Endothelial cells genetically selected from differentiating mouse embryonic stem cells incorporate at sites of neovascularization in vivo

The formation of new blood vessels, called angiogenesis, is involved in many physiological and pathological situations, such as embryogenesis, tissue remodeling and regeneration, wound healing and tumor neovascularization. In the adult, this process involves the recruitment and proliferation of either endothelial cells from pre-existing vessels or circulating endothelial progenitor cells (EPCs) originating from bone marrow(Asahara et al., 1997;Folkman, 1995).

In different situations, including advanced age, diabetes or hypercholesterolemia, such neovascularization is impaired, due to the reduction of vascular endothelial growth factor (VEGF) expression and to endothelial cell dysfunction (Couffinhal et al., 1999; Rivard et al.,1999a; Rivard et al.,1999b; Van Belle et al.,1997). Consequently, endothelial cell transplantation as a potential method for treatment of patients with vascular defects has become attractive and challenging in the past few years. This approach could also be used to improve tissue grafting after injury or reperfusion of ischemic tissues.

Several studies have described protocols to isolate and expand the population of circulating EPCs from blood(Asahara et al., 1997;Kalka et al., 2000;Shi et al., 1998). In different animal models of neovascularization, EPCs were shown to participate and enhance the formation of new blood vessels, and consequently increase tissue salvage (Asahara et al.,1999a; Asahara et al.,1999b; Kawamoto et al.,2001; Takahashi et al.,1999). However, EPCs represent a small proportion (0.1-0.5%) of circulating blood cells, and their expansion ex-vivo after harvest takes a considerable amount of time. This aspect is a concern for treating life-threatening conditions in patients. In this context, the development of a strategy for large scale purification of vascular endothelial cells (ECs) is of great interest.

Embryonic stem (ES) cells are derived from the inner cell mass of mouse blastocysts and have the potential of generating all embryonic cell lineages,including endothelial cells (Doetschman et al., 1985; Risau et al.,1988). ES-derived endothelial cells express most known endothelial cell markers, including CD31, VEGFR-2 (Flk-1), VE-cadherin, Tie-1 and Tie-2,and have the potential to form pseudo-vascular structures when differentiated into embryoid bodies (Vittet et al.,1996).

ES cells have a high proliferation rate and can easily be modified to express transgenes. In this respect, they could be used to genetically select ES-derived endothelial cells for grafting. Approaches employing genetic selection have already been used in ES cells in order to select for cardiomyocytes and neural-precursor cells(Klug et al., 1996;Li et al., 1998). Importantly,ES-derived cardiomyocytes were shown to form stable intracardiac grafts after injection into mice (Klug et al.,1996).

In this report, we used a similar strategy to yield large numbers of ECs from genetically modified ES cells. To this end, undifferentiated ES cells were transfected with a fusion gene consisting of the tie-1 promoter,specifically expressed in endothelial cells(Iljin et al., 1999;Korhonen et al., 1995),followed by sequences encoding the selection marker puromycin. Expression of the puromycin-resistance gene in ES-derived endothelial cells facilitated their purification during their in vitro differentiation. In addition, we observed that the release of puromycin selection was accompanied by the occurrence of cells positive for α-smooth muscle actin. We also show that purified endothelial cells participate in tumor angiogenesis in vivo.

Reporter gene constructs containing the tie-1 promoter were generated by subcloning the HindIII-ApaI fragment of the mouse tie-1 promoter (Iljin et al., 1999; Korhonen et al.,1995) into the same sites of either the pGL2 basic vector(luciferase reporter plasmid, Promega, Madison, WI) or the pEGFP vector(Clontech, Palo Alto, CA). The pEGFP vector contains the neomycin (G418)resistance gene.

The luciferase gene construct containing the VE-cadherin promoter was generated by subcloning the SalI(blunted)-XhoI fragment of the mouse VE-cadherin promoter [a gift from P. Huber, CEA,Grenoble, France (Gory et al.,1999)] into the SmaI-XhoI sites of the pGL2 basic vector.

A BamHI-KpnI fragment corresponding to the puromycin-N-acetyl-transferase (puro^r) cDNA was introduced into the same sites of the pGEM-3Zf(+/-) vector containing the HindIII-ApaI fragment of the mouse tie-1 promoter(Korhonen et al., 1995).

Rat antibodies against mouse CD31 (clone MEC 13.3), mouse CD34 (clone RAM34) and mouse VE-cadherin (clone 75) were from BD Pharmingen (Los Angeles,CA). Rat anti-VEGFR-2 was purchased from Clinisciences (France). Anti-α-smooth muscle actin was purchased from Sigma (Saint-Louis,MO).

Alexa Fluor-conjugated goat anti-mouse antibody, FITC and Alexa Fluor-conjugated streptavidin were from Molecular Probes (Eugene, OR). The biotin-conjugated donkey anti-rat antibody was from Jackson ImmunoResearch Laboratories (West Grove, PA). Biotin-conjugated anti-rabbit antibodies were from Amersham Pharmacia Biotech (Sweden). The FITC-conjugated rabbit anti-goat antibody was from Dako (Denmark).

OLA 129 ES cells, a gift from A. Smith (Centre for Genome Research,Edinburgh, UK) were maintained in Dulbecco's modified Eagle's medium (DMEM)with Glutamax-1 and NaPyruvate (Gibco BRL, Germany) containing 10% fetal calf serum (FCS; Dutscher, France), 50 U/ml penicillin, 50 μg/ml streptomycin,0.1 mM β-mercaptoethanol and non essential amino acids (all reagents were from Gibco BRL). This medium will be referred to as complete DMEM. The cells were kept undifferentiated by the addition of recombinant leukemia inhibitory factor (LIF) either purchased from Sigma or produced from COS cells(Smith, 1991). ES cells(5×10⁶ cells) were transfected by electroporation with a gene pulser (Eurogentec, Belgium) set at 500 V and 40 μF, in a total volume of 500 μl of PBS.

The established Chinese hamster lung fibroblast line CCL39 (American Type Culture Collection) and PS120 MEK S222D(Brunet et al., 1994) were maintained in DMEM supplemented with 7.5% FCS, 50 U/ml penicillin and 50μg/ml streptomycin. Bovine aortic endothelial cell (BAEC) were a generous gift of H. Drexler (Max-Planck-Institut, Bad Nauheim, Germany). They were maintained in low glucose DMEM supplemented with 5% FCS, 50 U/ml penicillin and 50 μg/ml streptomycin. HEK 293 cells were maintained in DMEM supplemented with 8% heat inactivated FCS, 50 U/ml penicillin and 50 μg/ml streptomycin. Cells were maintained at 37°C in a humidified atmosphere of 5% CO₂.

BAEC (1.5×10⁵ cells/well), CCL39 (10⁵cells/well) and HEK 293 (2×10⁵ cells/well) in 12-well dishes were transiently transfected by CaPO₄ precipitation with the indicated plasmids (600 ng/well of the reporter plasmid and 300 ng/well of the CMV β-galactosidase as a control for transfection efficiency). Two days later, cells were washed with cold PBS, and luciferase assays were performed according to the Promega protocol and application guide. Briefly, cells were lysed in lysis buffer (25 mM Tris-phosphate, pH 7.8, 2 mM DTT, 2 mM 1,2-diaminocyclohexane-N,N,N′,N′-tetraacetic acid, 10% glycerol, and 1% Triton X-100) for 15 minutes at room temperature, and the lysate was cleared by centrifugation. Luciferase activity was quantified in a luminometer(EG&G Wallac, Turku, Finland) in a buffer containing 20 mM Tricine, 1.07 mM (MgCO₃)Mg(OH)₂, 5H₂O, 2.67 mM MgSO₄, 0.1 mM EDTA, 33.3 mM DTT, 270 mM coenzyme A, 470 mM luciferine, and 530 μM ATP. The protein concentrations were measured using the Biorad protein assay (Hercules, CA).

Two protocols were used, to induce growth and differentiation. In the first one, freshly EDTA/trypsin (Biowhitaker, Walkersville, MD) dissociated ES cells were aggregated into embryoid bodies (EBs) in hanging drops of complete DMEM lacking LIF. 20 μl drops of cell suspension (4×10⁴cells/ml) were placed on the inside of lids of bacteriological Petri dishes. The lids were then placed over PBS-filled Petri dishes and incubated at 37°C; this was designated day 0. After 3-4 days, the resulting EBs were transferred to gelatin-coated 24-well tissue culture plates. In some cases,the medium was supplemented with human rVEGF-165 either purchased from Sigma or produced in our laboratory from Pichia pastoris after purification on heparin binding affinity columns, using the pPICZαA vector(Scheidegger et al., 1999)kindly provided by K. Ballmer-Hofer (University of Zürich,Switzerland).

In the second differentiation protocol, freshly dissociated ES cells were plated (3000 cells/cm²) in a 2D plane, on gelatin-coated dishes in complete DMEM lacking LIF. This method was used for selection of ES-derived endothelial cells. The concentration of puromycin to be applied on tie-1-puro^r expressing clones was determined using the non-resistant tie-1-EGFP cells as control cells. As tie-1-EGFP cells were killed by 0.25 μg/ml antibiotic, 1 μg/ml puromycin was added to the medium of differentiated tie-1-puro^r cells during 4 days. When indicated, selected cells were cultured on mouse laminin-1 (Sigma, 20 μg/ml) or in the presence of 5 ng/ml TGF-β1 (TEBU) or 10 ng/ml PDGF-BB (TEBU).

For spheroid formation, purified endothelial cells were aggregated into hanging drops (20 μl drops, 10⁵ cells/ml) for 5 days, in complete DMEM lacking LIF and in the presence of 10 ng/ml VEGF.

PCR amplification on genomic DNA was carried out with the Taq PCR master mix (Qiagen, France) containing Taq DNA polymerase, PCR buffer(1.5 mM MgCl₂) and dNTPs (200 μM for each dNTPs). After 35 cycles, the products were analyzed on ethidium-bromide-stained 1% agarose gels.

For PCR detection of the tie-1-EGFP transgene, we used the following pair of primers: 5′-CCCAACCATCCCCAGATCTG-3′ and 5′-TCCTCGCCCTTGCTCACCAT-3′, that anneal to the 3′ region of the tie-1 promoter and the 5′ region of the EGFP cDNA,respectively. For PCR detection of the tie-1-puro^rtransgene, we used the same tie-1 promoter primer and the primer 5′-GCGACCCACACCTTGCCG-ATG-3′, which anneals to the 5′ region of the puro^r cDNA. The annealing temperature was set at 57°C for both PCR reactions.

Total RNAs from undifferentiated and from purified B9TP cells were isolated using the RNA-now procedure (Biogentex, France). First-strand cDNAs were generated using reverse transcriptase (Roche, France) and oligo dT (Gibco BRL)using the manufacturer's instructions. For PCR amplification, cDNAs were amplified using the Taq PCR master mix (Qiagen). The amplification parameters were 95°C for 45 seconds, 55°C for 45 seconds and 72°C for 30 seconds, for 35 cycles. To ensure that the RT-PCR assay was semiquantitative,the quantity of equivalent reverse-transcribed RNA chosen was in the linear range of amplification. Hypoxanthine phosphoribosyl-transferase (HPRT) was used as an internal standard. The sequences of the primers used are the following:

• VEGFR-2 5′-AGCTCTCCGTGGATCTGAAA-3′ and 5′-CCAAG-AACTCCATGCCCTTA-3′;

• VEGFR-1 5′-CGGAAGCTCTGATGATGTGA-3′ and 5′-TATCT-TCATGGAGGCCTTGG-3′;

• Tie-1 5′-CTCACTGCCCTCCTGACTGG-3′ and 5′-CGATGTA-CTTGGATATAGGC-3′;

• HPRT 5′-GCTGGTGAAAAGGACCTCC-3′ and 5′-CACAGGAC-TAGAACACCTGC-3′.

The incorporation of ES-derived endothelial cells into sites of neovascularization was analyzed by using a tumor transplantation model in athymic nude mice (Harlan, France). To this end, 10⁶ ES-derived endothelial cells were co-injected subcutaneously with 10⁶ PS120 MEK S222D cells into the left flank of male athymic nude mice (n=6). For controls, 10⁶ tumoral cells were injected into the right flank of the same animal, and 10⁶ ES-derived endothelial cells were injected into the flank of five other mice.

Mice were euthanized after macroscopic tumor identification and frozen sections (4 μm) were processed for hematoxylin and immunofluorescence staining (see below).

Differentiated ES cells, gelatin-bound embryoid bodies, spheroids and tumor sections were fixed in 4% paraformaldehyde for 20 minutes at room temperature,permeabilized with 0.2% Triton X-100 for 5 minutes and blocked in 2% BSA/PBS for 2 hours. Cells were then incubated with the appropriate antibody for 1 hour at room temperature. After three washes in PBS, cells were incubated with the appropriate fluorescent-conjugated secondary antibody or streptavidin conjugates. Preparations were mounted in PBS:glycerol (1:9) and viewed under a Leica microscope.

The purpose of this study was to purify endothelial cells from differentiating ES cells through the use of a puromycin-resistance gene driven by an endothelial-specific promoter. We first compared the strength of the tyrosine kinase receptor Tie-1 and the vascular endothelial-cadherin(VE-cadherin) promoters. Both promoters were previously shown to result in vascular endothelial-specific expression in mice(Gory et al., 1999;Iljin et al., 1999;Korhonen et al., 1995).

To this end, luciferase reporter vectors controlled by either the tie1 or the VE-cadherin promoter were transiently transfected in bovine aortic endothelial cells (BAEC) and in non-endothelial cells, namely the fibroblastic CCL39 and the epithelial HEK 293 cells. The constructs were co-transfected with a vector encoding the β-galactosidase gene to monitor transfection efficiency. Relative luciferase activities were normalized to the values obtained with the empty pGL2-basic vector.Fig. 1 shows that the tie-1 promoter was active in BAEC and, to a lesser extent, in CCL39 cells. The low expression observed in CCL39 cells could reflect the lack of specificity reached in transient expression. In contrast, no activity could be detected in HEK 293 cells. A similar pattern was observed for the VE-cadherin promoter, but its relative activity was always lower than that of the tie-1 promoter. Transient expression of an EGFP reporter gene controlled by the same promoters gave similar results regarding their relative strength (data not shown). Based upon these results, we decided to test the cell-type specificity of the tie-1 promoter in the embryoid body model.

When differentiated into embryoid bodies (EBs), ES cells reproduce the stages of embryonic development, including vasculogenesis and angiogenesis(Risau et al., 1988). We first characterized vascular-like structures in wild-type EBs treated or not with 10 ng/ml VEGF, 5 and 8 days after leukemia inhibitory factor (LIF) removal to induce differentiation. Three different endothelial markers were used: the adhesion molecule CD31, VEGFR-2 and VE-cadherin.

Five days after LIF removal, control EBs contained numerous CD31-positive(CD31⁺) cell clusters (Fig. 2A), which also stained for VEGFR-2 but not for VE-cadherin (not shown). These cell clusters, in which CD31 staining was concentrated at cell-cell contacts, probably represent colonies of endothelial progenitors. After 8 days of differentiation, stained cells started to reorganize into primitive cord-like structures, and CD31 redistributed from cell-cell contacts to the entire cell membrane. The addition of VEGF accelerated the formation of the vascular network, already apparent at day 5 and fully developed after 8 days of differentiation (Fig. 2A). Cells forming these networks also expressed VEGFR-2 and VE-cadherin (Fig. 2B),indicating that VEGF promoted endothelial cell maturation. The time course of marker expression, in which VEGFR-2 and CD31 are expressed before VE-cadherin,is in agreement with previous studies(Hirashima et al., 1999;Vittet et al., 1996).

To test the cell-type specificity of the tie-1 promoter in the EB model, undifferentiated ES cells were transfected with the tie-1-EGFP construct. G418-resistant clones were screened by PCR to verify transgene integration. Two independent clones, designated A1TG and E3TG, were differentiated into EBs in the presence of VEGF. EGFP protein expression was undetectable until day 6 or 7 of differentiation, whereas CD31 and VEGFR-2 appeared earlier, at day 4 (not shown). Again, this was in agreement with the work of Vittet et al., who showed that expression of CD31 and VEGFR-2 transcripts preceded that of tie-1(Vittet et al., 1996). As shown in Fig. 3A, extensive vascular-like structures were visualized by CD31 staining in 10-day-old EBs derived from the A1TG clone (Fig. 3A, top). Cells forming these structures also expressed EGFP. All EGFP-positive (EGFP⁺) cells were found in the vascular network,thus demonstrating the endothelial specificity of the tie-1 promoter in the EB model. CD31-positive (CD31⁺) cell clusters that were negative for EGFP could be detected in some areas. As CD31 is an earlier marker than Tie-1 in the process of endothelial cell maturation, these clusters are likely to represent endothelial progenitors. In contrast, in EBs derived from a selected clone that did not carry the tie-1-EGFP transgene, vascular-like structures expressed only CD31 and not EGFP. The punctate green staining apparent in these control EBs represent autofluorescent dead cells (Fig. 3A, bottom).

To further check the tissue specificity of the tie-1 promoter,cells from embryoid bodies were enzymatically dissociated and replated on coverslips for 1 day. All EGFP⁺ cells also expressed CD31 as shown in Fig. 3B and all autofluorescent dead cells were lost. Changes in the cell morphology that correlate with EGFP expression are most obvious in this figure. Whereas CD31⁺EGFP^- cells were rounded and tightly bound to each other, CD31⁺EGFP⁺ cells were more elongated and CD31 was found on the entire cell membrane, instead of being concentrated at cell-cell contacts.

As an alternative, we also tested another ES differentiation system that was initiated by plating cells in 2D culture without LIF and in the presence of VEGF. This model was previously shown to support endothelial cell differentiation (Hirashima et al.,1999; Nishikawa et al.,1998). Thus, A1TG cells were analyzed for the expression of CD31,CD34 (another marker of the endothelial lineage) and EGFP after 8 days of differentiation in the presence of VEGF. A population of cells expressing CD31, EGFP (Fig. 3C) and CD34(not shown) was identified. DAPI analysis shows the heterogeneity of the differentiated population, which includes cells that express neither CD31,EGFP (Fig. 3C) nor CD34 (not shown). Since all EGFP⁺ cells also expressed CD31, this experiment confirmed the endothelial specificity provided by the tie-1 promoter in the 2D differentiation system.

Stable clones carrying both tie-1-EGFP and tie-1-puro^r transgenes were obtained by co-transfecting the A1TG clone described above with the tie-1-puro^rtransgene and a hygromycin resistance gene. Hygromycin-resistant clones were screened by PCR for transgene integration(Fig. 4A). Note that positive clones also carry the tie-1-EGFP transgene.

Two independent clones, B9TP and C4TP, were induced to differentiate in EBs and in 2D culture in the presence of 10 ng/ml VEGF. We found that puromycin addition to EBs was not appropriate for proper endothelial cell selection. First, the antibiotic could not efficiently penetrate into the whole 3D structures and, second, endothelial cells became surrounded by dying cells,which affected their growth capacity. Prior enzymatic dissociation of EBs was not efficient, since it damaged the cells and gave rise to only small numbers of selected cells. For these reasons, 2D cultures were preferentially used for endothelial cell selection.

Fig. 4B shows the efficiency of selection on differentiated B9TP cells after a 4-day puromycin treatment (1μg/ml puromycin), started at day 7 of differentiation after LIF removal. Endothelial cells were visualized by both CD31 staining and EGFP expression. Nuclei were stained with DAPI to detect cells present in the field, including CD31-EGFP-cells. On the left panel, corresponding to 10-day-old differentiated B9TP cells without puromycin treatment, EGFP⁺CD31⁺ cells are encompassed in a monolayer of cells negative for both markers. After a 4-day puromycin-selection followed by 1 day without the antibiotic to allow cells to recover, 96-98% of cells were positive for CD31 and EGFP. Upon selection, EGFP⁺CD31⁺ cell spreading was less efficient and cells were more tightly associated(Fig. 4B). EGFP protein expression also decreased in some cells when selection was released, but all selected cells retained the CD31 marker, which was then found concentrated at cell-cell contacts. It is possible that environmental changes caused by the removal of all other cell types results in a partial dedifferentiation of endothelial cells towards an endothelial progenitor phenotype.

The C4TP clone gave results similar to B9TP, which clearly demonstrates the efficiency of ES-cell-derived endothelial cell purification through genetic selection using the tie-1 promoter. Subsequent analyses were performed on the B9TP clone.

Purified cells were further characterized for the expression of various endothelial markers. We did not visualize VE-cadherin and VEGFR-2 proteins in 2D cultures with conventional immufluorescence techniques. Reasoning that the detection of these antigens could be easier in pseudo-vascular structures than in selected colonies, we aggregated purified CD31⁺EGFP⁺cells into spheroids. After 3 days, spheroids were plated on gelatin, and stained for endothelial markers 2 days later.Fig. 5A shows that although most cells were CD31⁺, EGFP expression was restricted to certain cells organized into vascular networks. Such structures were also found to express VE-cadherin, VEGFR-2 and CD34. However, VEGFR-2 and CD34 were also expressed in cell clusters negative for EGFP. These clusters probably represent endothelial progenitors.

Expression of three other endothelial markers, VEGFR-1, Tie-1 and ICAM-2,was monitored by a semi-quantitative RT-PCR approach.Fig. 5B shows that VEGFR-2 and VEGFR-1 were expressed at basal levels in undifferentiated cells. However,their expression was strongly increased in purified endothelial cells. In contrast, Tie-1 and ICAM-2 mRNA could be detected only in selected cells.

Interestingly, selected cells contained a small percentage (2-4%) of CD31^-/EGFP^- cells when analyzed 24 hours after puromycin removal. These cells were highly spread and expressed α-smooth muscle actin (α-SMA; Fig. 6). Their proportion was strongly enhanced at high cell density and in spheroids. In the light of recent results (Yamashita et al., 2000), we formulated the hypothesis thatα-SMA⁺ cells may be derived from purified endothelial cells themselves or their progenitors. Thus, we tested the effects of TGF-β and laminin-1, two known inducers of SMC differentiation(Arciniegas et al., 1992;Hayashi et al., 1998;Hirschi et al., 1998;Morla and Mogford, 2000) on the purified population. Fig. 6A shows that providing 5 ng/ml of TGF-β or plating cells onto laminin-1 dramatically increased the number of α-SMA⁺cells after 2 days. Laminin-1 had the strongest effect, even when cells were cultured in the presence of 10 ng/ml of VEGF. PDGF, a potent growth factor for SMC, was less efficient than TGF-β or laminin-1, suggesting that the occurrence of α-SMA⁺ cells in the purified endothelial cell population was due to the differentiation of endothelial cells or their progenitors towards an SMC phenotype rather than to their proliferation. Finding cells that expressed both CD31 and α-SMA⁺(Fig. 6B) further supported this hypothesis.

Interestingly, and unlike previous results(Yamashita et al., 2000), we did not obtain a pure population of α-SMA⁺ cells, even after a treatment with either TGF-β1, laminin-1 or PDGF for several days. This might be due to the different stages of cell differentiation at the time of purification. Whereas Yamashita et al. selected VEGFR-2⁺progenitors by flow cytometry at day 4 of differentiation, we waited until day 7 before adding puromycin for 4 additional days. It is possible that our selection protocol allows endothelial cells or their progenitors to reach a more advanced stage of differentiation that could prevent some of them from entering the smooth muscle cell lineage.

To investigate the capacity of purified endothelial cells to participate in the process of neovascularization in vivo, we used a tumor transplantation model in athymic nude mice. For that purpose, we chose a clone of PS120 cells expressing an active mutated form of MEK-1 (MEK S222D) conferring a weak tumoral potential in nude mice (Brunet et al., 1994). Six mice were injected subcutaneously with a mixture of 10⁶ selected endothelial cells and 10⁶ PS120/MEK S222D cells on the left flank, and 10⁶ PS120/MEK S222D cells alone on the right flank. Other controls were provided by five other mice injected only with 10⁶ endothelial cells. Mice injected with 10⁶endothelial cells alone never developed tumors, even after 5 months. PS120 MEK/S222D cells produced small tumors, if any, visible 3-4 weeks after inoculation; their weight ranged from 0.3 to 70 mg, confirming their previously described low tumorigenic potential. In contrast, co-injection of selected endothelial cells and PS120 MEK S222D resulted in the formation of tumors visible 2 weeks after injection. Three weeks after injection, their weight was almost 10 times higher than control tumors (ranged from 75 to 610 mg). This experiment was performed three times and always gave similar results.

When stained with hematoxylin and eosin, sections of 3-week-old co-injected and control tumors were homogenous and revealed no overt difference (not shown). These tumors were next analyzed for the presence of EGFP⁺endothelial cells in the neovasculature(Fig. 7A). Immunofluorescence staining for CD31 showed that the extent of neovasculazisation was similar in both types of tumors. EGFP⁺ cells were found in microvessels at the periphery of tumors resulting from the co-injection of the two cell types. Surprisingly, there was no EGFP staining in the center of the tumors. Whether this means that injected cells are simply absent in this location or that the tie-1 promoter is no longer active in these particular cells is not known and will be discussed below.

We also did not detect any EGFP⁺ cells in tumors resulting from the injection of PS120/MEK S222D alone in the same mice, indicating that injected endothelial cells did not migrate to the other flank of the animals through the circulation.

As puromycin-selected cells could also give rise to α-SMA⁺cells in vitro, tumor sections were co-stained for α-SMA and CD31(Fig. 7B). First, we detected few, if any, α-SMA⁺ blood vessels in control tumors,suggesting that these vessels were not mature. In contrast, most blood vessels in co-injected tumors were lined by α-SMA⁺ cells. It is unclear whether these α-SMA⁺ cells were of host or ES cell origin. In addition, we detected, in the co-injected tumors only, areas rich in α-SMA⁺ cells, probably derived from the injected endothelial cells. The size difference between co-injected and control tumors might thus be due to the growth of these α-SMA⁺ cells. This hypothesis and alternatives will be discussed below.

Subcutaneously injected ES cells are known to give rise to teratomas, which are tumors containing various types of differentiated tissues. Even if we found that the expression of oct-4, a marker for undifferentiated ES cells(Yeom et al., 1996), was downregulated in purified cells (data not shown), we thought it was important to check that selected endothelial cells do not give rise to teratomas in vivo. Although we did not detect any tumor formation in mice injected with 10⁶ endothelial cells alone, we repeated the experiment using 10⁷ endothelial cells instead. This protocol did not result in teratoma formation but hemangiomas did develop 3 weeks after injection (not shown). Hemangiomas consisted of both blood-filled cavities and tumor masses,with cavities surrounded by several layers of endothelial cells positive for EGFP, CD31 and CD34, while the rest of the tumor mass contained mostly CD34⁺ cells, with occasional CD31 staining.

We show here that the tie-1 promoter drives endothelial-specific EGFP expression in the embryoid body model of vasculogenesis and angiogenesis. We also present a novel approach for genetically selecting endothelial cells from differentiating ES cells, based on the expression of a puromycin resistance gene controlled by the tie-1 promoter. Selected cells expressed several endothelial markers and were recruited at sites of neovascularisation in vivo in a tumor transplantation model.

Both tie-1 and VE-cadherin promoters were previously found to drive endothelial-specific expression in vivo(Gory et al., 1999;Korhonen et al., 1995). In the present study, we compared the activity of these two promoters in various cell lines. In agreement with previous reports(Gory et al., 1999;Iljin et al., 1999), both promoters displayed functional activity in endothelial cell lines, but they also exhibited a weaker activity in fibroblast cells in transient transfection experiments. Whereas previous studies have not directly compared their respective activity, our experiments demonstrate that the tie-1promoter allows a more potent induction of reporter genes in endothelial cells. Therefore, we chose the tie-1 promoter for the rest of the study.

In spite of its weak activity in fibroblast cell lines in transient experiments, the tie-1 promoter retained its cellular specificity when expressed in differentiating ES cells. Reporter EGFP expression could be detected from day 5 of differentiation in VEGF-treated EBs and from day 6 in 2D cultures. This occurred only 1 day later than the expression of CD31, an early marker for endothelial precursors and endothelial cells(Nishikawa et al., 1998;Vittet et al., 1996). In embryoid bodies, EGFP expression was restricted to cells organized into cord-like structures also expressing CD31. VEGF enhanced the number of both EGFP⁺ cells and vascular-like structures. However, EGFP was absent from clusters of CD31⁺VEGFR-2⁺ cells, which probably represent endothelial progenitors. This confirms previous reports showing that the VEGF receptor R2 is an early marker for endothelial cell precursors(Hirashima et al., 1999;Vittet et al., 1996;Yamaguchi et al., 1993),whereas VE-cadherin and tie-1 transcript expression start at a later stage of differentiation(Hirashima et al., 1999;Vittet et al., 1997;Vittet et al., 1996).

Interestingly, the onset of tie-1 promoter activity correlated with changes in cell morphology. Whereas CD31⁺EGFP^-cells were round and formed clumps, CD31⁺EGFP⁺ cells displayed elongated processes and reorganized into pseudo-vascular structures. It is not clear whether these morphological changes are induced by Tie-1 expression or whether they are simply a part of a more general maturation program, but it would be interesting to analyze the effects of Tie-1 expression on endothelial cell migration and cytoskeleton organization.

The availability of ES cell clones carrying an integrated tie-1-EGFP transgene is of great interest for several reasons. First,we have confirmed the endothelial specificity of the tie-1 promoter in the EB model. This finding is in agreement with recently published work(Gustafsson et al., 2001). Second, this model can be used to trace the activity of the tie-1promoter in single cells within EBs whereas, in most studies published so far, tie-1 expression was assayed only by RT-PCR in a mix of different cell types. Third, tie-1-EGFP clones constitute a good model for easily monitoring the effects of potential pro- or anti-angiogenic factors. Two examples are given by our work, in which VEGF (this study) and bFGF (S.M.,unpublished) strongly promoted the formation of EGFP⁺ vascular-like structures. Finally, these cells can be used to generate mice expressing EGFP in their vasculature. This may help to understand the molecular mechanisms regulating tie-1 promoter activity, in particular, physiological and pathological conditions.

We took advantage of the cellular specificity of the tie-1promoter to genetically select endothelial cells from differentiating ES cells, using puromycin as a selection marker. Such an approach has been previously used for selecting neuronal and cardiac cells from a population of differentiating ES cells (Klug et al.,1996; Li et al.,1998). Starting from a few undifferentiated ES cells, large numbers of cells expressing the endothelial markers CD31, CD34, VEGFR-1,VEGFR-2, Tie-1, VE-cadherin and ICAM-2 can be obtained using this method.

While this work was under way, an alternative method was proposed to isolate vascular progenitors by flow cytometry cell sorting based on VEGFR-2 expression (Yamashita et al.,2000). VEGFR-2⁺ cells were isolated at day 4 of differentiation and counter-selected for E-cadherin expression, which is a marker for undifferentiated ES cells. VEGFR-2 is an early marker for endothelial progenitors (Hirashima et al.,1999; Vittet et al.,1996; Yamaguchi et al.,1993) and we detected its expression by RT-PCR in undifferentiated ES cells. This was in agreement with the weak vegfr-2 promoter activity previously observed in undifferentiated ES cells by the same experimental method, using lacZ as a reporter gene(Hidaka et al., 1999). Although LIF removal is theoretically sufficient to induce the differentiation of most ES cells over a few days, we thought it might be advantageous to wait longer than 4 days to avoid contamination by undifferentiated ES cells. For this reason, we chose a promoter regulating the expression of a late marker of endothelial cell maturation, [i.e. tie-1(Hirashima et al., 1999;Vittet et al., 1996)]. Because EGFP expression driven by the tie-1 promoter was not detectable before day 5 or 6 in differentiating ES cells, we waited until day 7 to start selection with puromycin. After 4 days selection, almost all cells (96-98%)were CD31⁺EGFP⁺. The other cells expressed α-SMA,a marker for smooth muscle cells. When maintained in culture,α-SMA⁺ cells later expressed transcripts for two other smooth muscle markers, SM22 and calponin (S.M. and C.G., unpublished). Two known inducers of SMC differentiation, TGF-β1 and laminin-1, strongly increased the proportion of α-SMA⁺ in the cell population. The fact that some cells express both CD31 and α-SMA suggested the existence of either a common progenitor for endothelial and SMC or a transdifferentiation mechanism from endothelial towards SMC. This finding is reminiscent of previous works by other groups (Arciniegas et al., 1992; DeRuiter et al.,1997; Hirschi et al.,1998; Yamashita et al.,2000). A few days after puromycin removal, CD31⁺ cells selected with our method adopted a morphology more characteristic of endothelial progenitors than more mature endothelial cells(Bautch et al., 2000). We interpreted this phenomenon as being a possible dedifferentiation of endothelial cells towards a more immature phenotype. Whether a dedifferentiation process is required for further differentiation into smooth muscle cells or whether transdifferentiation may occur between both lineages needs to be explored in more depth. If the common origin for both lineages is confirmed, our differentiation system might serve as a new model for studying the effects of TGF-β1 and laminin-1 on SMC differentiation.

Yamashita et al. showed that FACS-selected VEGFR-2⁺ cells participated in the formation of the vascular network in the developing chick embryo, and that they differentiated into both endothelial and vascular smooth muscle cells in vivo (Yamashita et al.,2000). In the present study, we showed that puromycin-resistant cells could be incorporated into microvessels at sites of neovascularisation in a tumor transplantation model in athymic nude mice, as revealed by EGFP fluorescence. Surprisingly, there were no EGFP⁺ endothelial cells in the vessels found in the core of the tumors, rather they were incorporated into peripheric vessels. The simplest explanation for this phenomenon would be that EGFP⁺ cells are not integrated at this particular location for unknown reasons. Alternatively, the activity of the tie-1 promoter might be downregulated in microvessels found in the tumor center. Another member of the Tie family, Tie-2, was shown to be preferentially expressed in vessels at the periphery of certain tumors(Asahara et al., 1999a;Peters et al., 1998). Because Tie-1, like Tie-2, is involved in angiogenesis and blood vessel maturation, it is tempting to speculate that the two promoters might be similarly regulated.

We consistently found that the size of the PS120/MEK S222D tumors was larger when tumor cells were co-injected with purified endothelial cells. Providing exogenous endothelial cells may have promoted the vascularization and subsequent growth of the tumors, as suggested by the presence of EGFP⁺ cells in newly formed blood vessels. In this respect, it should be noted that, although vascular densities were comparable in control and co-injected tumors, overall angiogenesis was increased as the co-injected tumors were larger. However, there might be other causes for this difference in tumor size. First, we found numerous α-SMA⁺ cells in co-injected tumors. Although we do not have a marker for these cells (the tie-1 promoter is not active in α-SMA⁺ cells), it is possible that they are derived from the injected endothelial population, as they were absent in the control tumors. α-SMA⁺ cells were found both located around blood vessels and organized in cell aggregates. On the one hand, blood vessel lining by α-SMA⁺ cells, observed only in the co-injected tumors, may have an additional impact on angiogenesis and subsequent tumor growth. On the other hand, the size of SMC-containing aggregates could be increased by tumor-secreted cytokines, such as members of the TGF family. Alternatively, injected endothelial cells may themselves secrete growth factors, such as PDGF, triggering the survival and proliferation of PS120 MEK S222D cells at the initial stage following injection. Although it is likely that the tumor size difference results from a combination of all these effects, the fact remains that puromycin-selected endothelial cells can be incorporated in tumors at sites of neovascularization. Importantly, experiments performed in mice indicated that endothelial cells selected by our protocol did not produce teratomas over a period of several months when subcutaneously injected in large amounts,suggesting that the purified cells did not contain undifferentiated cells.

Further studies will be necessary to assess the ability of endothelial cells selected by our method to promote revascularization of wounded or ischemic tissues. Their potential ability to give rise to both endothelial and smooth muscle cells in vivo could be particularly advantageous if mature vessels are to be formed. If they prove to be competent, ES-cell-derived endothelial cells could provide an appealing alternative to adult stem cells in the context of pro-angiogenic therapy. As we have shown here, a large number of endothelial cells or progenitors can be obtained from few undifferentiated ES cells and they can also be easily genetically modified. Hence, a combination of stem cell and gene therapy could be of great benefit for treating degenerative diseases in which endothelial cells are involved.

We thank M. C. Brahimi-Horn, R. Tournaire-Binetruy, M. P. Simon, and E. van Obberghen-Schilling for helpful discussions and critical reading of the manuscript. We thank our colleagues for their generous gifts of cell lines and constructs. We gratefully acknowledge Eric Couchi and Frank Paput for mice husbandry and the Department of Anatomo-Pathology of the Centre Antoine Lacassagne for their help with histological staining. The CNRS, the University of Nice-Sophia Antipolis, the AFM/INSERM, the Foundation de France, the Association for the Research against Cancer (ARC) and the Ministère de la Recherche are gratefully acknowledged for their financial support.