Msx genes define a population of mural cell precursors required for head blood vessel maturation

The cardiovascular system is composed of the heart, arteries, arterioles, capillaries, veinules and veins. Two distinct layers compose the blood vessels: the inner endothelium and the external mural layer. Endothelial cell (EC) assembly takes place first. Within the mouse embryo, mesodermal progenitors give rise to the angioblast, an endothelial precursor. The angioblasts start to aggregate as early as embryonic day (E) 7.5, forming the ventral and dorsal aortas as well as the vitelline vascular tree. The expansion and specialization of this initial basic system is achieved by angiogenesis, involving EC sprouting, vessel branching and intussusception from existing blood vessels (Risau, 1997).

Vascular maturation then takes place and confers endothelial tube contractility and resistance (Jain, 2003). Its primary mediators are the mural cells that form a multi-layered vascular smooth muscle around arteries and veins, whereas in capillaries and veinules the mural layer is composed of sparse pericytes. These mural cells originate from different sources depending on their position within the body. Most vascular smooth muscle cells (VSMCs) of the head and aortic arch have a neural crest origin (Jiang et al., 2000; Etchevers et al., 2001; Korn et al., 2002) (for a review, see Majesky, 2007), whereas those of the trunk primarily derive from the somitic and lateral plate mesoderm (Esner et al., 2006; Majesky, 2007; Santoro et al., 2009; Wiegreffe et al., 2009). Vascular maturation further involves the development of an elastic lamina between endothelial cells and VSMCs and of an extracellular matrix (ECM) that embeds both layers (Jain, 2003).

Vascular maturation involves the proliferation, migration and differentiation of VSMCs. These are orchestrated by multiple signaling pathways, including PDGFβ/PDGFRβ (Hellstrom et al., 1999), TGFβ (Majack, 1987; Battegay et al., 1990; Nishishita and Lin, 2004), angiopoietin 1 (Angpt1)-Tie2 (Fukuhara et al., 2008; Saharinen et al., 2008) and BMP (El-Bizri et al., 2008; Spiekerkoetter et al., 2009; Bai et al., 2010).

In the mouse, the Msx gene family is composed of three homeodomain transcription factors. Msx1 and Msx2 play important and often overlapping roles in the development of craniofacial structures, neural tube and limb (Alappat et al., 2003; Bach et al., 2003; Lallemand et al., 2009). In humans, MSX1 mutations lead to cleft palate and lips, as well as to tooth agenesis (Vastardis et al., 1996; Kapadia et al., 2007). MSX2 has been associated with craniosynostosis (Wilkie, 1997). Msx3 is absent from the human genome (Finnerty et al., 2009) and in the mouse its expression is restricted to the dorsal aspect of the neural tube (see Ramos and Robert, 2005).

Recently, we have shown that in the adult mouse, both Msx1 and Msx2 are expressed in a subset of peripheral artery VSMCs. Furthermore, Msx1 is expressed in pericytes of capillaries and Msx2 in ECs of the dorsal aorta. In the embryo, both Msx1 and Msx2 are detected in some ECs of the aorta from E14.5 and Msx1 additionally in mural cells of the intersomitic arteries from E10.5 (Goupille et al., 2008). We have undertaken the present study to clarify the role of Msx genes in vascular development. Phenotypic analyses and gene expression assays demonstrate that VSMC coverage is reduced in the Msx1^−/−; Msx2^lox/−; Sm22α-Cre mutant (hereafter referred to as Sm22Cre Msx1/2), resulting in an increase in vessel diameter and impairment of endothelium maturation. By genetic lineage tracing, we show that the primary defect lies in a population of VSMC precursors. BMP signaling and its targets Mmp2 and Mmp9 are downregulated in Msx1; Msx2 mutant VSMCs, leading to impaired incorporation of Msx1-expressing vascular smooth muscle precursors into the mural layer.

We previously reported the generation of Msx1 null (Msx1^lacZ), Msx2 null (Msx2^GFP, Msx2^lacZ) and conditional (Msx2^lox) mutant alleles (Houzelstein et al., 1997; Lallemand et al., 2005; Bensoussan et al., 2008). The Msx1^lox conditional mutant allele (Fu et al., 2007) was a generous gift from Dr Robert Maxson (Los Angeles, CA, USA) and the Tie2-Cre transgenic mouse (Kisanuki et al., 2001) from Dr Masashi Yanagisawa (Dallas, TX, USA). The Sm22α-Cre (Zhang et al., 2006) and Rosa^mT/mG (Muzumdar et al., 2007) engineered mice were purchased from The Jackson Laboratory (Bar Harbor, ME, USA). The Msx1^CreERT2 mouse was generated by introducing the CreERT2 coding sequence (Feil et al., 1996) (a kind gift of Pierre Chambon, Illkirch, France) at the initiator ATG site of Msx1 by homologous recombination in ES cells. CreERT2 interrupts the Msx1 coding sequence, thus creating a null allele. After Southern blot selection, recombinant cells were injected into blastocysts and these were reimplanted using standard protocols. All mice were maintained on an NMRI outbred background. Genotyping primers are listed in Table S1 in the supplementary material. All studies were conducted using mutant embryos with littermates as controls.

Tamoxifen (Sigma) was dissolved in ethanol, emulsified in sunflower oil (Sigma) and then sonicated three times for 5 seconds each at a final concentration of 10 mg/ml. Tamoxifen was intraperitoneally injected either three times at 3.5 mg per injection or twice at 2.5 mg per injection per pregnant female (weight, 30 g). Successive injections were performed at 12-hour intervals.

Embryos were fixed overnight in 4% paraformaldehyde and non-specific epitopes blocked overnight in 0.1% sodium azide, 1 mM MgCl₂, 1% BSA, 10% goat serum, 0.5% Triton X-100 and 0.5% Tween 20. A 1-week incubation with rat anti-mouse CD31 (BD Pharmingen) antibody was performed followed by a 5-day incubation with a secondary anti-rat antibody (Alexa 546, Invitrogen). Acquisition and treatment of images using OPT was performed by Bioptonics MRC Technology (Edinburgh, UK) according to published protocols (Sharpe et al., 2002).

RNA was extracted using the RNeasy Mini Extraction Kit (Qiagen). RT-PCR was performed on a StepOnePlus machine (Applied Biosystems, Warrington, UK) using SYBR PCR Master Mix (Applied Biosciences). Primers are listed in Table S2 in the supplementary material. Gapdh was used as a reference. PCR efficiency was in the range 98-100% for all assays. PCR cycle parameters were: 10 minutes at 95°C initial incubation, followed by 15 seconds at 95°C and 1 minute at 60°C for 40 cycles. For each gene studied, multiple RNA samples were analyzed; n=5 for single Msx1 and single Msx2 mutants, n=4 for Sm22Cre Msx1/2 mutants, each in duplicate. Data are expressed as fold changes [2^{−(Δctmutant–Δctreference)}] using Msx1^+/−; Msx2^lox/+; Sm22α-Cre triple heterozygotes as a reference.

E11.5 embryos were fixed for 2 hours in 4% paraformaldehyde (Sigma) then immersed in 15% sucrose and O.C.T. Compound (Tissue-Tek) before being frozen in liquid nitrogen and cryostat sectioned at 20 μm. The Bmp4 probe was a gift from Dr B. Hogan (Durham, NC, USA). Automated in situ hybridizations were performed with an InsituPro VSi apparatus (Intavis Bioanalytical Instruments, Köln, Germany).

Immunohistochemistry was performed as described (Goupille et al., 2008) except that X-Gal staining was omitted. Primary antibodies are listed in Table 1. Secondary antibodies (Invitrogen) were: Alexa Fluor 488 goat anti-mouse and goat anti-rabbit, Alexa Fluor 568 goat anti-rabbit, Alexa Fluor 647 goat anti-mouse and Alexa Fluor 635 goat anti-rabbit at 1/300 and Alexa Fluor 488 streptavidin at 1/1000. A Zeiss Axioplan microscope equipped with an Apotome and with Axiovision software (Carl Zeiss, Jena, Germany) was used for vessel section surface measurements. All images were assembled in Photoshop and Illustrator (Adobe Systems, San Jose, CA, USA).

The E12.5 embryos used were either heterozygous or homozygous at the Sm22α-Cre; Rosa^mT/mG; Msx1^lacZ; Msx2^lacZ loci. They were genotyped before cell dissociation. Cells were dissociated mechanically with a 1-ml 26 gauge syringe and resuspended in Dulbecco's Modified Eagle's Medium (Invitrogen) containing 2% fetal bovine serum (Invitrogen). VSMCs were sorted on the basis of GFP expression. For ECs, the cell suspension was labeled with Phycoerythrin-conjugated anti-CD31 antibodies (BD Biosciences). Cells were then washed with PBS and fluorescence was quantified as relative fluorescence units on a MoFlo High Performance Cell Sorter (Beckman Culture, Krefeld, Germany). Values are reported as mean fluorescence units.

VSMCs were FACS sorted as described above and maintained in DMEM supplemented with 10% fetal bovine serum and streptomycin on collagen I (BD Biosciences)-coated plates. The medium was changed every 48 hours and cells were passaged a maximum of six times (i.e. up to 20 days) using accutase digestion (Gibco). Recombinant mouse Bmp4 (Roche) was added to Msx1^−/−; Msx2^−/− cells and these were incubated for a further 48 hours, then lysed and RNA extracted using the RNeasy Mini Extraction Kit (Qiagen). Msx1^+/−; Msx2^+/− cells were transfected with Msx1 and Msx2 small interfering RNA (siRNA; siGENOME SMARTpool, Dharmacon) and with siGLO Risc-Free control siRNA (Dharmacon) as a control for transfection efficiency. DharmaFECT transfection reagent (Dharmacon) was used according to the manufacturer's instructions. After 48 hours, cells were lysed and RNA extracted as described above.

For results from qRT-PCR, diameter measurements and cell number quantification, means ± s.e.m. were calculated. For qRT-PCR experiments, results from Msx mutant analysis were expressed as fold change relative to Msx double-heterozygous littermates or as absolute values. Cell counts and area measurements were performed using ImageJ version 1.43g (NIH). One-way ANOVA was used to compare independent experiments. Comparison between data groups was performed using a non-parametric Dunnett test. All statistical analyses were performed using GraphPad Prism version 5.0 for Apple (GraphPad software, San Diego, CA, USA).

We previously demonstrated that, in the adult mouse, Msx1 and Msx2 are expressed in the VSMCs of peripheral arteries (Goupille et al., 2008). In E11.5 embryos, Msx1^lacZ was detected in a few ECs of the carotid artery (CA) (Fig. 1A). Msx1^lacZ expression was also observed in mesenchymal cells outside the CA, some of which co-expressed Msx2^GFP (Fig. 1C). Msx2^GFP was mainly expressed in differentiated VSMCs of the CA (Fig. 1B). At E12.5, Msx1^lacZ was detected mainly in the endothelial layer of the CA and vertebral artery (VA) but not in veins (see Fig. S1 in the supplementary material). Expression of Msx1^lacZ was also observed in a few cells of the most external VSMC layer, and again in mesenchymal cells further away from the CA (see Fig. S1 in the supplementary material). Msx2^GFP expression was maintained in the VSMCs at this stage.

To inactivate both genes in blood vessels, we combined a null (Lallemand et al., 2005) and a floxed (Bensoussan et al., 2008) Msx2 allele and the Sm22α-Cre (Sm22α is also known as transgelin – Mouse Genome Informatics) transgene (Zhang et al., 2006), together with Msx1^lacZ null alleles. Using this strategy, Msx1 is inactivated in the two layers of the blood vessel, whereas Msx2 is inactivated only in the VSMCs. The Sm22α-Cre transgene that we used was chosen to inactivate Msx2^lox because of its early activation in mural cells (El-Bizri et al., 2008). The Rosa^mT/mG allele ubiquitously produces a membrane-bound Tomato red fluorescent protein, which is replaced by membrane-bound GFP after Cre-mediated recombination (Muzumdar et al., 2007). When associated with this allele, the Sm22α-Cre transgene drives expression of GFP in VSMCs at embryonic stages before they integrate into the CA and, consequently, before they express differentiation markers such as alpha-smooth muscle actin (α-Sma; Acta2 – Mouse Genome Informatics) (see Fig. S2 in the supplementary material). Sm22Cre Msx1/2 mutants die a few hours after birth due to Msx1 deficiency (Houzelstein et al., 1997), in contrast to constitutive double mutants that die at E14.5 (Lallemand et al., 2005).

By optical projection tomography (OPT), we first observed overbranching of the CA at E11.5 in the Sm22Cre Msx1/2 mutant (Fig. 1D-G; see Movies 1 and 2 in the supplementary material) but not in the Msx1 or Sm22Cre Msx2 single mutants (data not shown). The number of primary and secondary CA branches was increased by ~1.4-fold in the mutant (Fig. 1H). No branching defect was detected in capillaries by immunofluorescence on sections (data not shown). Furthermore, a 2-fold increase in arterial caliber was observed in the Sm22Cre Msx1/2 mutant. At E11.5, the CA section surface increased from ~1400 μm² in the control to ~2900 μm² in the mutant (Fig. 1I-L). The VA caliber was also increased (data not shown). No change was observed in single Msx1 or Sm22Cre Msx2 mutants. In contrast to the CA, no significant change was observed in dorsal aorta sections of the double mutant (~13,200 μm²) relative to the control (~12,800 μm²) (Fig. 1M-P).

Strikingly, quantification of α-SMA-positive cells in the CA revealed a halving in the number of VSMCs in the double mutant (Fig. 2A,D) as compared with the control (Fig. 2G). RNA was extracted from heads of E11.5 embryos (Fig. 2H) and the expression level of a set of genes measured by qRT-PCR. In the Sm22Cre Msx1/2 mutant, VSMC-specific transcripts were decreased in proportion to the reduced coverage of the vessels (see Fig. S3 in the supplementary material). The number of ECs was not reduced (Fig. 2G), and, accordingly, expression of Cd31 (Pecam1 – Mouse Genome Informatics) and jagged 1 (Jag1), two genes that are expressed early in the endothelium, was not affected by the mutation (Fig. 2B,E,I). Unexpectedly, the VE-cadherin (VE-cad; cadherin 5 – Mouse Genome Informatics) level appeared markedly decreased on sections (Fig. 2C,F). According to qRT-PCR (Fig. 2I), VE-cad transcripts were reduced by 26% in the Msx1 single mutant and by 57% in the Sm22Cre Msx1/2 double mutant. Similarly, von Willebrand factor (Vwf) transcripts were reduced by 25% and 39%, respectively. VE-cad and Vwf are among the genes that are expressed late in the endothelium (Dejana et al., 1989; Navarro et al., 1995), suggesting an impairment in maturation. The basal lamina, analyzed using collagen IV, laminin 1 and fibronectin antibodies, appeared normal in the Sm22Cre Msx1/2 mutant (see Fig. S4 in the supplementary material).

At birth, vascular anomalies were observed in the head of Sm22Cre Msx1/2 mutants. The superficial temporal artery showed more branching than in the control; aneurysms and hemorrhages were frequently observed (see Fig. S5 in the supplementary material). No major branching defect or hemorrhage was observed in either Msx1 or Sm22Cre Msx2^lox/− single mutants.

To identify the cell layer in which Msx1 is required, we took advantage of a conditional allele (Fu et al., 2007) to selectively inactivate Msx1 in the endothelium and in the VSMC lineages (see Fig. S6 in the supplementary material). This showed that Sm22Cre-driven inactivation of Msx1 and Msx2 results in the same phenotype as the Sm22Cre Msx1/2 mutation, i.e. an increase in the CA section surface and a reduction in VSMC coverage (see Fig. S6A,B,D in the supplementary material). By contrast, Tie2-Cre-driven Msx1 and Msx2 inactivation did not affect the CA diameter or VSMC coverage (see Fig. S6B,C,D in the supplementary material). Therefore, Msx1 is required in cells of the smooth muscle lineage, whereas there is no evidence for a role for Msx genes in the endothelium.

According to our results, the Msx1 and Msx2 transcription factors are essential for proper formation of the mural cell layer. However, the weak expression of Msx1^lacZ in this layer precludes a role for Msx1 in mature VSMCs. An alternative explanation is that Msx genes are required in VSMC precursors before they are recruited to the vessel wall. To investigate this hypothesis, we constructed a Tamoxifen-inducible Msx1^CreERT2 allele (Y.L., J. Moreau, C.S.C., F. Langa-Vives and B.R., unpublished). Properties of this allele are briefly reported in Fig. S7 in the supplementary material. Msx1^CreERT2 was used in conjunction with the Rosa^mT/mG allele (Muzumdar et al., 2007).

Pregnant dams were injected three times with Tamoxifen, at E8.0, E8.5 and E9.0, and the embryos analyzed at E10.5, E11.5 or E12.5. At E10.5 and E11.5, we observed a number of GFP-positive mesenchymal cells in the region between the neural crest-derived root ganglion and the CA (Fig. 3A,B). Furthermore, from E10.5 to E12.5, the VSMC layer was progressively populated by GFP-positive cells. Many of them co-expressed α-Sma, demonstrating that they had differentiated into bona fide VSMCs (Fig. 3C-F). The CA is just forming at E9.0 (Walls et al., 2008) and is unlikely to be covered by mural cells. Therefore, the GFP-positive cells we observed in the mural layer at E12.5 after Msx1^CreERT2 activation at E8.0-E9.0 are unlikely to derive from pre-existing mural cells proliferating in situ.

Using Msx1^CreERT2; Rosa^mT/mG, we further analyzed the covering of blood vessels by GFP-positive cells in an Msx null context, taking advantage of the fact that Msx1^CreERT2 is an Msx1 null allele (Fig. 3G-J). Tamoxifen was injected at E8.0 and E8.5 and embryos analyzed at E11.5 or E12.5. At E11.5, when at least one functional allele for Msx1 and Msx2 remained, we observed GFP-positive cells in close proximity to the CA and a population of GFP-positive cells that had reached the mural layer (Fig. 3G, arrows). When both Msx1 and Msx2 were inactivated, GFP-positive cells similarly migrated to the CA region, but very few were found close to, or in, the mural layer (Fig. 3H, arrow), to which they failed to attach. At E12.5, significantly fewer GFP-positive cells were observed immediately adjacent to the CA and even fewer in the mural layer, in the mutant versus control (Fig. 3I,J).

Msx1-positive precursors were also observed to express Msx2^lacZ before they integrated into the mural layer (see Fig. S8A in the supplementary material). Furthermore, we used the Msx1^CreERT2 allele to inactivate Msx2 in an Msx1 mutant context (Msx1^CreERT2/−; Msx2^lox/lox). Tamoxifen was injected at E8.0, E8.5 and E9.0. At E11.5, the CA exhibited the same abnormal phenotype as in Sm22Cre Msx1/2 mutants, i.e. an increase in vessel diameter and the depletion of mural cells (see Fig. S8B,C in the supplementary material). We conclude that Msx2 was inactivated by Msx1^CreERT2 in Msx1-expressing precursors before they reached the mural layer.

The migration, survival and proliferation of VSMCs have been shown to depend on the matrix metalloproteinases Mmp2 and Mmp9 (reviewed by Newby, 2006). We therefore evaluated the level of Mmp2 protein by immunofluorescence on transverse sections of E12.5 embryos, in which Msx1-expressing precursors had been labeled with GFP at E8.0-E9.0 using Msx1^CreERT2 and Rosa^mT/mG alleles (Fig. 4A-F). In the control, many GFP-positive cells accumulated Mmp2 and integrated in the mural layer, whereas very few cells did so in the mutant (compare Fig. 4A,C,E with 4B,D,F). Mmp reduction in the mutant was confirmed by qRT-PCR on cephalic VSMCs, which were FACS sorted from E12.5 embryo heads using the Sm22α-Cre transgene together with Rosa^mT/mG (see Fig. S9 in the supplementary material). In Msx1^−/−; Msx2^−/− mutant VSMCs, we observed a severe reduction in the expression of both Mmp2 (52% of control) and Mmp9 (40%) (Fig. 4G). Mmp2 expression was also reduced in the Msx1^−/− mutant (56% of control), although in this genotype this was compensated by an increase in Mmp9. No reduction in Mmp expression was observed in the Msx2^−/− mutant.

Mmp2 and Mmp9 are crucial for the survival and proliferation of VSMCs (Newby, 2006). However, we observed no change in apoptosis, using Lysotracker staining (see Fig. S10A,B in the supplementary material), or in proliferation rate, using an anti-phospho-histone H3 antibody (see Fig. S10C,D in the supplementary material), between controls and Sm22Cre Msx1/2 mutants. Phospho-histone H3-positive cells represented 2.7% of the total cells in either genotype (see Fig. S10E in the supplementary material).

Mmp2 and Mmp9 have previously been characterized as targets of the BMP signaling pathway in mural cells (El-Bizri et al., 2008). Msx genes are involved in BMP signaling at several sites during development. We therefore analyzed the expression levels of a number of BMP ligands and receptors in VSMCs, FACS sorted as previously described (Fig. 5A). Bmp4 showed the most significant downregulation in Sm22Cre Msx1/2 mutant cells. Bmp2, Bmp6 and Bmp7 transcripts were also reduced, albeit to a lesser extent. The Bmp2 expression level proved to be low and its variation is not expected to account for a reduction in overall BMP signaling. By contrast, Bmp7 is expressed at a higher level than Bmp4 and might contribute to this reduction in the mutant. No significant change in expression was observed for the BMP receptor genes Bmpr1a, Bmpr1b and Bmpr2.

The reduction of Bmp4 expression around the CA was confirmed by in situ hybridization on sections from E11.5 Sm22Cre Msx1/2 embryos (Fig. 5Bb). The reduction in BMP signaling was further demonstrated by analyzing Smad1/5/8 phosphorylation by immunohistochemistry at E11.5. The number of phospho-Smad-positive cells was conspicuously reduced in the double mutant (Fig. 5Cb). Quantification of phospho-Smad-positive cells on sections showed that 22% of the cells were stained with the anti-phospho-Smad antibody in control embryos versus 13% in the double mutant, corresponding to a 40% decrease (Fig. 5D).

To further demonstrate that, in the VSMC lineage, Msx genes act upstream of BMPs, VSMCs possessing one active allele of Msx1 and Msx2 were sorted using Sm22α-activated Rosa^mT/mG, and Msx gene inactivation was mimicked in culture using Msx-targeted siRNAs. VSMCs retained their phenotype when cultured, based on α-Sma expression and morphological characteristics (Fig. 6A). Transfection efficiency was assessed using a fluorescent non-targeted siRNA (NT siRNA) (Fig. 6A), which did not interfere with Msx1 or Msx2 expression (Fig. 6B). Increasing amounts of Msx-specific siRNAs progressively reduced Msx1 and Msx2 transcript levels, reaching maximal efficiency at 25 ng/ml (Fig. 6B). Under these conditions, 88% of Msx1 and Msx2 transcripts were lost. Bmp4 and Bmp7 transcripts were correspondingly reduced, as were those of Mmp2 (Fig. 6C). To obtain further confirmation, VSMC precursors were sorted in the same way from Msx1; Msx2 double-null mutants and, after a few days in culture, they were demonstrated to express low levels of Mmp2 and insignificant levels of Mmp9 transcripts. After 48 hours in culture with increasing amounts of exogenous Bmp4, the cells did not show phenotypic changes (Fig. 6D) but the Mmp2 expression levels were raised significantly and in a dose-dependent manner (Fig. 6E).

Altogether, these results confirm that Msx genes act upstream of Bmp4 and Bmp7 in VSMC progenitors and that, as previously demonstrated (El Bizri et al., 2008), BMP signaling is required for Mmp2 expression. In culture, Mmp9 expression decreases dramatically and so no conclusion could be drawn for this gene.

As mentioned above, whereas the number of ECs and the expression of early endothelial markers were not decreased in the Sm22Cre Msx1/2 mutant, late-expressed endothelial markers were downregulated (Fig. 2G). We verified that this did not correlate with a change in arteriovenous identity of the CA using antibodies against neuropilin 1, an artery-specific marker (Klagsbrun et al., 2002) (see Fig. S11 in the supplementary material). Downregulation of endothelial markers suggested that reduced VSMC coverage leads to impairment in signaling between the two layers, resulting in defects in endothelium maturation. By qRT-PCR, we did not observe a significant difference in the expression of ligands produced by the endothelium, such as Tgfb1 or Pdgfb, in single or double Msx mutants as compared with controls (Fig. 7A). By contrast, mRNAs for mural cell-secreted factors were detected at lower levels in the Sm22Cre Msx1/2 mutants. Vegfa and Angpt1 levels were decreased by 37% and 36%, respectively (Fig. 7B). This can be readily explained by the reduction in mural cell coverage, which would impact on the endothelium. It has recently been demonstrated in culture that Tie2 (Tek – Mouse Genome Informatics) is concentrated at the cell membrane by its ligand Angpt1 (Fukuhara et al., 2008; Saharinen et al., 2008). In keeping with these results, we observed in normal embryos a high concentration of Tie2 at the EC membrane on the external side, facing the mural cells, and on the lateral side adjacent to neighboring ECs (Fig. 7C,C′,D). By contrast, Tie2 was diffusely distributed over the EC membrane in the Sm22Cre Msx1/2 mutant (Fig. 7E,E′,F), correlating with the decreased Angpt1 secretion by the thinner mural layer.

To confirm these results, we sorted ECs by FACS (using a CD31 antibody) from E12.5 heads (see Fig. S9 in the supplementary material). In ECs, Kruppel-like factor 2 (Klf2) expression is dependent on Tie2 activation (Sako et al., 2009). Accordingly, we detected a 50% decrease in Klf2 expression in the constitutive Msx1; Msx2 double-mutant ECs (Fig. 7G). Fig. 7H summarizes the consequences for the endothelium of the reduction in mural cell coverage associated with Msx deficiency.

We report for the first time that combined Msx1 and Msx2 gene deficiencies lead to major defects in blood vessels: an excess of branching and increase in caliber of major head arteries such as the CA, hemorrhages and aneurysms. None of these defects is observed in Msx1 or Msx2 single mutants. These vascular defects are related to a reduction in the mural cell coverage of head arteries.

Similar to the cardiac outflow tract (Rentschler et al., 2010), VSMCs in most head vessels derive from neural crest cells (NCCs), whereas the endothelium is of mesodermal origin (Jiang et al., 2000; Etchevers et al., 2001; Korn et al., 2002). In particular, in the mouse embryo, the contribution of NCCs has been demonstrated for the CAs, whereas neural crest-derived VSMCs could not be detected at any stage in the dorsal aorta (Jiang et al., 2000), for which VSMCs have been shown to share a lineage with somitic mesoderm (Esner et al., 2006; Wiegreffe et al., 2009). Msx1 and Msx2 are strongly expressed in NCCs (for reviews, see Bendall and Abate-Shen, 2000; Ramos and Robert, 2005). Unlike Xenopus, in which Msx1 is mandatory for neural crest formation (Monsoro-Burq et al., 2005), combined deficiencies of Msx1 and Msx2 do not preclude neural crest formation or migration in the mouse. However, they severely affect neural crest subpopulations, resulting in pleiotropic defects in NCC derivatives (Ishii et al., 2005). Thus, the Msx1; Msx2 double mutation impairs the differentiation, but not migration, of cranial NCCs that form the frontal bones (Han et al., 2007). In the outflow tract of the heart, Msx genes are required to inhibit excessive proliferation of post-migratory NCCs (Chen et al., 2007). By contrast, Msx genes do not seem to play a role in the proliferation of the neural crest-derived cells that contribute to the cranial ganglia and the first pharyngeal arch, but instead function in preventing their apoptosis (Ishii et al., 2005).

We propose that the head smooth muscle defects reported here are linked to a specific neural crest subpopulation that depends on functional Msx genes to give rise to smooth muscle progenitor cells. Indeed, some images (Fig. 3A,B) strongly suggest that the Msx1-expressing VSMC precursors migrate from dorsal regions. Of note, some VSMCs form in Msx mutant head vessels, confirming that only a subpopulation of precursors is affected. This is in agreement with the heterogeneity of origin of VSMCs (reviewed by Majesky, 2007). In addition, we have not observed branching defects in the intersomitic vessels or a reduction in the mural coverage of the dorsal aorta (data not shown). Thus, Msx deficiency does not seem to affect non-neural crest-derived VSMCs or the vessels that they cover.

At E12.5, Msx1 is predominantly expressed in the endothelium. Defects in the double mutant might therefore result from the conjunction of Msx1 deficiency in the endothelium and Msx2 deficiency in the VSMCs. However, the same abnormal vascular phenotype was observed in Sm22α-Cre; Msx1^lox/lox; Msx2^lox/lox mutants, whereas the specific mutation of Msx1 and Msx2 in the endothelium did not lead to vascular defects. Our data show that the primary defect in the Sm22Cre Msx1/2 mutant lies in mural cell precursors before they reach the blood vessels. The Sm22Cre transgene that we used is expressed early in the mural lineage (see Fig. S2 in the supplementary material), allowing inactivation of Msx1 and Msx2 before VSMCs attach to the mural layer. Taking advantage of the Msx1^CreERT2 allele and an inducible reporter, we could trace back cells that once expressed Msx1. Tamoxifen injections were performed at stages (E8.0-E9.0) when mural cells have not yet differentiated in the head vessels (Walls et al., 2008). Under these conditions, many GFP-positive cells were observed at E11.5 and even more at E12.5 in the mural layer. This implies that these cells derive from VSMC precursors that express Msx1, which migrate to progressively populate the mural layer. VSMCs derived from Msx1-expressing precursors no longer express Msx1 after differentiation. Similarly, switch-off of Msx1 expression before differentiation has been reported for VSMC adventitial progenitors in the aorta (Passman et al., 2008).

Cell lineage studies clearly demonstrate that cells derived from Msx1-expressing precursors accumulate in normal amounts in the region around the CA, in a context of Msx deficiency. However, most of these cells fail to reach the artery and to integrate into the mural layer. This suggests late migration defects in the VSMC precursors, linked to the reduction in Mmp2 and Mmp9 expression in the mutant. Mmp2 and Mmp9 are known to free smooth muscle cells from the cell-matrix contacts that normally restrict their migration (Kenagy et al., 1997; Newby, 2006). The majority of Msx1-expressing VSMC precursors that fail to reach the CA do not accumulate detectable amounts of Mmp2 and might therefore be impeded in making their way through the vessel ECM. Of note, some cells manage to integrate into the mural layer even though they do not express Mmp2. These cells possibly express other proteinases, in keeping with the phenotypic heterogeneity of VSMCs (Majesky, 2007).

Msx genes have been previously associated with BMP signaling at several sites during development. Functional analysis of tooth or palate formation further demonstrated that these genes can act either upstream or downstream of Bmp4 (Chen et al., 1996; Zhang et al., 2002). Sm22Cre Msx1/2 double mutants exhibit a severe downregulation of the BMP pathway in the VSMC lineage. Bmp7 is the prevalent ligand of this family in the head VSMCs and its depletion, together with that of Bmp4, should result in a strong reduction in global BMP signaling in mutant VSMCs. The BMP pathway has pleiotropic effects on the vasculature, as discussed by Abe (Abe, 2006). Manipulations that reduce BMP levels usually result in impairment of VSMC coverage and in a correlative dilation of the vessel, e.g. Flk1-driven deletion of Bmpr1a (Park et al., 2006), Sm22α-driven deletion of Bmpr1a (El-Bizri et al., 2008), knockdown of Bmpr2 (Liu et al., 2007) and mutation of Smad5 (Yang et al., 1999). These phenotypes are strikingly similar to those we describe in the head vessels of the Sm22Cre Msx1/2 mutant. Interestingly, the BMP pathway plays a major role in Mmp2 and Mmp9 expression. Sm22α-Cre-driven inactivation of Bmpr1a severely affects Mmp2 and Mmp9 expression in VSMCs, which leads to a reduction in mural cell coverage and consequent vessel dilation (El-Bizri et al., 2008). In head VSMCs, Msx genes seem to act upstream of BMP expression and consequently before BMP receptor activation, as the level of Smad phosphorylation is reduced in the double mutant. Indeed, we show that Mmp2 expression can be rescued in mutant VSMCs by exogenous Bmp4. Furthermore, Roybal et al. (Roybal et al., 2010) have shown that Bmp4 expression is similarly decreased in NCC-derived osteogenic cells upon inactivation of both Msx1 and Msx2 by a Wnt1-Cre transgene, although the overall level of BMP signaling is enhanced. Altogether, these results suggest an autocrine or paracrine mechanism among VSMCs producing their own BMPs. This is in keeping with the observation that gut smooth muscle precursor cells also produce BMPs and their receptors concurrently (Torihashi et al., 2009).

In the Sm22Cre Msx1/2 mutant, markers expressed late in the endothelium (Vwf, VE-cad) are downregulated. This suggests that impairment of the endothelium maturation process derives from defects in mural cells. Indeed, reduction in the Angpt1 expression level correlates well with the reduction in the number of VSMCs, which predominantly secrete this factor. A decrease in Angpt1 expression leads to a change in the membrane localization of Tie2 in ECs, as demonstrated in cell culture (Fukuhara et al., 2008; Saharinen et al., 2008) and which we confirm in vivo. Furthermore, the expression of Klf2, which is induced by activation of Angpt1 (De Val and Black, 2009; Sako et al., 2009), is significantly reduced in mutant ECs. Our interpretation is that reduced coverage by mural cells leads to a reduction in the secretion of mural factors, such as Angpt1. Consequently, there is an incomplete activation of endothelial receptors (such as Tie2), resulting in impairment of EC maturation.

Our in vivo data show that the Msx family of homeodomain transcription factors plays a major role in head vascular maturation and that their mutation leads to mural layer defects. We demonstrate Msx1 expression in VSMC precursors, which require both Msx1 and Msx2 in order to integrate into the mural layer. We propose the following mechanism (Fig. 8): in head VSMCs, Msx factors act upstream of the BMP expression that controls the expression of Mmp2 and Mmp9; deficiency in these metalloproteinases prevents VSMCs from reaching the mural layer, leading to reduced mural coverage, vessel dilation and impairment of endothelium maturation.

We are very grateful to Drs Margaret Buckingham, Colin Crist, Stéphane Vincent and Didier Montarras for critical reading of the manuscript, to Dr Robert Maxson and Dr Masashi Yanagisawa for generously sharing mouse strains, to Prof. Pierre Chambon for the gift of CreERT2, and to Pierre-Henri Commere from the flow cytometry platform of the Institut Pasteur for cell sorting. This work was supported by the Institut Pasteur, the CNRS and grants from the French Association pour la Recherche sur le Cancer (ARC) and Ligue contre le Cancer (LCC). M.L. is the recipient of a fellowship from the Portuguese Fundação Ciência e Tecnologia (FCT).