Beckman Coulter). FACS analysis was carried out using LSR Fortessa or FacsCANTO (BD). At least 8-10 embryos, yolk sacs (YS), placentas or fetal livers were pooled for each experiment. Single stains and fluorescence minus one (FMO) controls were used to validate the position of the gates. Appropriate isotype controls were used. Dead cells were excluded by Hoechst 33258 (Invitrogen) or 7-aminoactinomycin D (Sigma) uptake. Data were analyzed using FlowJo software (TreeStar). Details on the antibodies are provided in supplementary material Table S2.

Dissection and culture of explants of embryonic vessels were performed as described previously (Tonlorenzi et al., 2007). Sorted EYFP⁺ cells were resuspended in DMEM with 20% FBS, bFGF (5 ng/ml) and β-mercaptoethanol (5 mg/ml) and spotted onto calf-skin collagen-coated dishes (sigma). Cells were kept in 3% oxygen.

Differentiation to smooth muscle, bone-like cells, adipocytes and endothelial networks was obtained as described previously (Tagliafico et al., 2004; Balconi et al., 2000). Infection of sorted EYFP⁺ cells with a lentiviral vector directing nLacZ expression (pRRL.sin.PPT.CMV.nLacZ.pre), in vitro myogenic differentiation and in vivo engraftment assay of at least 5×10⁵ cells nLacZ-EYFP⁺ cells in in TA muscles of C57/Bl6 or α-SGKO mice were performed as described previously (Dellavalle et al., 2007; Pessina et al., 2011).

Freshly sorted EYFP⁺ cells from E12.5 embryos were plated in a methylcellulose-based medium (M3434, StemCell Technologies) according to the manufacturer's instructions. Colonies were evaluated at day 2, 4, 7 and 12. Mesenchymal-like colonies included adherent, spindle-shaped cells.

Fluorescent and phase-contrast images were taken using the following microscopes: Nikon Eclipse E600; Nikon ACT-1 acquisition software; Leica AF6000; Leica AF600 acquisition software; Leica TCS SP2 Laser Scanning Confocal; Zeiss LSM 780 Inverted Confocal. Images were assembled in panels using Adobe Photoshop CS4 and Adobe Illustrator CS4.

Data were analyzed with Microsoft Excel 12.2.3 and GraphPad Prism 5. Values are expressed as mean±s.e.m. or mean±s.d. To assess statistical significance, unpaired two-tailed Student's t-tests were used (where not specified, *P<0.05; **P<0.01; ***P<0.001).

Part of this work was carried out in the ALEMBIC imaging center (San Raffaele Scientific Institute). We thank Elisabetta Dejana, Marella de Bruijn and Patrizia Rovere-Querini for comments and suggestions, and Mario Tirone for technical support.