Combinatorial interaction between CCM pathway genes precipitates hemorrhagic stroke

Endothelial cell-cell adhesion is important for proper blood vessel formation, maintaining the integrity of the vasculature, and regulating diffusion of molecules between the luminal and abluminal faces of blood vessels. Weakened endothelial junctions are implicated in intracranial hemorrhage (ICH), which are severe and debilitating forms of stroke in humans that include the cerebral cavernous malformation (CCM) disorders. CCM occur either sporadically or with an autosomal dominant inheritance pattern. So far, three CCM genes have been identified in the heritable disorders: CCM1/KRIT1 (Laberge et al., 1999), CCM2/malcavernin (Denier et al., 2004) and CCM3/programmed cell death 10 (PDCD10) (Bergametti et al., 2005), accounting for approximately 40%, 20% and 40% of the autosomal dominant disorders, respectively (Craig et al., 1998).

The CCM1 locus codes for a protein called KRIT1 that contains four ankyrin repeat domains and a FERM (protein 4.1, ezrin, radixin, moesin) domain that mediates its interaction with Rap1, a small GTPase that facilitates KRIT1 localization to cell-cell junctions. In Drosophila melanogaster, Rap1 is not required for cell proliferation and cell fate specification, but is required for normal tissue morphogenesis and cell shape maintenance (Asha et al., 1999). Rap1 is localized to the adherens junctions and is actively required for their even distribution; loss of Rap1 leads to disrupted epithelial cell behavior (Knox and Brown, 2002). KRIT1/CCM1 also binds CCM2, CCM3, and a variety of additional interacting proteins, and it is thought that these proteins all function together in large multiprotein complexes regulating cell-cell junction formation (Hilder et al., 2007).

Despite the identification of the defective genes for the three CCM loci, little is known about the molecular mechanisms underlying CCM lesion formation. Furthermore, not all individuals harboring defective CCM genes develop ICH, reflecting incomplete penetrance of these mutations and/or involvement of additional genetic modifiers predisposing to lesion formation (Lucas et al., 2003). It has been suggested that multiple genetic factors play important roles in predisposition to hemorrhagic stroke and influence the likelihood of ICH events in both familial and sporadic disorders. Recent evidence that CCM genes act together in common intracellular complexes and/or signaling pathways (Dupre et al., 2003; Hilder et al., 2007; Voss et al., 2007; Zawistowski et al., 2005) suggested to us that minor functional perturbations of different genes in this pathway might act together to precipitate ICH. However, demonstrating multigene association is not possible in the small number of available human hemorrhagic stroke pedigrees, nor has functional evidence substantiating this idea been readily available from other model organisms.

We sought to explore the possibility that very minor, otherwise silent deficits in genes that function together to regulate endothelial junctional integrity might synergistically initiate hemorrhagic stroke. We turned to the zebrafish, a useful model organism for analysis of human disease genes and their pathophysiology (Lieschke and Currie, 2007). Zebrafish embryos are readily amenable to simultaneous functional knockdown of multiple genes (Pham et al., 2007), making them very useful for functional analysis of multigene families or interacting proteins. In this study, we show that a combined minor reduction in the expression of multiple CCM pathway genes can precipitate hemorrhagic stroke. Our results provide a rationale for understanding the variability in appearance and onset of this disease.

We began our analysis by studying rap1b, a gene implicated in the CCM pathway in endothelial cells (Cullere et al., 2005; Glading et al., 2007; Kehrer-Sawatzki et al., 2002; Laberge-le Couteulx et al., 1999) but not yet characterized in zebrafish. We obtained full-length clones for two RAP1-related zebrafish genes, rap1a and rap1b. The rap1a gene is expressed ubiquitously during early development (data not shown), but rap1b shows vascular-enriched expression by in situ hybridization (Fig. 1A). To knockdown rap1b function in zebrafish, we designed morpholino oligonucleotides (modified antisense oligonucleotides) (Nasevicius and Ekker, 2000), targeting either the ATG initiator codon or the second exon-intron boundary, to interfere with rap1b translation or splicing, respectively. Embryos injected with either one of these morpholinos appeared morphologically normal up to 4 days postfertilization (dpf) and were indistinguishable from control morpholino-injected embryos (data not shown and supplementary material Fig. S1A,B). However, by 48 hours postfertilization (hpf) more than 70% of embryos injected with either of the rap1b morpholinos (n=380) developed ICH (Fig. 1E,F). Most hemorrhage foci were localized in, and around, the hindbrain ventricle (Fig. 1E,F; supplementary material Fig. S1). There was also a left-side bias, with 62% (n=25/40) of hindbrain hemorrhages in rap1b morpholino-injected fish occurring on the left side. The efficacy of the rap1b splice morpholino was verified by reverse transcription (RT)-PCR (supplementary material Fig. S2). We also confirmed that rap1b morpholino-injected animals do not have defects in proper patterning of the nervous system (supplementary material Fig. S3A,B), endothelial specification, or arterial-venous differentiation (supplementary material Fig. S3C-J). The specificity of the observed phenotypic effects of rap1b knockdown was further verified by co-injection of synthetic human RAP1B mRNA with the zebrafish rap1b morpholino. Although most of the injected RNA would probably have been degraded by the time hemorrhage began to appear at 2 dpf, only 44% of the co-injected embryos (n=72) manifested cranial hemorrhages compared with 74% of the embryos injected with rap1b ATG morpholino alone (n=127) (Fig. 1G). These results indicate that the rap1b morpholinos target rap1b and specifically abrogate its function, and suggest that the human and zebrafish Rap1b proteins have a conserved function. To investigate whether Rap1b is specifically required in endothelial cells for maintenance of vascular integrity, we carried out a rescue experiment using the zebrafish fli1 promoter (Villefranc et al., 2007) to drive endothelial expression of human Rap1b. When we co-injected a fli1:hRap1b DNA construct together with the zebrafish rap1b morpholino, cranial hemorrhage was reduced from 74% to only 30% of injected embryos, despite the mosaic expression of injected DNA (Fig. 1G). These results indicated that rap1b function is required in endothelial cells for vascular integrity.

We performed additional experiments to further explore the nature of the endothelial defects caused by Rap1b depletion. Ultrastructural examination of the cranial endothelium revealed normal well-formed endothelial junctions in control morpholino-injected animals (Fig. 2A). In contrast, most endothelial cell-cell junctions in rap1b morpholino-injected animals were either poorly formed or even absent, with open gaps observed between some endothelial cells (Fig. 2B-D), as also noted in human CCM patients (Clatterbuck et al., 2001). In order to further examine the role of Rap1 in endothelial junction formation, we used small interfering RNA (siRNA) to knockdown Rap1b in human umbilical vein endothelial cells (HUVECs) in vitro. We used Rap1b siRNA to reduce Rap1b and total Rap1 levels in HUVECs to 90% and 63% of normal, respectively (supplementary material Fig. S4A). Residual Rap1 failed to localize properly to junctions in Rap1b siRNA-treated HUVEC monolayers, similar to HUVECs treated with siRNA against vascular endothelial (VE)-cadherin or VE-cadherin null cells (supplementary material Fig. S4B-F). Transwell permeability tests showed increased permeability across Rap1b siRNA-treated endothelial monolayers, similar to endothelial monolayers treated with VE-cadherin siRNA (Fig. 2E). Furthermore, we observed loss or mislocalized staining of a variety of endothelial junction markers in Rap1b siRNA-treated monolayers, including the adherens junction markers VE-cadherin and β-catenin (Fig. 2F-I), and markers of proper tight junction formation (supplementary material Fig. S5A-F). These results indicate that Rap1 function is required for proper endothelial junction formation and maintenance both in vivo and in vitro.

We took advantage of the ability to perform rapid combinatorial knockdown of gene function in the zebrafish (Pham et al., 2007) to examine the synergistic effects of combined partial loss-of-function for multiple members of the CCM/Rap1 signaling cascade. Previous reports have shown that cool1/β-pix, a guanine nucleotide exchange factor, and Pak2a, a downstream p21-activated kinase, control directed cell migration (Cau and Hall, 2005; Schwamborn and Puschel, 2004) and vascular permeability in zebrafish embryos (Buchner et al., 2007; Liu et al., 2007), and potentially function in a signaling pathway downstream from ccm gene products and Rap1b. Like the rap1b gene, pak2a is also preferentially expressed in the zebrafish vasculature, and mutants or morpholino-injected animals for either pak2a or β-pix display intracranial bleeding (Buchner et al., 2007; Liu et al., 2007) (supplementary material Fig. S6). We titrated doses of the respective morpholinos for each of these genes to levels that resulted in hemorrhage in less than 10% of injected animals, and then injected pairwise combinations of the morpholinos. At these very low doses of the splice-targeting morpholinos, we could not detect a significant reduction in correctly spliced mRNA levels by semi-quantitative RT-PCR (supplementary material Fig. S7). Nevertheless, injection of any two morpholinos resulted in strong synergistic effects, with dramatically elevated numbers of animals experiencing ICH (Fig. 3A). For example, co-injecting doses of rap1b and pak2a morpholinos, which, when given individually, caused hemorrhage in only 3% and 8% of animals, respectively, resulted in 80% of animals developing ICH. To investigate whether functional interaction between rap1b and pak2a is specifically required in endothelial cells for maintenance of vascular integrity, we carried out another endothelial cell-specific rescue experiment. The zebrafish fli1 promoter was used to drive mosaic endothelial expression of pak2a. When we co-injected this fli1:pak2a DNA construct together with the zebrafish rap1b morpholino, ICH was reduced from 75% with morpholino alone to only 35% (Fig. 3B). These results indicate that excess Pak2a in endothelium can complement the defect in rap1b morphants. Co-injection of morpholinos against all three genes (rap1b, pak2a, β-pix), at doses of two-thirds of the amount used for the double morpholino injections, resulted in 86% of animals developing ICH (Fig. 3A). These results indicate that very minor deficits in multiple members of the Rap1b/Pak2a/β-pix pathway lead to unexpectedly dramatic hemorrhagic outcomes.

To examine whether combined partial functional reduction in a ccm gene product and Rap1b could precipitate hemorrhagic stroke, we used morpholinos to partially knockdown zebrafish ccm1 together with rap1b. Loss of CCM1 function in either mice (Whitehead et al., 2004) or zebrafish (Mably et al., 2006) has been previously shown to cause a defect in cardiac function, precluding analysis of potential vascular integrity defects. Co-injection of low doses of ccm1 and rap1b morpholinos resulted in synergistic enhancement of the cardiac phenotype, similar to previously reported combined partial knockdown of ccm1 and ccm2 (Mably et al., 2006). Interestingly, however, co-injection of even lower doses of ccm1 and rap1b morpholinos resulted in 36% of animals developing ICH, a phenotype not previously reported for knockdown of ccm1 in zebrafish (Fig. 4). To rule out the possibility that the observed effects were because of genetic cross-regulation rather than functional synergy between gene products, we carried out semi-quantitative RT-PCR for ccm1, pak2a, rap1b, β-pix and mapk1 on embryos injected with high doses of either ccm1 or rap1b morpholinos. We did not detect any changes in transcript levels of any of the genes, other than ccm1 or rap1b that were being directly targeted by their respective morpholinos (supplementary material Fig. S8).

Formation of seamless endothelial tubes is crucial for the development and function of blood vessels, and requires proper assembly of endothelial cell-cell junctions. Defects in the maintenance of endothelial junctions in cranial vessels lead to ICH, with devastating consequences. The RAP1 gene has been shown to be important for cadherin-mediated cell adhesion in other epithelial cell types, including Drosophila epithelial cells (Knox and Brown, 2002). Our findings demonstrate that Rap1b is essential for junctional integrity in endothelium. Depletion of Rap1b in cultured endothelial cells leads to defective endothelial cell-cell junctions and loss of barrier function. Rap1-depleted cultured endothelial monolayers display defective localization of markers of both adherens junctions and tight junctions, similar to defects in the distribution of adherence junctions that have been observed in Rap1-deficient Drosophila epithelial cells (Knox and Brown, 2002). Rap1b knockdown in zebrafish embryos leads to defective formation of endothelial junctions and ICH, with lesions appearing in similar anatomical locations to those commonly recognized for ICH lesions. Generally, in humans, stroke occurs more frequently on the left side of the brain than on the right side (Foerch et al., 2005), and we found that zebrafish embryos injected with rap1b morpholino also display slightly more hindbrain hemorrhages on the left side. Together, our results suggest that Rap1 has a conserved role in regulating cell-cell junction formation from invertebrates to higher vertebrates including humans.

Active Rap1 localizes to cell-cell contacts and physically interacts with Krit1/CCM1, and it has been proposed that Rap1 is important for recruiting Krit1 to junctions (Voss et al., 2007). In addition to showing that other junctional markers failed to localize properly, our results also demonstrate that residual Rap1 (from Rap1a) does not localize properly to cell-cell junctions in Rap1b siRNA-treated endothelial monolayers in culture, indicating that junctional targeting of Rap1 is dependent on its own function.

In addition to demonstrating that Rap1b is a key component in stabilizing cell-cell junctions in vivo, our results also confirm that it acts in a common molecular pathway with the ccm genes. In humans, hereditary CCM disorders are associated with mutations at the CCM1/KRIT1 (Laberge et al., 1999), CCM2/malcavernin (Denier et al., 2004) and CCM3/PDCD10 (Bergametti et al., 2005) loci. The familial CCM disorders all show dominant inheritance patterns with variable penetrance and expressivity, and virtually all CCM alleles are loss-of-function mutations (Felbor et al., 2006). However, the underlying basis for lesion formation in these disorders is still unclear. Haploinsufficiency has been documented in other human genetic vascular disorders including collagen 3A1 deficiency (Schwarze et al., 2001; Wenstrup et al., 2000), and it may be similarly important in CCM (Cave-Riant et al., 2002). One report documented a loss of heterozygosity by second mutation in the same CCM locus that was associated with a CCM lesion (Gault et al., 2005). However, it has been difficult to establish the factors that initiate most lesion formation in CCM and why the appearance of these lesions is so variable.

Second-site genetic mutations have also been suggested as a possible mechanism for initiating lesion formation in CCM (Gunel et al., 1996). Second-site genetic modifier(s) have been well documented for many other human genetic diseases including Huntington’s disease (Wexler et al., 2004) and sickle cell anemia (Sebastiani et al., 2005). In sickle cell anemia, 6–8% of affected individuals also show overt stroke, a phenotype caused by a genetic modifier from the TGF-β signaling pathway (Sebastiani et al., 2005). In the case of CCM, the proteins encoded by the three known disease loci have been shown to physically interact with one another (Voss et al., 2007; Zawistowski et al., 2005), and with additional proteins including Rap1 (Beraud-Dufour et al., 2007; Hilder et al., 2007; Serebriiskii et al., 1997). As noted above, Rap1 itself has been implicated in junctional localization of CCM1/KRIT1 (Voss et al., 2007). Based on these results it seemed possible that the phenotypic consequences of reduced levels of CCM proteins in human heterozygotes could be readily influenced by mutations in, or reduced levels of, some of these interacting proteins, and that these second-site defects could either trigger lesion formation or influence the penetrance, onset, or severity of the disease.

As a first test of this idea, we used the zebrafish to examine whether minimal deficits in some of these genes, which are individually phenotypically silent, could combine to yield defects in junction formation and hemorrhage. Highly conserved ccm1 and ccm2 orthologs are expressed in the zebrafish vasculature, and animals with homozygous null mutations in either of the genes (the santa and valentine mutants, respectively) develop enlarged nonfunctional hearts and lack blood circulation, as do animals with full-dose knockdowns in either gene (Mably et al., 2006), precluding analysis of potential vascular integrity defects. Similar cardiac defects have been reported in mice with homozygous targeted deletion of Ccm1/Krit1 (Whitehead et al., 2004). Combined partial knockdown of ccm1 and ccm2 by morpholinos also ‘phenocopies’ the cardiac phenotypes (Mably et al., 2006). However, our results show that very low-level depletion of ccm1 and rap1b synergistically enhances the cranial hemorrhage phenotype seen in rap1b morphants. Strongly synergistic hemorrhage defects are also seen when zebrafish embryos are injected with combinations of very low doses of morpholinos against other genes implicated in the CCM pathway, including rap1, ccm1, β-pix and pak2a.

It remains to be seen whether similar second-site defects in interacting genes influence lesion formation, or disease manifestation, in human patients heterozygous for CCM mutations. This is likely to be difficult to determine given the small patient population sizes and large number of potential interacting genes. Furthermore, the extraordinary level of synergy that we have documented in our zebrafish studies suggests that maintenance of junctional integrity is extremely sensitive to even minimal combined functional disruption. This may make it difficult to determine whether a second site polymorphism detected in human CCM patients is only a harmless polymorphism, or a significant genetic modifier of the disease. However, evidence that these genes interact closely, both physically and functionally, in junction formation and vascular integrity leads to the hope that approaches designed to modulate the downstream activity or function of the CCM pathway might have positive benefits for a large number of affected individuals.

Zebrafish lines used in the study are wild-type EK, Tg(fli1:EGFP)^y1 and Tg(flk:GFP); Tg(gata:DsRed). All lines were maintained as previously described (Westerfield, 2000).

Zebrafish rap1a and rap1b orthologs were identified based on identity searches to human genes. A rap1a cDNA fragment was amplified using primers rap1aF: AAAAACACACGCCTTCC-CGCCTTG and rap1aR: TGGAGTAAACCAGAGCAAAACCC, and cloned into pTOPO (Clontech). The zebrafish rap1b in situ clone was obtained from ZIRC (cb1026). In situ hybridization was carried out as previously described (Pham et al., 2007). A complete cDNA clone for human Rap1b was obtained from Open Biosystems and used for synthetic mRNA synthesis and for generation of an endothelial specific pTolfli1ep:human Rap1b construct (Villefranc et al., 2007). To detect axonal projections (pan-neuronal), a monoclonal antibody generated against acetylated tubulin (Sigma T6793) was used.

Morpholinos were injected at described doses into 1–2 cell stage zebrafish embryos. Morpholinos used in this analysis are listed below:

rap1b translation start site (start site underlined) MO 5′ GGACTACTAACTTGTATTCACGCAT 3′

rap1b exon 2 splice donor site MO 5′ AAATGATGCAGAACTTGCCTTTCTG 3′

rap1b mismatch (mismatches underlined) MO 5′ GGAGTAGTAACTTCTATTGACGGAT 3′

pak2a exon 8 splice donor site MO 5′ AATAGAGTACAACATACCTCTTTGG 3′

β-pix exon 6 splice donor site MO 5′ GCGCATCTCTCTTACCACATTATAG 3′

ccm1 exon 1 splice donor site MO 5′GCTTTATTTCACCTCACCTCATAGG 3′

Total cellular RNA from morpholino-injected embryos was extracted using Trizol reagent and treated with DNAase I. cDNA was synthesized using the ThermoScript kit (Invitrogen). Semi-quantitative PCR was performed using cDNA generated from equal amounts of RNA from different samples. The linear range of amplification cycles was determined for each target gene. Target and reference (β-actin) genes were amplified to the same cycle number for each sample. PCR primers used to analyze zebrafish genes are listed in the supplementary material.

Zebrafish embryos were fixed at room temperature in 2.5% glutaraldehyde (pH 7.3) made in 0.13 M sodium cacodylate buffer. Embryos were embedded in Epoxy resin and semi-thin 1 μm plastic sections were cut and stained with Toluidine Blue O stain. Thin sections were prepared on a Reichert-Jung Ultracut-E ultramicrotome. The grids were post-stained with uranyl acetate and lead citrate.

Human endothelial cells from umbilical cord vein (HUVEC) were seeded on gelatin-coated tissue culture vessels. They were cultured for 24 hours in MCDB 131 with 20% FCS, ECGS and heparin (as described in Lampugnani et al., 2006) and were transfected with siRNA (supplementary material).

Antibodies used were: Rap1 for immunofluorescence and western blot (rabbit polyclonal, code sc-65, Santa Cruz); Rap1b for western blot (rabbit, code 36E1, Cell Signaling); VE-cadherin for immunofluorescence and western blot (goat polyclonal, code sc-6458, Santa Cruz); KRIT1 for immunofluorescence and western blot (goat polyclonal, code sc-23997, Santa Cruz); beta-catenin (mouse monoclonal, clone 14, BD transduction laboratories); claudin 5 (rabbit polyclonal, kindly donated by Dr H. Wolburg, Institute of Pathology, University of Tubingen, Tubingen, Germany); cingulin (rabbit polyclonal, code 36–4401, Zymed); ZO-1 (rabbit polyclonal, code 61–7300, Zymed); vinculin (mouse monoclonal, clone h-VIN1, Sigma); and tubulin for western blot (mouse monoclonal, clone B-5-1-2, Sigma).

Immunocytochemistry, transwell permeability assays and western blots were carried out as previously described (Lampugnani et al., 2006). Details of the protocols can be found in the supplementary material.

We thank Drs I. Dawid, E. Tournier-Lasserve and K. Sampath for comments on the manuscript. We thank members of the Weinstein Lab for technical assistance and critical suggestions. This work was supported by the intramural program of the National Institute of Child Health and Human Development (B.M.W.), by the Associazione Italiana per la Ricerca sul Cancro and European Community (EUSTROKE consortium; M.G.L., E.D.), and by the Leducq Fondation (B.M.W., M.G.L., E.D.). Deposited in PMC for immediate release.