The guanine nucleotide exchange factor C3G is necessary for the formation of focal adhesions and vascular maturation

Cells receive cues from the extracellular matrix and from growth factor receptors to activate cellular functions such as cell division, cell migration, cell differentiation and cell survival. These signals can be transmitted through adaptor proteins and guanine nucleotide exchange factors to small GTPases. Small GTPases activate, through mediators, a number of downstream targets, including mitogen-activated protein kinase (MAPK) and Jun kinases. These signalling pathways are often perturbed in cancer cells causing excessive cell proliferation, abnormal cell migration and aberrant cell adhesive properties.

C3G is a guanine nucleotide exchange factor transmitting signals from the extracellular matrix or growth factors to members of the Ras family of GTPases. Specifically C3G has been shown to respond to B and T cell receptor activation (Reedquist et al.,1996; Smit et al.,1996), insulin and EGF (Okada and Pessin, 1997), integrin binding(Arai et al., 1999;Uemura and Griffin, 1999),erythropoietin and interleukin 3 (Nosaka et al., 1999), and interferon γ(Alsayed et al., 2000). Activation of C3G can lead to the activation of Jun kinase(Mochizuki et al., 2000;Tanaka et al., 1997) or MAP kinase (Nosaka et al., 1999). C3G stimulated guanine nucleotide exchange predominantly on Rap1 and, to a lesser extend, on R-Ras (Ichiba et al.,1997; Ohba et al.,2001). However, C3G-mediated activation of Jun kinase appears to occur solely through activation of R-Ras(Mochizuki et al., 2000). The rough eye phenotype caused by constitutive over-activation of DC3G in D. melanogaster in vivo can be suppressed by reducing the gene dosage ofRas1, rolled (MAPK) and Rap1(Ishimaru et al., 1999). Taken together, these reports suggest an activation of both, the Ras/Rap/MAPK-pathway and the Ras/Jun kinase-pathway, by C3G.

C3G was initially isolated as a protein binding to the amino-terminal SH3 domain of the adaptor protein, chicken retrovirus kinase (Crk)(Knudsen et al., 1994;Tanaka et al., 1994). C3G contains a carboxy-terminal CDC25-type catalytic domain stimulating the guanine nucleotide exchange on Ras-family members(Tanaka et al., 1994). Centrally located are four proline-rich domains that confer binding to the amino-terminal SH3 domain of Crk and the adaptor protein Grb2(Kirsch et al., 1998;Tanaka et al., 1994). Amino-terminal to these is one proline-rich region that confers binding to the SH3 domain of the focal adhesion molecule p130^Cas(Kirsch et al., 1998).

Crk and p130^Cas have been shown to be integral parts of focal adhesions. They form a complex with the docking protein paxillin and link filamentous actin of the cytoskeleton to integrin receptors (reviewed byTurner, 2000;Schaller, 2001). Crk, CrkL and Grb2 have been shown to enhance C3G activity(Ichiba et al., 1997). Crk does so by recruiting C3G to the cytoplasmic membrane. The SH2 domain of Crk that confers binding to paxillin or the adaptor protein Shc is essential for membrane recruitment of C3G (Ichiba et al., 1997). Paxillin, in turn, binds to integrins via src and focal adhesion kinase (reviewed bySchaller, 2001).

A recently reported mutation of the C3G (Grf2) gene in mice resulted in early post-implantation lethality thereby excluding the study of the role of C3G later in development(Ohba et al., 2001). We have generated a mouse strain carrying a hypomorphic C3G allele,C3G^gt. Here we reveal three previously unreported functions of C3G. Firstly, C3G is required for the formation or stabilisation of integrin β1- and paxillin-positive focal adhesions. Secondly, C3G is necessary for a normal response to PDGF. Thirdly, C3G is crucial for vascular maturation. The observed defects in cell adhesion and PDGF-response may explain the blood vessel development defect.

The promoter-less reporter construct pGT1.8geo [kindly provided by W. Skarnes (Skarnes et al.,1995)] was electroporated into the parental murine embryonic stem(ES) cell line MPI-II (Voss et al.,1997) as described previously(Voss et al., 1998b).

Total RNA was isolated from ES cells heterozygous for the insertion ofpGT1.8geo into the murine genome (clone F82). 5′ rapid amplification of cDNA ends was performed as described previously(Voss et al., 1998b) using oligonucleotides complementary to the lacZ gene of pGT1.8geoas gene-specific primers. RACE products were cloned into pGemT and sequenced. The gene-specific sequence obtained was used to screen databases. To identify the site of integration within the C3G locus a series of probes were used for Southern analysis including several probes cloned by PCR from intron 1. These probes were generated through the use of the Celera Discovery System and Celera Genomics' associated databases.

To test for the presence of C3G mRNA 3′ of the gene trap insertion site, RT-PCR was performed as described previously(Voss et al., 1998b) on total RNA isolated from homozygous embryos using oligo(dTTP) for the reverse transcriptase reaction and using 5′ TGT TTT CCG CCT TGT TAT GTT CCT 3′ as the forward and 5′ AGT GGC AGC GGC AGA CCT CAG 3′ as the reverse primer. This reaction generated a product complementary to bases 268-647 of a mouse EST (GenBank accession no. BG916265) spanning exons 21-24 of the C3G locus. To test for the presence of full-length protein coding mRNA in homozygous embryos, RT-PCR representing all coding exons was performed using primers in exon 21 (5′ TGT TTT CCG CCT TGT TAT GTT CCT 3′) and exon 1 (5′ GCCCGGAAATGTCCGGCAAGATCG 3′).

Chimeras were produced as describes previously(Voss et al., 1998b) and mated to wild-type females of the mouse strains C57B1/6 and Random Swiss. In timed pregnant female mice, noon of the day on which the vaginal plug was observed was termed day 0.5 of gestation (E0.5). Genotyping was carried out using DNA isolated from tail biopsies or embryonic membranes by C3Ggene-specific Southern analysis using an [α-³²P]dATP-labelled probe near the insertion site of pGT1.8geo in intron 1 (probe 1,Fig. 1A). Probe 1 comprised 203 bp sequence located 22 kb 3′ of exon 1. It was PCR generated using 5′ GGA GCA GGC AGG ACT CCT CAC T 3′ and 5′ TCT CCC ATC TAA GAG GCT CTG G 3′ as primers.

Embryos were dissected from the uteri, placed in phosphate-buffered saline(PBS), viewed and photographed using a low-magnification stereomicroscope(Zeiss) and a digital camera (Axiocam, Zeiss). Embryonic membranes were used for genotyping by C3G-specific Southern analysis. For histological analysis, embryos were fixed in 4% paraformaldehyde, dehydrated, infiltrated and embedded in paraffin. 8 μm serial sections were cut, deparaffinised and stained with Haematoxylin and Eosin using standard techniques. Slides were viewed and photographed using a compound microscope (Zeiss) and a digital camera. Photographs of mutant and control embryos were compared in detail.

A lacZ- or C3G-specific[α-³²P]dATP-labelled probe 3′ of the insertion site(probe 2, Fig. 1A) was used for northern analysis of total RNA isolated from embryos, adult tissues and cell lines and, in vitro transcribed and labelled with ³⁵S-labelled UTPαS, for in situ hybridisation as described previously(Voss et al., 2000). Probe 2 was PCR generated using 5′ CGC CCC TGC TTC TCA ATG GTT AGC C 3′and 5′ CCA TCC AAG AAG GGA AAG CCA GCC G 3′ and was complementary to 431 bases spanning exons 2-5 of the C3G locus. Probe 2 was cloned into pBluescript KS+ and sequenced.

Lysates of primary murine embryonic fibroblasts (MEFs) isolated from E10.5 embryos of heterozygous intercrosses were separated by 8% to 16% SDS-PAGE,electrotransferred to membrane (Immobilon P), blocked in 1% BSA, 0.1% Tween 20 PBS, incubated with primary rabbit polyclonal antibody (1:200 or 1:1000, Santa Cruz), washed, incubated with secondary horseradish peroxidase-conjugated antibody (1:3000 to 1:20000, Biorad), washed and detected by chemiluminescence(Pierce). Densitometry was carried out using a computing densitometer(Molecular Dynamics).

Horseradish peroxidase-immunohistochemistry was performed as described previously (Thomas et al.,2000). In addition, horseradish peroxidase-immunohistochemistry was performed on cryosections for the detection of platelet and endothelial cell adhesion molecule 1 (PECAM 1). Fluorescence immunohistochemistry was carried out using standard techniques. In brief, after treatment with 0.25%gelatine and 0.1% Triton X-100 in PBS paraformaldehyde-fixed or acetone-fixed cryosections (PECAM 1 and smooth muscle α-actin (SMαA) staining)or deparaffinised sections (SMαA and all other antigens) were treated with intervening PBS washes with first antibody solution, second fluorescent antibody solution and then counter-stained with 0.1 μg/ml bis-benzimide in 0.25% gelatine in PBS (Molecular Probes). The sections were then mounted with Mowiol, viewed and photographed using a fluorescence microscope (Zeiss) and a digital camera. Immunocytochemistry was carried out as described previously(Voss et al., 2000).

The first antibodies and dilutions used were mouse anti-smooth muscleα-actin (SMαA; Sigma, 1:400), rat anti-mouse CD31 (PECAM 1;Pharmingen, 1:200 for cryosections and 1:20 for whole mounts), rabbit anti-nidogen (kindly provided by M. Dziadek, 1:200), biotinylated hamster anti-rat CD29 (Integrin β1 chain, Pharmingen, 1:250), mouse anti-chick paxillin (BD Transduction, 1:250), rabbit anti-human integrin β3(Chemicon, 1:20), mouse anti-chicken vinculin (Sigma, 1:50). Secondary antibodies and streptavidin used were biotinylated goat anti-rat (Vector,1:150), biotinylated horse anti-mouse and rabbit (Vector, 1:50), Alexa Fluor 546 goat anti-mouse (Molecular Probes, 1:1000), Alexa Fluor 488 goat anti-mouse (Molecular Probes, 1:1000), TRITC goat anti-rabbit (Jackson,1:200), TRITC-streptavidin (Southern Biotechnology, 1:50).

Primary murine embryonic fibroblasts (MEFs) were isolated essentially as described previously (Voss and Thomas,2001) omitting the glass bead dissociation step. Briefly, E10.5 embryos were transferred individually into 250 μl of trypsin/EDTA solution,incubated for 5 minutes at 37°C followed by mechanical dissociation by pipetting and plated into 6-well tissue culture plates in Dulbecco's minimum essential medium with 10% foetal bovine serum (DMEM 10% FBS). The cells were used at passage 3 and 4. Ninety-six-well plates or cover slip inserts for 24-well plates were coated for 1 hour with 10 μg/ml human plasma fibronectin (Roche), 10 μg/ml gelatine (Sigma) or 10 μg/ml natural mouse laminin (Life Technology). Cells were plated at subconfluent density (20,000 cells per well, 24-well plate) with or without 0.1% or 10% FBS.

For PDGF treatment, cells were serum starved for 16 hours and then treated with or without 2 ng/ml or 10 ng/ml platelet-derived growth factor-AA(PDGF-AA) or PDGF-BB (both recombinant human; Pepro Tech). Numbers of actin rings were counted per treatment in more than 500 cells per cultures isolated from 3 C3G^gt/gt homozygous and 3 wild-type embryos in duplicates and compared by χ² test. Results are given as mean± standard error of the mean with P values.

For adhesion assays 7,000 and 14,000 cells were plated into 96-well plates. After 1 hour or 2 hours, non-adherent cells were washed off twice with PBS and cells were fixed with 3% paraformaldehyde, 2% sucrose at room temperature for 5 minutes. Adherent cells were stained with 0.5% Crystal Violet in 20%methanol. Plates were washed in H₂O and air-dried. Stained adherent cells were solubilised in 0.1 M citric acid in 50% ethanol, pH 4.2. The optical absorbance was determined at 570 nm using an ELISA reader(SpectraFluor Plus, Tecan). Results were compared by two-factor analysis of variance with genotype and animal as the two factors. Results are given as mean ± standard error of the mean with P values.

Cell migration assays were performed as described previously(Lallemand et al., 1998). In brief, a rectangular wound was created in monolayers of cells on the day they became confluent using a 20-200 μl micropipette tip. Dislodged cells were washed off and medium was replaced. Images of the monolayer wounds were taken at the time of wounding (0 hours) and 7 hours later. The area free of cells was measured using image analysis software (NIH Image 1.62). The area covered by cells within 7 hours after wounding was calculated. Results were compared by two-factor analysis of variance with genotype and experiment as the two factors. Results are given as means ± standard error of the mean withP values.

We generated a mutant C3G allele, C3G^gt, in a gene trap screen in murine embryonic stem cells(Voss et al., 1998b). 5′RACE of RNA isolated from clone F82 revealed the fusion of coding sequences of the gene trap construct pGT1.8geo(Skarnes et al., 1995) to exon 1 (Fig. 1A) of the murineC3G locus (base 297 of the sequence with GenBank accession no. NM_504050) (Zhai et al.,2001). The protein deduced from this fusion mRNA contained the first 19 amino acids of C3G fused to β-galactosidase and neomycin phosphotransferase (Fig. 1A). This fusion protein lacked the carboxy-terminal CDC25-type catalytic domain,the 4 central Crk SH3-binding domains and the central p130^CasSH3-binding domain. These domains have been shown to be essential for C3G function (Kirsch et al.,1998). No functional domains have been identified within the first 19 amino acids of C3G (Ichiba et al.,1999). Therefore no abnormal function is to be expected from this fusion protein.

The gene trap insertion was located in the 59 kb long intron 1,approximately 16 kb 3′ of exon 1(Fig. 1A). Southern analysis ofEcoRV-digested wild-type, heterozygous and homozygous DNA showed that the insertion of the gene trap construct caused a shift in the size of the third EcoRV fragment of intron 1 (14830 bases to 26997 bases from exon 1, Fig. 1B). Southern analysis of the 5′ and 3′ adjacent EcoRV fragments of intron 1 revealed no polymorphism in heterozygous or homozygous DNA,confirming the absence of rearrangements of the C3G locus outside of the gene trap insertion site. Loss of sequences 5′ or 3′ of the gene trap insertion was ruled out by 5′RACE (see above) and RT-PCR(below).

Northern analysis of total RNA isolated from E10.5 embryos wild type,heterozygous or homozygous for the C3G^gt mutant allele showed C3G/β-gal/neo fusion mRNA of the expected size of about 4.7 kb and no detectable mRNA 3′ of the gene trap insertion site(Fig. 2A). Likewise radioactive in situ hybridisation on homozygous and wild-type E11.5 embryos did not reveal mRNA 3′ of the gene trap insertion (compareFig. 3K and L with I and J). However, limiting dilution RT-PCR revealed C3G mRNA 3′ of the gene trap insertion in homozygous mutant embryos at less than 1% of wild-type levels (Fig. 2B). This 3′mRNA was generated presumably by splicing around the gene trap insertion in intron 1. RT-PCR using primers to exons 21 and 1 revealed the presence of C3G mRNA containing all coding exons in homozygous mutant embryos(Fig. 2C). We have documented this form of splicing for one of our other gene trap mouse strains previously(Voss et al., 1998a). Finally,western analysis of total primary embryonic fibroblast lysates showed two prominent C3G bands in wild-type and heterozygous lysates. Instead of these,two weak bands were visible in homozygous lysate(Fig. 2D). Densitometrically these amounted to less than 5% of the normal amount of C3G protein. In conclusion, the C3G^gt allele generates less than 1% mRNA containing all coding exons and no more than 5% C3G protein and, therefore, is a hypomorphic rather than a null allele.

It was previously reported that the C3G gene is expressed ubiquitously in Drosophila melanogaster(Ishimaru et al., 1999) and in human foetal and adult tissues (Tanaka et al., 1994). Northern analysis of adult mouse tissues showed thatC3G is expressed at low levels ubiquitously, with higher levels of expression in testes (not shown). The activity pattern of theβ-galactosidase reporter inserted into the C3G locus(Fig. 3A-D) and radioactive in situ hybridisation of sections of E11.5 and E12.5 embryos confirmed this result (Fig. 3E-L). Notably,reporter activity and expression of the endogenous locus were observed in all cells including vascular endothelial cells and blood vessel surrounding cell,both in large (Fig. 3I) and small blood vessels (Fig. 3D,J).

We did not recover any homozygous C3G^gt/gt mutant offspring from heterozygous inter-crosses at weaning. Among 361 embryos (36 litters) recovered between E9.5 and E14.5, 26% were homozygous, 49%heterozygous for the C3G^gt allele and 25% wild-type at theC3G locus (Table 1 andFig. 1B). Nine percent of the implanted embryos were resorbed at an early stage and could not be genotyped. Given the Mendelian distribution of the C3G^gt allele,these early resorptions are likely to involve all three genotypes. The majority of the homozygotes recovered at E9.5 were phenotypically normal(Table 2). At E10.5, one third of the homozygotes showed phenotypic abnormalities or were dead. At E11.5 four fifths were abnormal or dead. At E13.5 and 14.5 all homozygotes were either dead or highly abnormal.

The majority of the C3G^gt/gt mutant embryos died between E10.5 and E11.5 exhibiting haemorrhage into the lumen of the neural tube (Fig. 4A-C). Mutant embryos dying between E13.5 and E14.5 showed massive subcutaneous oedema or haemorrhagic oedema (Fig. 4D-F). Most likely hypovolaemic cardiovascular failure was the cause of death. The vascular system defects are the subject of this report. Apart from the defects observed in the vascular system the homozygous embryos also showed defects in the nervous system (A. K. V. and T. T.,unpublished).

Histological analysis of serial sections of 19 homozygous embryos and 19 controls at E9.5, E10.5 and E11.5 revealed that haemorrhage initiated from small cephalic blood vessels most often in the vicinity of the hindbrain neural epithelium (Fig. 4G-I),occasionally near the olfactory epithelium. Small blood vessels burst leading to an accumulation of blood in the interstitial space until the neural epithelium gave way and ruptured. The blood then drained into the lumen of the neural tube, probably causing hypovolaemia leading to death.

Blood vessel maturation occurs in mice between E10.5 and E11.5(Takahashi et al., 1996). Blood vessel maturation involves the recruitment of supporting cells of mesenchymal and other origin that differentiate into pericytes and smooth muscle cells (reviewed by Carmeliet,2000). This process lends mechanical stability to blood vessels.

In order to test if the fragility in C3G^gt/gt blood vessels was due to a lack of blood vessel maturation we examined the expression of smooth muscle α-actin (SMαA) in 8 pairs of C3G homozygous mutant embryos and littermate controls at E10.5 and E11.5. We found few or no SMαA-positive cells around small cephalic blood vessels in homozygous mutant embryos (compare Fig. 5A with B and C with D). Those few SMαA-positive cells that were found did not form a tight ring around the blood vessels as seen in the controls. Rather they were loosely associated with the blood vessels and with their environment. They had a rounded appearance with minimal contacts to neighbouring cells. Larger blood vessels, e.g. the dorsal aorta or the basilar artery, did acquire SMαA-positive cells. However, the coating of the larger blood vessels was less complete in C3G mutant embryos than in controls(basilar artery in Fig. 5G compared to H). Significantly, the dorsal aorta was deficient in SMαA-staining cells in its dorsal aspects which are known to develop blood vessel supporting cells later than the ventral aspects(Takahashi et al., 1996). The mutant vascular endothelial cells expressed platelet and endothelial cell adhesion molecule 1 (PECAM 1) normally and secreted nidogen normally(Fig. 5E,F,I,J,K,L). Structure and arboration of blood vessels appeared normal in embryos stained for PECAM 1 as whole mounts (not shown). These data suggested that the primary defect inC3G mutant vasculature was localised to vascular supporting cells. Vascular endothelial cells, in contrast, appeared to be functioning normally.

The initial stimulus for recruitment of vascular supporting cells has been proposed to be an increase in shear force experienced by blood vessel endothelial cells due to a gradual increase in blood pressure and flow as the embryos grow and their heart function becomes more efficient. It has been hypothesised that, in response to the increase in shear force, the endothelial cells release PDGF, which signals to perivascular cells(Risau, 1997). These then migrate to and along the blood vessels and differentiate into supporting cells. It appears that PDGF-BB is required for recruitment of pericytes to capillaries, but is not required for recruitment of PDGFRβ-positive cells to large blood vessels (Leveen et al.,1994; Lindahl et al.,1997).

In order to study in more detail the possible phenotypic aberration of mesenchymal cells, one source of blood vessel supporting cells, we isolated primary murine embryonic fibroblasts (MEFs) from 3C3G^gt/gt and 3 wild-type littermate control E10.5 embryos. We exposed them to PDGF, which is one of the signals that mesenchymal cells surrounding blood vessels would receive from endothelial cells in vivo. Immortilised fibroblast cell lines have been reported to respond to PDGF-BB with a prominent reorganisation of actin filaments into actin rings. The mutant MEFs showed a much higher frequency of actin ring formation than control cells in response to PDGF-BB, but not to PDGF-AA (compare(Fig. 6A,B with C and D). Mutant MEFs showed not only one actin ring more frequently, they also often formed multiple rings per cell (Fig. 6B) or actin rings spanning the whole perimeter of the cell(Fig. 6A). In contrast, the majority of control MEFs did not form actin rings. In 2% of these cells one small actin ring was seen (Fig. 6C). After exposure to 10 ng/ml PDGF-BB, actin rings occurred with a frequency of 0.33 per cell in mutant MEFs versus 0.02 in control MEFs(P<0.001; Fig. 6D). The frequency of actin rings in mutant and control MEFs in the absence of PDGF-BB was low and not significantly different (0.01 per cell compared with 0.002 per cell). PDGF-AA had no effect on MEFs of either genotype in this system. The frequencies of actin ring formation in response to PDGF-BB observed in this study in primary embryonic fibroblasts were consistently lower than those observed in immortalised fibroblast cell lines. The frequency of actin rings has been reported to vary between cell lines and to be lower in fibroblast cell lines established from isolates obtained prenatally than postnatally (Hedberg et al.,1993). In conclusion, the aberrant response of C3G mutant fibroblasts to PDGF-BB may be one of the mechanisms involved in the pathogenesis of the abnormal blood vessel phenotype.

In C3G mutant small cephalic blood vessels, the few remaining SMαA-positive cells appeared to be only loosely attached to their environment (see above). To examine the formation of cell adhesions inC3G mutant cells we stained mutant and control MEFs for paxillin and vinculin. The number and the size of paxillin-positive focal adhesions were greatly reduced in C3G mutant MEFs(Fig. 6E compared with F). Likewise, the number and size of vinculin-positive focal adhesions was reduced(Fig. 6G compared with H). The fact that the mutant cells generated a few small paxillin-positive focal points may be explained by the presence of residual wild-type C3G generated from the mutant allele. Alternatively, C3G may be important in the stabilisation of paxillin-positive focal adhesions rather than the initial establishment. Paxillin does not bind directly to C3G. Rather they both bind to the adaptor protein Crk. Paxillin binds to the SH2 domain and C3G to the amino-terminal SH3 domain of Crk (Senechal et al., 1998). Therefore, the paucity of paxillin-positive focal adhesions in C3G^gt/gt mutant MEFs suggests that the C3G molecule may be essential for the assembly of the protein complex containing,among other proteins, C3G, Crk and paxillin. Alternatively, signalling through C3G and activation of downstream cascades may be important for the stabilisation of focal adhesions.

To study the effects of the C3G^gt mutation on cell adhesion, homozygous mutant and control MEFs were plated onto laminin,gelatine or fibronectin in the absence of serum or onto gelatine or fibronectin in the presence of 10% serum. The plating efficiency ofC3G^gt/gt mutant was significantly reduced on laminin and gelatine with and without serum (Fig. 7A). In contrast, cell adhesion to fibronectin was not affected by the C3G mutation. Over a period of 2 days the majority of the mutant cells grown on laminin or gelatine in the absence of serum rounded up and lifted off the plate. However, the mutant cells grown on fibronectin remained attached, although they did not spread in a normal manner (without serum shown in Fig. 7B). These data suggested that C3G was required for binding to specific components of the extracellular matrix and for cell spreading in general.

In monolayer wounding assays of isolates of primary fibroblasts from 3 C3G mutant E10.5 embryos and 3 littermate control embryos we observed a significant increase in cell migration activity in the mutant cells(Fig. 7C,D). An increase in cell motility had been observed previously in C3G-deficient fibroblasts(Ohba et al., 2001). The requirement of C3G for cell adhesion and the regulation of cell migration may not be unrelated events.

In order to explore the effects of the C3G^gt mutation on integrin expression or distribution we stained mutant and control MEFs for integrin β1 and paxillin (Figs8 and9) or integrin β3 and paxillin (Fig. 9). Control cells showed a punctate staining for integrin β1(Fig. 8B,H;Fig. 9F) and β3(Fig. 9B,D). Both integrinβ1- and β3-positive foci coincided with paxillin staining(Fig. 8F,L;Fig. 9K; and data not shown). The integrin β1 foci were large and sparse, whereas the integrin β3 foci were more numerous and generally smaller.C3G^gt/gt mutant cells exhibited a marked reduction in the number of integrin β1 foci(Fig. 8A compared with B, and G with H). In contrast integrin β3 foci appeared to be unaffected by the C3G^gt mutation(Fig. 9A-D). Integrin β1 and paxillin double staining visualised the highly abnormal morphology ofC3G^gt/gt mutant cells in the absence of serum,particularly on laminin (Fig. 8E) and gelatine (not shown) and, to a lesser extend, on fibronectin (Fig. 8K). Integrinβ1 immunostaining was markedly abnormal in the mutant cells. Instead of the punctate staining distributed evenly across the cell as seen in the control cells, integrin β1 appeared to form abnormal aggregates usually near the nucleus. Abnormal aggregates of integrin β1 were most pronounced on laminin (Fig. 8A), but also obvious on gelatine (not shown) and fibronectin(Fig. 8G).

To enable us to examine C3G^gt/gt mutant cells in a healthier state we grew cells for 2 days on gelatine in the presence of 10% FBS before starving them in 0.1% serum overnight. In addition, we grew cells in 10% serum on tissue culture plastic before plating them for 40 minutes onto fibronectin in the presence of 10% serum. In both cases the cells attached, spread more efficiently and exhibited a healthier morphology than without serum. However, integrin β1 immunostaining was abnormal even under these conditions. Instead of the evenly distributed punctate staining seen in the control cells (Fig. 9F), mutant cells exhibited a few or no integrin β1-positive foci (Fig. 9E). The few integrin β1-positive foci sometimes coincided with paxillin staining in the C3G mutant cells (Fig. 9J compared with K). These data suggested that C3G was instrumental in the assembly or stabilisation of integrin β1 clusters. Although adhesion to laminin was most severely affected, adhesion and spreading on other substrates was also abnormal. The observed abnormalities indicated that C3G is important as an integral part of the complex formation around integrinβ1, but not integrin β3. Alternatively, signalling through C3G may be important for the stabilisation of integrin β1 clusters.

We have shown that C3G was essential for fibroblast cell adhesion, for the formation of paxillin- and integrin β1-, but not integrinβ3-positive focal adhesions, for a normal response to PDGF and for blood vessel maturation. Migration or differentiation (or both) of the perivascular cells was disturbed by the C3G mutation. This led to haemorrhages in the majority of the cases (80%). In the remaining 20% it led to disrupted vascular integrity with resulting oedema. The difference in the C3Gmutant phenotype observed by us and by Ohba and co-workers is probably attributable to the ameliorating effect of the less than 1% residual normal mRNA observed in our C3G^gt/gt embryos. Expression of normal mRNA at only 5% of the normal level has been shown previously to lead to a disproportionally large correction of a mutant phenotype(Dorin et al., 1996).

Our hypothesis is that in C3G mutants PDGF signals from blood vessel endothelial cells did not elicit a normal response in perivascular cells. PDGF-B mutant embryos have been reported to fail to develop normal numbers of vascular smooth muscle cells in microvasculature(Lindahl et al., 1997), whereC3G^gt/gt mutants also show defects in smooth muscle cells. In contrast, larger blood vessels develop smooth muscle cells at least to some extent in PDGF-B, PDGF-Rβ and C3G mutant embryos(Leveen et al., 1994;Lindahl et al., 1997;Soriano, 1994) (and our data). However, the C3G mutant phenotype manifests itself earlier and shows a wider variety of defects than mutations affecting the PDGF signalling pathway only. This is presumably due to the fact that C3G is required for other signalling pathways, such as integrin signalling.

Integrins mediate interactions between cells and extracellular matrix components such as laminin and fibronectin. Integrins are heterodimers consisting of an α and a β subunit. There are a large number ofα and β subunits, many of which bind multiple substrates(Hynes, 1992). Cell spreading requires the cytoplasmic domain of the β subunits(Berrier et al., 2000;Ylanne et al., 1993) and involves activation of the Ras signalling pathway(Berrier et al., 2000). In this context the requirement of C3G specifically for integrin β1 focal points was particularly interesting. In contrast to integrin β1, integrinβ3 foci formed without or with very little C3G. Although integrin β1 is normally involved, adhesion to fibronectin can be mediated by integrinβ3. In contrast, adhesion to laminin and collagen relies on integrinβ1 for the β subunit of the integrin heterodimers(Hynes, 1992). Accordingly,C3G mutant cells, which form integrin β3-positive cell adhesions, but are deficient in integrin β1-positive cell adhesions,adhere to fibronectin, but only poorly to laminin or gelatine.

3T3 cell adhesion to fibronectin, laminin, collagen I and vitronectin results in CrkL-mediated C3G phosphorylation and, presumably, activation(de Jong et al., 1998). Adhesion of other cell types is at least partly mediated by C3G(Arai et al., 2001;Arai et al., 1999). While these studies used cell adhesion to extracellular matrix components as their end point, we showed here that C3G was crucial for the assembly or stabilisation of integrin-positive focal adhesions. Moreover, we showed that lack of C3G affected specifically integrin β1, and not integrin β3. Interestingly, the reported phenotype of C3G null mutant mice(Ohba et al., 2001) is similar to that of integrin β1 null mutants(Fassler and Meyer, 1995;Stephens et al., 1995). Taking our data into consideration it is possible that the peri-implantation lethality of C3G null mutants may in part be caused by a deficiency in integrin β1-mediated cell adhesion.

Focal adhesions are large extracellular matrix-induced clusters of integrins linked to the ends of large actin bundles. One of the proteins making the link is paxillin (Turner,2000; Schaller,2001). C3G deficiency caused a greatly reduced number of much smaller paxillin-positive focal adhesions. The reduction in focal adhesions is most likely the cause of reduced cell attachment properties and failure of cell spreading. Previously, it had been shown that C3G binds to Crk and it was known that Crk in turn binds to paxillin(Knudsen et al., 1994;Salgia et al., 1995;Schaller, 2001;Tanaka et al., 1994). We showed here for the first time that C3G is crucial for the formation and/or re-enforcement of focal adhesions. Lack of focal adhesions has been observed in fibroblasts expressing a negative regulator of Src. Consistent with our finding, over-expression of C3G reverses this phenotype (Li, 2002). This together with our data suggests that signalling through C3G rather than the C3G molecule itself is important for the formation and/or stabilisation of focal adhesions.

The loss-of-function phenotype observed in our study shows considerable overlap with mutant phenotypes caused by loss of function of other members of the same signalling pathway. Besides activation through Crk-C3G, members of the Ras family of GTPases can be activated through the phosphotyrosine docking molecule ShcA, the adaptor protein Grb2 and the guanine nucleotide exchange factors Sos1 and Sos2. Embryos lacking ShcA succumb to cardiovascular defects between E10.5 and E11.5 (Lai and Pawson,2000). Similar to C3G mutants, in ShcA mutant embryos the mature vasculature is not established efficiently. Reduced staining for smooth muscle α-actin and impaired cell-cell and cell-extracellular matrix adhesions were observed in the ShcA mutant embryos (Lai and Pawson,2000). The loss-of-function phenotype of Sos1 was reported as cardiovascular defects, haemorrhage and death between E11.5 and E12.5 (Wang et al., 1997) or as death due to placental defects between E9.5 and E11.5(Qian et al., 2000).Sos2 mutant mice are viable(Esteban et al., 2000). Similarities between the loss-of-function phenotypes of the Crk-C3G and the Shc-Sos aspects of the signalling pathway were somewhat unexpected. These two aspects of the Ras-signalling pathway had been observed to act antagonistically in vitro (Okada et al.,1998). In contrast, the observed similarities in the loss-of-function phenotypes would suggest synergistic action, at least with respect to blood vessel maturation.

Taken together our data show that C3G is essential for normal PDGF signalling and for normal response to the extracellular matrix environment. C3G is critical for the process of vascular myogenesis. These results place C3G in a central role for cell-cell and cell-matrix interactions during blood vessel maturation.

We would like to thank W. Skarnes for providing the pGT1.8geo gene trap construct and M. Dziadek for the anti-nidogen antisera. We gratefully appreciate the excellent technical assistance of C. Cowled, S. Mihajlovic, E. Loza and A. Conrad. We thank V. Foletta and H. Cooper for fruitful discussions. This project was funded by the Walter and Eliza Hall Institute,the Max Planck Society and Amgen Inc.