Ccm1 is required for arterial morphogenesis: implications for the etiology of human cavernous malformations

Stroke is an important cause of morbidity and mortality in children where it ranks in the top 10 causes of death(Arias and Smith, 2003; Lynch et al., 2002). Approximately 20% of pediatric stroke cases are due to intracranial hemorrhage, the most common underlying cause of which is vascular malformations (Lynch et al.,2002). Cerebral cavernous malformations (CCM) represent an important subset of such vascular malformations. They are common, being found in 0.5% of the population in autopsy and MRI surveillance studies(Del Curling et al., 1991; Otten et al., 1989; Robinson et al., 1991). These lesions are classically defined by enlarged and thin-walled vascular structures in the central nervous system (CNS) without intervening brain parenchyma, lined by endothelial cells and lacking supporting vascular smooth muscle cells.

Recently it has been shown that a monogenic form of CCM is linked to loss-of-function mutations in the CCM1 locus on chromosome 7q, which encodes the KRIT1 protein (Craig et al.,1998; Laberge-le Couteulx et al., 1999; Marchuk et al.,1995; Moriarity et al.,1999; Rigamonti et al.,1988; Sahoo et al.,1999). KRIT1 is an intracellular protein with ankyrin repeats and a FERM domain (Serebriiskii et al.,1997; Zawistowski et al.,2002), initially cloned on the basis of a yeast two-hybrid interaction with KREV1/RAP1a, a small RAS family GTPase(Serebriiskii et al., 1997). Subsequently, additional 5′ sequences were found that extend the open reading frame (Sahoo et al.,2001; Zhang et al.,2000). The longer protein did not interact with KREV1/RAP1a, but instead showed strong interaction with integrin cytoplasmic domain associated protein-1-alpha (ICAP1α) (Faisst and Gruss, 1998; Zawistowski et al., 2002; Zhang et al.,2001), which binds a similar NPXY motif on both KRIT1 andβ1-integrin (Zawistowski et al.,2002; Zhang et al.,2001). In addition, KRIT1 interacts with the plus ends of microtubules, suggesting a possible role in microtubule targeting(Gunel et al., 2002). Although these biochemical data have provided useful insights into KRIT1 function in vitro, the in vivo role of KRIT1 remains unclear, particularly with respect to vascular disease.

Using in situ hybridization, others have reported that Ccm1 is expressed ubiquitously until E10.5, at which point the expression starts to become restricted to neural and epithelial tissues(Denier et al., 2002; Kehrer-Sawatzki et al., 2002). These studies suggest that Ccm1 may play a role in neural development or function, and have led some authors to suggest that Ccm1 does not play a role in cardiovascular development(Kehrer-Sawatzki et al.,2002). The neural and epithelial expression of Ccm1 in adulthood suggests that cavernous malformations may be the result of primary neural defects.

Human vascular malformation syndromes, such as CCM, affect the junction between veins and arteries. Arteries and veins form separately but follow parallel trajectories and can be distinguished on the basis of molecular markers, even prior to the onset of flow at E8.5(Ji et al., 2003; McGrath et al., 2003; Wang et al., 1998). Recent studies in zebrafish have demonstrated a genetic pathway governing arterial identity, in which the expression of vascular endothelial growth factor (Vegf) is necessary for the expression of arterial markers and morphology (Lawson et al.,2002). Specifically, Vegf is necessary for the expression of Notch family members in arteries, and Notch signaling, in turn, is required for the arterial expression of ephrin B2(Efnb2) and correct arterial-venous relationships(Lawson et al., 2001). A specific role for this genetic pathway in arterial identity has not yet been confirmed in mammalian systems.

The expanding link between human vascular disorders such as hereditary hemorrhagic telangiectasia (Li et al.,1999; Park et al.,2003; Sorensen et al.,2003; Urness et al.,2000) and CADASIL (Gridley,2001; Joutel et al.,1996) to molecular pathways implicated in arterial identity and vascular patterning suggests that Ccm1 may have an essential role in arterial development. Here we demonstrate, using murine gene targeting, that Ccm1 is required for vascular development. Our data indicate that Krit1-associated vascular defects are not secondary to disrupted neural patterning. We also demonstrate impaired arterial identity in embryos lacking Ccm1, and provide evidence that Krit1 lies upstream of Notch signaling in the vasculature in mice and humans. Our studies suggest that mutations in Ccm1 impact vascular development and disease by disrupting a genetic pathway important in establishing arterial identity.

We designed a construct to delete the first ankyrin repeat of Ccm1by replacing most of coding exon 6 and all of exon 7 with an internal ribosomal entry site and the E. coli lacZ gene. Genomic fragments of 2.0 kb and 4.5 kb from the 5′ and 3′ ends of the targeted region,respectively, were cloned into the pPNT vector(Tybulewicz et al., 1991),together with the IRES-lacZ gene from plasmid pENβ16-B (both plasmids were a gift from Dr Yuan Zhuang). Transfected R1 embryonic stem cells were screened for correct integration of the targeting vector by PCR and sequencing, as well as by Southern blotting. A founder that transmitted the Ccm1^tm1Dmar allele to its progeny was obtained. Mouse genotypes were determined by PCR (Barrow et al., 2003) using allele-specific primers (sequences available upon request). Mice heterozygous for the allele were backcrossed for at least five generations to C57BL/6J mice.

We carried out in situ hybridization at 70°C as previously described(Urness et al., 2000). Probes were generated by in vitro transcription of appropriate plasmids. We were provided plasmids for Hand1, Hand2 (Dr Deepak Srivastava), Dll4 (Dr John Shutter, Amgen), Notch4 (Dr Thomas Gridley)and Efnb2 (Dr David Anderson). Using RT PCR on mouse cDNA with appropriate primers (sequences available upon request) we generated probe templates for Ccm1, IRES-lacZ, Nppa, Six3, Otx2, Fgf8, Gbx2, Shh, Pax7,Nkx2-2, Ntn1, Brachyury and Vegfa. A probe for MafB was generated by in vitro transcription of an EST (IMAGE: 2811217).

Mouse tissues were studied with antibodies to Pecam (clone MEC13.3, BD Biosciences) and α-smc actin (clone 1A4, DAKO). Human tissues were studied with antibodies to PECAM (clone JC70A, DAKO) and NOTCH4 (polyclonal H-225, SantaCruz).

Frozen sections were stained first with Pecam antibody (clone MEC13.3, BD Biosciences), followed by antibodies against phosphohistone H3 or cleaved caspase 3 (Cell Signaling Technology). Finally, appropriate fluorescent secondary antibodies (Jackson ImmunoResearch) were applied. Slides were counterstained with DAPI (Molecular Probes).

Total endothelial nuclei and mitotic endothelial nuclei were counted from the dorsal aorta, branchial arch arteries, and vitelline arteries, and data were labeled as originating from either caudal or rostral regions. All sections distal to the sinus venosus of the heart were considered caudal. The mean percentages of rostral or caudal mitotic nuclei for each of three separate experiments were compared using an unpaired t-test (InStat,by GraphPad software).

Embryos were prepared for confocal immunofluorescent detection of PECAM antigen as previously described (Drake and Fleming, 2000). Images were obtained using an Olympus FluoView™ 300 confocal microscope (University of Utah Cell Imaging Core).

Ink injections were carried out in E8.5-E9.5 embryos as previously described (Urness et al.,2000). Following ink injection, yolk sacs were removed and embryos were photographed directly.

Mice heterozygous for Efnb2 expressing a tau-lacZtransgene under the control of the Efnb2 locus were previously generated and described (Wang et al.,1998). We crossed this transgene into Ccm1 heterozygous mice. Double heterozygous (both Ccm1 and Efnb2) male mice were mated with Ccm1 heterozygous females to generate embryos heterozygous for Efnb2 with all possible Ccm1 genotypes. Genotyping was performed with allele-specific primers (sequences available upon request). Staining for β-galactosidase expression was performed with X-gal as previously described (Sorensen et al., 2003). Littermates homozygous for Ccm1 and wild type at the Efnb2 locus were used to control for the negligible embryonicβ-galactosidase expression from the Ccm1 IRES-lacZconstruct.

RNA was isolated from single E8.8 (13 somite) embryos (RNAqueous 4PCR kit,Ambion) and used as template to make random primed cDNA (RetroScript kit,Ambion). Assays for Efnb2, Dll4, Notch4 and Vegfa were obtained, as well as for Ccm1 as a control for genotype(Assays-on-Demand, Applied Biosystems), and used according to the manufacturer's instructions on an Applied Biosystems 7900HT thermal cycler(University of Utah Genomics Core Facility). Transcripts were normalized in relation to Gapdh expression (rodent GAPDH, Applied Biosystems). Comparisons were made between three separate pairs of wild-type and homozygous mutant embryos, and reported in relation to wild-type expression (samples run in triplicate).

To explore the role of Ccm1 in vascular development, we generated a Ccm1 mutant allele (Ccm1^tm1Dmar) by disrupting the gene at the first ankyrin repeat with our targeting vector(Fig. 1A-C). At E8.5, no Ccm1 expression was detected in homozygous mutant mice, although we observed ubiquitous Ccm1 transcript expression in wild type(Fig. 1D), similar to previous reports (Denier et al., 2002; Kehrer-Sawatzki et al., 2002). Heterozygous mutant mice were phenotypically normal on the C57BL/6J genetic background. Homozygous mutant embryos were grossly indistinguishable from their littermates prior to E9.0. Mutant embryos suffered from generalized developmental arrest after E9.5, and retained a primitive yolk sac vascular network, failed to complete turning, and developed pericardial effusions and atrial enlargement. Ccm1^tm1Dmar/Ccm1^tm1Dmar embryos(hereafter referred to as Ccm1^-/-) were disintegrating by E10.0 and no embryos survived beyond E11.0.

Defective development in Ccm1^-/- embryos was first observed in the vascular system. From E8.5-E9.5, the vessels of the cephalic mesenchyme became progressively dilated(Fig. 2A-D). These vessels will invade the neural tube and start to form the intracranial vascular network at E10.0 (Ruhrberg et al., 2002),a stage in development that Ccm1^-/- embryos fail to reach. The dilatation in the precursor vessels of the brain is reminiscent of the dilated endothelial sacs observed in the intracranial vasculature of patients with functional hemizygosity at the CCM1 locus(Fig. 2E,F).

To determine developmental defects that could be directly attributed to Ccm1, we searched for the earliest histologic and morphologic defects in mice lacking Ccm1. We observed vascular dilatation starting at E8.5. In a cross-sectional survey through multiple developmental stages,staining for Pecam revealed ectatic vessels in the embryo(Fig. 3A-D). Dilatation occurred throughout the aorta just distal to the heart and extended to the most caudal regions of the embryo. Shortly after their formation between E8.5 and E9.0, the intersomitic arteries of the embryo also enlarge (data not shown). We observed no enlargement of the adjacent cardinal veins, and no evidence of arteriovenous shunts. Thus, loss of Ccm1 leads to arterial dilatation that begins after the axial vascular pattern has been established by vasculogenesis.

We sought to determine whether this aortic enlargement was attributable to increased endothelial proliferation or decreased apoptosis. We assessed proliferation by double immunolabeling embryos with antibodies against Pecam and phosphorylated histone 3 (H3P), a specific marker of the mitotic phase of the cell cycle (Brenner et al.,2003; Hendzel et al.,1997; Nechiporuk and Keating,2002). We observed a twofold increase in mitotic endothelial cells from the dilated distal aortae in Ccm1^-/- embryos, as compared with wild-type controls (Fig. 3E-G). There were no differences in H3P staining of more rostral regions of the Ccm1^-/- aortae and branchial arch arteries compared with wild-type controls. There was also no difference in vascular apoptosis observed in Ccm1^-/- embryos, as assessed by immunofluorescence using antibodies against cleaved Caspase 3. Thus, mice lacking Ccm1 show extensive vascular dilatation following vasculogenesis associated with an increased endothelial proliferative rate.

To obtain a more complete view of the vasculature, we performed whole-mount confocal immunofluorescence using Pecam antibodies(Fig. 4A-D). The enlargement of the distal dorsal aortae was again observed (arrow in Fig. 4B); however, in contrast to the extensive vascular dilatation, we observed a discrete region of vascular narrowing in the first branchial arch artery and adjacent proximal dorsal aorta that develops at E8.5 (Fig. 4D). All Ccm1^-/- embryos at E8.5 have this narrowing; however, the extent and severity of involvement is variable.

To understand the effects that Ccm1^-/- vascular defects have on circulation, we performed India ink microangiography. Injection of India ink into the ventricle of an E8.5 heart filled the first branchial arch artery and entire dorsal aorta of the wild-type embryo(Fig. 4E). By contrast, no ink passed through the branchial arch arteries of Ccm1^-/-embryos and the dilated portions of the dorsal aorta were not opacified(Fig. 4F). Similar observations were made at E9.5, at which point the normal passage of ink is through the second and third pairs of branchial arch arteries(Fig. 4G). A small amount of ink was observed in the narrowed first and second branchial arch arteries of Ccm1^-/- embryos, with minimal filling of the adjacent dorsal aorta. The dilated vascular malformation of the caudal embryo was not seen angiographically (Fig. 4H). This is analogous to human CCM lesions, which are typically not observed on cerebral angiography(Robinson et al., 1993),implying that such lesions may also have narrowed arterial inflow.

We further characterized the time-course of this narrowing with a developmental series of embryo cross-sections stained for Pecam. These studies indicated that the proximal aorta and the first branchial arch artery formed normally at E8.0 (Fig. 5A,B). At E8.5, the first branchial arch artery had failed to enlarge(Fig. 5C,D). At E9.0, the vestige of the first arch artery and adjacent aorta remained severely narrowed(Fig. 5E,F), and the formation of the second branchial arch artery was abnormal(Fig. 4G,H; and data not shown). We postulated that this narrowing might be due to increased vascular apoptosis, yet we found no immunological evidence of increased cell death at E8.5 (data not shown). We also did not observe differential endothelial proliferation by immunostaining against phosphorylated histone H3 in this discrete region of the embryo, although this method may lack sufficient power to detect a difference in such a small number of cells. Thus, in addition to vascular dilatation, we observed narrowing of the branchial arch arteries and rostral dorsal aorta in Ccm1^-/- embryos at E8.5.

Although vascular defects are first observed at E8.5 and progress with age,we found no evidence of cardiac defects in embryos from E8.0-E9.5. During these stages, mice lacking Ccm1 completed cardiac looping, developed appropriate endocardial cushions in the AV canal, and showed similar myocardial trabeculation to wild-type embryos (see Fig. S1AD at Supplementary Information). The expression of the cardiac transcription factors Hand1 (also known as eHand) and Hand2 (also known as dHand), as well as the marker Nppa (natriuretic peptide precursor type A,also known as atrial natriuretic factor) was found to be similar at E8.5 in both wild-type and Ccm1^-/- embryos (see Fig. S1E-J). As mutant embryos age and begin to suffer from generalized developmental arrest after E9.5, we have frequently observed pericardial effusions and atrial enlargement (data not shown). These studies indicate that vascular defects are the primary defects affecting Ccm1^-/- embryos and suggest that cardiac defects observed later in development are secondary.

Concurrent with early vascular development, the embryonic neural tube becomes organized and structured in both anteroposterior (AP) and dorsoventral(DV) aspects as an early step in neural development(Briscoe and Ericson, 2001; Liu and Joyner, 2001). The neural expression of Ccm1 and the CNS predilection of cavernous malformations led us to investigate neural patterning in Ccm1^-/- embryos. In order to determine whether vascular defects occur secondary to neural defects in mice lacking Ccm1, we used RNA in situ hybridization to study the AP and DV organization of the developing CNS at E8.5. From anterior to posterior, the CNS at this stage can be divided into the forebrain, midbrain and hindbrain. The hindbrain itself can be subdivided further into 8 rhombomeres(Liu and Joyner, 2001). Molecular markers can also distinguish the midbrain-hindbrain junction, or isthmus, which has an important organizer function(Liu and Joyner, 2001). The homeobox gene Six3 (Fig. 6A,B) is expressed in the forebrain at E8.5(Oliver et al., 1995). The transcription factor Otx2 (Fig. 6C,D) is also expressed in the forebrain and extends into the midbrain up to the isthmus, where expression abruptly stops(Liu and Joyner, 2001). The secreted molecule Fgf8 (Fig. 6E,F) has a domain of expression restricted to the isthmus at the midbrain-hindbrain junction (Liu and Joyner, 2001). The homeobox gene Gbx2(Fig. 6G,H), which helps establish the posterior boundary of the isthmus, is then expressed throughout the hindbrain (Liu and Joyner,2001). Within the hindbrain the leucine zipper transcription factor MafB (Fig. 6I,J) is limited in expression to rhombomeres 5 and 6(Grapin-Botton et al., 1998). Along the AP axis of the neural tube, the secreted protein Shh(Fig. 6K,L) is expressed in the notochord and floor plate where it performs an inductive role(Briscoe and Ericson, 2001). The homeobox gene Pax7 (Fig. 6M,N) is expressed in the dorsal neural tube(Fu et al., 2003; Mansouri et al., 1996). The homeobox gene Nkx2-2 (Fig. 6O,P) is expressed in the lateral floor plate(Charrier et al., 2002; Fu et al., 2003). The axon guidance molecule Netrin1 (Fig. 6Q,R) is expressed in the medial floor plate as well as in adjacent somites (Serafini et al.,1996). No alterations in the normal expression pattern of these genes were observed in embryos lacking Ccm1. We conclude that AP and DV patterning of the neural tube in Ccm1^-/- embryos is intact at the onset of vascular defects. These studies indicate that the vascular developmental defects observed in Ccm1^-/- embryos are not secondary to a primary neural developmental defect.

After the initial endothelial tubes form, arteries are distinguished from veins morphologically and by the expression of genetic markers. The establishment of arterial identity is an important early event in angiogenesis, with relevance to vascular defects and malformations(Sorensen et al., 2003; Urness et al., 2000). Starting at around E9.0, vascular smooth muscle cells are recruited by the arterial endothelium (Li et al., 1999). This recruitment occurs at an earlier stage of development in arteries than veins. We analyzed arterial smooth muscle recruitment by immunostaining for alpha smooth muscle actin (α-Sma). At E9.5 α-Sma staining in wild-type embryos is seen in arterial vessels and in cardiac myocytes, but, as expected, is not observed in veins (Fig. 7A). In Ccm1^-/- embryos, only α-Sma staining of arteries is lost; expression in the heart is maintained(Fig. 7B). Thus, there is a loss of α-Sma staining that is specific to Ccm1^-/-arteries, indicative of a failure to recruit arterial smooth muscle cells.

Prior to arterial smooth muscle cell recruitment, and prior to the onset of flow (Ji et al., 2003; McGrath et al., 2003), the arterial endothelium is specified and can be distinguished from venous endothelium by genetic markers (Wang et al., 1998). To identify whether Ccm1^-/-arteries are appropriately specified, we examined the expression of the arterial-specific gene Efnb2 by in situ hybridization, and Efnb2-tau-lacZ transgene (Wang et al., 1998) expression by X-Gal staining. Vascular expression of Efnb2 is downregulated in both the embryo and yolk sac. In the embryo, early Efnb2 transgene expression is detected in the arterial endothelium, somites, nephrogenic cords and hindbrain. In the absence of Ccm1, arterial expression of Efnb2 is not detected; however,expression in the somites, nephrogenic cord and brain is unaffected(Fig. 7C,D; and data not shown). In the yolk sac, Efnb2 is expressed only in the endothelium. This is most clearly seen by X-gal staining at E9.0, after the primary plexus has remodeled into an arborizing vascular network. Efnb2 transgene expression is detected in the arterial vessels of the posterior yolk sac(Fig. 7E). In the absence of Ccm1, no staining is observed(Fig. 7F). A similar loss of arterial endothelial expression of Efnb2 transcript is observed in Ccm1^-/- yolk sacs before the onset of circulation at E8.5,when the vasculature is composed of a meshwork of homogenously sized endothelial tubes (Fig. 7G-J). The diminished expression of arterial markers and subsequent arterial morphologic defects suggest that arterial identity is impaired in mice lacking Ccm1.

A genetic pathway has been described in zebrafish placing Efnb2expression genetically downstream of Notch gene signaling. Notch signaling, in turn, is genetically downstream of Vegfain the control of arterial identity(Lawson et al., 2001; Lawson et al., 2002). The expression of the arterial-specific mammalian Notch genes, Dll4 and Notch4 was significantly downregulated in Ccm1^-/- embryos by E8.5(Fig. 8A-D), prior to the gross appearance of the mutant phenotype and prior to the onset of circulation. The expression of the pan-endothelial markers Pecam(Fig. 4A,B) and Kdr(data not shown) remained unperturbed. We observed similar intensity and distribution of expression of Vegfa(Fig. 8E,F) in Ccm1^-/- embryos, by in situ hybridization, compared with wild type. The downregulation of arterial markers with intact expression of Vegfa in mice lacking Ccm1 was confirmed and quantified by real-time quantitative RT-PCR (Fig. 8G). These results would suggest that Ccm1 lies genetically downstream of Vegf in the control of Notchsignaling and arterial morphogenesis.

Our experiments indicate that arterial identity and development are impaired in Ccm1^-/- mice. In view of the observed similarities between enlarged vessels in Ccm1^-/- embryos and human cerebral cavernous malformations, we speculated that a similar disruption of arterial identity (reduced expression of DLL4, NOTCH4and EFNB2) might underlie the pathology in human cavernous malformations. Because antibodies that recognize human EFNB2 and DLL4 are not available, antibodies against human NOTCH4 were used to examine arterial identity in previously excised vascular malformations from three affected individuals of a family with a frameshift mutation(Sahoo et al., 1999) in CCM1. Control experiments using autopsy-derived, formalin-fixed brain tissue from unaffected individuals showed prominent and specific expression of NOTCH4 in arterial endothelium and smooth muscle cells(Fig. 9A). In affected individuals, we were not able to detect NOTCH4 expression in cavernous lesions, and there is a marked reduction of NOTCH4 in the arteries of the brain tissue adjacent to the vascular malformation(Fig. 9B). Thus, similar to mice lacking Ccm1, humans with loss-of-function mutations in CCM1 have decreased expression of NOTCH4 in association with vascular malformations.

The neural and epithelial expression of Ccm1 in adulthood had suggested that cavernous malformations might be the result of primary neural defects. In this study we demonstrate an essential role for Ccm1 in vascular development. The earliest defects in mice lacking Ccm1 are vascular, and result in thin-walled, endothelial caverns that are reminiscent of human cerebral cavernous malformations. Dilation is observed in the vessels of the cephalic mesenchyme and the dorsal aorta of the caudal embryo, with simultaneous narrowing of the branchial arch arteries and proximal dorsal aorta. No defects in neural patterning or cardiac development are observed prior to the onset of vascular defects. The vascular defects are associated with a loss of arterial endothelial markers (Efnb2, Dll4 and Notch4) prior to the onset of flow, and a failure of arterial smooth muscle recruitment to the developing arteries. In human patients, we similarly observe loss of NOTCH4 expression in arteries associated with cerebral cavernous malformations. We conclude from our data that the loss of Ccm1 leads to primary vascular defects and disrupts the molecular pathway regulating arterial identity.

Vasculogenesis, the de novo formation of endothelial tubes from angioblast precursors (Risau, 1997), is intact in Ccm1^-/- mice, as the vascular pattern is formed correctly by E8.0. In this study, we demonstrate that Ccm1 is essential for angiogenesis, the process of vessel maturation and network remodeling wherein new vessels arise from preexisting vessels(Risau, 1997). Mice lacking Ccm1 die by E11 with prominent vascular defects associated with inappropriate angiogenic remodeling. Arteries in Ccm1^-/-embryos develop morphologic defects starting at E8.5, with marked enlargement of the caudal aorta and the vessels of the cephalic mesenchyme. We observed increased endothelial proliferation in the dilated aorta of the caudal embryo compared with wild type. This dilatation and proliferation is not a consequence of increased flow as demonstrated by a lack of circulation of India ink when injected into the cardiac ventricle. Narrowing of the branchial arch arteries and the proximal aortae is also present at E8.5. Cardiac development appears entirely normal until E9.5, when generalized developmental arrest occurs in Ccm1^-/- mice. We conclude that Ccm1 is required for angiogenesis.

The neural and epithelial expression of Ccm1 in adulthood suggests that cerebral cavernous vascular malformations may be secondary to neural defects. Our results indicate that loss of Ccm1 expression results in primary vascular defects in the embryo. Vascular defects in mice lacking Ccm1 are uniformly present at E8.5. We demonstrate that the anteroposterior and dorsoventral patterning of embryos lacking Ccm1is normal at E8.5, based on the expression of a broad panel of genes. The optic and otic anlagen are both present, and no morphologic neural tube defects are observed despite vascular enlargement throughout the embryo. These results suggest that endothelial cells require Ccm1. To determine whether Ccm1 expression is necessary within the endothelial cell itself, or whether endothelial cells depend upon interactions with adjacent cells expressing Ccm1, further experiments using conditional mutagenesis will be important.

Arteries are distinguished from veins even at the earliest stages of development and prior to the onset of circulation. This distinction is first recognized on the basis of a unique set of molecular markers that label arteries and not veins. We examined the expression of the known arterial markers, Efnb2, Notch4 and Dll4, and found significant downregulation of the arterial expression of all three genes by in situ hybridization and by real-time quantitative RT-PCR. The extravascular expression of Efnb2 from somites, nephrogenic cord and hindbrain remains intact in Ccm1^-/- embryos, suggesting a specific effect of Ccm1 on arterial identity. These data demonstrate that arterial specification of endothelial tubes is disrupted in Ccm1^-/- mice.

Recently, much progress has been made in understanding the process by which arteries are distinguished from veins. This process is best understood in the zebrafish, a model organism particularly well suited to the establishment of genetic pathways (Fishman,2001). In the zebrafish, loss of Notch signaling leads to a reduction of arterial Efnb2 expression, and abnormal arterial-venous connections as seen angiographically(Lawson et al., 2001). The expression of Vegf was shown to be necessary for Notch signaling and Efnb2 expression (Lawson et al.,2002). This genetic pathway has not been confirmed in mammalian systems, in part because of the relative difficulty in genetically manipulating mice compared with zebrafish. In our studies, we found no detectable disruption of Vegf expression in mice lacking Ccm1. These data suggest that Ccm1 lies genetically downstream of Vegf, and indicates that it is genetically upstream of Notch and Efnb2 signaling in the control of arterial specification.

In mammalian systems, it has been demonstrated that Notchsignaling is important for vascular development(D'Amore and Ng, 2002; Gridley, 2001; Krebs et al., 2000; Leong et al., 2002). The combined loss of the Notch1 and Notch4 receptors leads to severe vascular defects including narrowed or collapsed dorsal aortae(Krebs et al., 2000). The vascular defects observed in Notch1 and Notch4 double-mutant embryos are more severe than those observed in mice lacking Notch1alone, whereas mice lacking Notch4 show no phenotype(Krebs et al., 2000). These studies have led others to hypothesize that genetic interactions between Notch4 and Notch1 are important for angiogenic vascular remodeling (Krebs et al.,2000). Our studies support a crucial role for Notchsignaling in angiogenesis, and suggest that disruption of Notch4,with its probable ligand Dll4, may contribute to the vascular phenotype observed in Ccm1^-/- mice. These studies lead to a hypothesis that is readily tested: whereas disruption of Notch4 may lead to no phenotype, ablating the endothelial expression of both Notch4 and Dll4 in mice will result in a severe vascular phenotype reminiscent of Ccm1^-/- mice.

The disruption of the Notch signaling pathway in Ccm1^-/- mice led us to postulate that a similar reduction in NOTCH expression may be important in the pathogenesis of human cavernous malformations. Our studies of affected individuals with known mutations in CCM1 (Marchuk et al.,1995; Sahoo et al.,1999) show that NOTCH4 is significantly reduced in arterioles associated with CCM lesions. We were unable to test the expression of EFNB2 and DLL4 because neither reliable antibodies, nor frozen sections for in situ hybridization from these affected individuals are available. We observe additional parallels between the human lesions and mice lacking Ccm1. Both exhibit enlarged, thin-walled vessels that lack vascular smooth muscle support. In both instances, the enlarged vessels do not opacify well angiographically, suggesting that human lesions are isolated from a significant arterial supply. We demonstrate that cavernous lesions in mice lacking Ccm1 are isolated from arterial inflow by vascular narrowing proximal to the lesion. This suggests that similar narrowing may be present in human lesions. The decrease in NOTCH4 and these other parallels lead us to further suggest that impaired arterial specification may contribute to the etiology of CCM.

This characterization of Ccm1^-/- mice exemplifies the efficacy of studying vascular malformation genes to better understand arterial development. We have previously characterized mice lacking the two genes known to cause Hereditary Hemorrhagic Telangiectasia, an autosomal dominant vascular dysplasia caused by loss-of-function mutations in either endoglin(Eng) (Li et al.,1999) or activin receptor-like kinase 1 (Acvrl1)(Urness et al., 2000). Affected individuals develop enlarged arteriovenous channels that bypass capillary beds, and are prone to rupture(Guttmacher et al., 1995). Mice lacking Acvrl1 and Eng fail to maintain distinct arterial and venous beds, developing similar arteriovenous shunts(Sorensen et al., 2003; Urness et al., 2000). These three examples suggest that a natural mutagenesis screen has occurred in humans, in which the readout of vascular malformations suggests disruption of genes involved in arterial and venous development.

We thank L. D. Urness and K. R. Thomas for helpful comments; D. N. Louis for cavernous malformation specimens; J. Lee and L. K. Sorensen for technical assistance; and C. Rodesch from the University of Utah Cell Imaging Core. Embryonic stem cell culture and generation of chimeric mice was performed by the Duke Comprehensive Cancer Center Transgenic Mouse Facility. We would also like to thank C. B. McKinney and the members of the Huntsman Cancer Institute Tissue Access and Tissue Imaging and Analytical Morphology Cores, supported by the Huntsman Cancer Foundation and grants CA2014 and CA73922 from the NIH. This work was supported by grants from the National Institutes of Health and the American Cancer Society, and a Technology Commercialization Grant from the University of Utah. K.J.W. is a Pfizer Postdoctoral Fellow.