1274 Research Article

IntroductionCerebral cavernous malformations (CCMs) affect over 0.5% of thepopulation (Labauge et al., 2007). They are vascular lesions in thebrain characterized by enlarged vessels (caverns) lined byendothelium, with no underlying smooth muscle or elastic tissue(Moriarity et al., 1999). Bleeding occurs frequently in these lesions,and can lead to headaches, seizures, focal neurological deficits andstroke (Moriarity et al., 1999; Zabramski et al., 1999).

The majority of CCMs are sporadic, but a significant minorityhave a genetic basis (Labauge et al., 2007). Three different CCMloci have been identified: CCM1/KRIT1 (Laberge-le Couteulx etal., 1999), CCM2/OSM (Liquori et al., 2003) and CCM3/PDCD10(Bergametti et al., 2005). Mutations are inherited in heterozygosisand loss of the wild-type allele in endothelial cells induces thedevelopment of the malformations (Akers et al., 2009; Pagenstecheret al., 2009). CCM1 and CCM2 are important in endothelial cellbiology and vascular development (Hogan et al., 2008; Kleavelandet al., 2009; Whitehead et al., 2004; Whitehead et al., 2009).Biochemically, CCM2 is known to balance the activities of the smallG proteins RhoA and cdc42 (Whitehead et al., 2009), probablythrough directed degradation of RhoA (Crose et al., 2009), whereasboth CCM1 and CCM2 have been shown to form a complex thatbinds to the type I transmembrane receptor heart of glass(Kleaveland et al., 2009).

CCM3 has also been found in the CCM1-CCM2 complex, albeitonly when one of the partners is overexpressed (Hilder et al., 2007;Stahl et al., 2008). Likewise, CCM3 has been found complexed tooverexpressed phosphoprotein phosphatase type 2A (Goudreault etal., 2008), although the functional consequences of this interactionare not known.

The germinal center kinase III (GCKIII) subfamily of proteinsconsists of SOK1 (also referred to as YSK1 and STK25), Mst3(also known as STK24) and Mst4 in mammals (Pombo et al., 2007;Ling et al., 2008). They belong to the Ste20 family of proteins,a large group of kinases that are characterized by a high degreeof homology in their catalytic domain (Dan et al., 2001). TheGCKIII kinases are involved in two important cellular processes:modulation of cell death and proliferation (Dan et al., 2002; Huanget al., 2002; Nogueira et al., 2008; Zhou et al., 2009), andregulation of cell migration (Lu et al., 2006; Preisinger et al.,2004). Specifically, Mst4 and SOK1 have been shown to associatewith the cis-Golgi matrix protein GM130, where they play a rolein determining the proper localization and morphology of the Golgicomplex. SOK1 is also important for the invasion of cells intotype I collagen and for cell migration in wound-healing assays,whereas overexpression of Mst4 seems to oppose these actions(Preisinger et al., 2004).

CCM3 has been shown to bind to GCKIII kinases, both in yeasttwo-hybrid assays (Ma et al., 2007; Rual et al., 2005) and inmammalian cells upon overexpression (Goudreault et al., 2008; Maet al., 2007; Voss et al., 2007). The reported functional andbiochemical consequences of this interaction are the modulation ofthe ERK pathway (Ma et al., 2007) and the phosphorylation ofCCM3 by SOK1 (Voss et al., 2007).

In this work, we show for the first time that CCM3 is locatedon the Golgi apparatus, forming a complex with a GCKIII proteinand GM130. CCM3 is important to maintain normal GCKIII proteinlevels and phosphorylation of the SOK1 substrate, 14-3-3. As aconsequence, cells with silenced CCM3 are impaired in theirorientation and migration capabilities. These results link Golgi

CCM3/PDCD10 stabilizes GCKIII proteins to promoteGolgi assembly and cell orientationMiguel Fidalgo1, María Fraile1, Ana Pires1, Thomas Force2, Celia Pombo1 and Juan Zalvide1,*1Laboratory of Cell Signalling and Cancer Research, Department of Physiology, University of Santiago de Compostela, San Francisco s/n 15705,Santiago de Compostela, A Coruña, Spain2Center for Translational Medicine, Thomas Jefferson University, Philadelphia, PA 19107, USA*Author for correspondence (juan.zalvide@usc.es)

Accepted 14 January 2010Journal of Cell Science 123, 1274-1284 © 2010. Published by The Company of Biologists Ltddoi:10.1242/jcs.061341

SummaryMutations in CCM3/PDCD10 result in cerebral cavernous malformations (CCMs), a major cause of cerebral hemorrhage. Despiteintense interest in CCMs, very little is known about the function of CCM3. Here, we report that CCM3 is located on the Golgi apparatus,forming a complex with proteins of the germinal center kinase III (GCKIII) family and GM130, a Golgi-resident protein. Cells depletedof CCM3 show a disassembled Golgi apparatus. Furthermore, in wound-healing assays, CCM3-depleted cells cannot reorient the Golgiand centrosome properly, and demonstrate impaired migration. Golgi disassembly after either depletion of CCM3 or dissociation ofCCM3 from the GM130-GCKIII complex is the result of destabilization of GCKIII proteins and dephosphorylation of their substrate,14-3-3. Significantly, the phenotype induced by CCM3 depletion can be reverted by expression of wild-type CCM3, but not by disease-associated mutants. Our findings suggest that Golgi dysfunction and the ensuing abnormalities of cell orientation and migration resultingfrom CCM3 mutations contribute to CCM pathogenesis.

Key words: SOK1/YSK1, CCM, Mst

Jour

nal o

f Cel

l Sci

ence

1275CCM3, Golgi and cell orientation

dysfunction and abnormalities in cell polarity to the developmentof cerebral cavernomas.

ResultsCCM3 is located on the Golgi apparatusBecause of its relevance to cavernoma pathogenesis, we wanted tostudy the cell biological role of CCM3 and the consequences of itsmutation. When we performed an immunofluorescence analysis ofCCM3 in cultured cells, we saw that the immunoreactivity of CCM3was concentrated in a spot near the nucleus, where it overlappedwith the cis-Golgi protein Golgi 58K (Fig. 1A) and also with anothercis-Golgi protein, GM130 (supplementary material Fig. S1). As acontrol for the specificity of the immunofluorescence, we transfectedSaOS2 cells with a small interfering (si)RNA against CCM3 (Fig.2A). Under these circumstances, the immunoreactivity of theCCM3 antibody diminished significantly, suggesting that thedetected signal is indeed CCM3. This localization of CCM3 wasunexpected, as it is at odds with its reported interaction with otherCCM gene products, such as CCM1 and CCM2, which have beenshown to form a complex at the cell periphery (Hilder et al., 2007).To confirm the location of CCM3 with a different technique, wepurified Golgi membranes from rat liver. CCM3 was readilydetectable in the purified Golgi membranes (Fig. 1B). The amountof CCM3 in purified Golgi membranes compared with the amountin whole tissue extracts was similar to the amount of the knownGolgi protein Golgi 58K, but was clearly lower than that of GM130.This could be due to either incomplete localization of CCM3 onthe Golgi apparatus or its dissociation from the Golgi membranesduring purification.

CCM3 is important for cell migration, and for Golgiassembly and repositioningWe next asked what is the biological function of CCM3. Therefore,we downregulated CCM3 using RNAi. Transient transfection ofCCM3 siRNA in SaOS2 cells consistently reduced CCM3 levelsby more than 90% (Fig. 2A, left panel). To rule out non-specificeffects of the siRNA, we also downregulated CCM3 by RNAi usinga lentivirally encoded small hairpin (sh)RNA directed against the3� untranslated region (UTR) of CCM3 mRNA in SaOS2 cells (Fig.2A, middle panel) and HeLa cells (Fig. 2A, right panel). In bothcell types, the shRNA was even more effective than the syntheticsiRNA in downregulating CCM3 protein levels.

We then asked what effect knockdown of CCM3 had on cellpolarity and migration, because its SOK1 binding partner has beenshown to regulate these events (Preisinger et al., 2004). To that end,we studied the effect of CCM3 knockdown on cell migration usinga wound-healing assay. We found that cell migration wassignificantly impaired in cells depleted of CCM3 (Fig. 2B,C). Thiswas true in both SaOS2 and HeLa cells (Fig. 2C), confirming thatthis is not a cell-specific effect.

The impaired migration correlated with the inability of cellsdepleted of CCM3 to reposition both the Golgi apparatus and thecentrosome towards the leading edge of the wound when treatedwith serum. This was in distinct contrast to control siRNA-treatedcells, which repositioned their Golgi apparatus and centrosome quiteefficiently (Fig. 2D-G). Significantly, this was the case for SaOS2cells transfected with siRNA or transduced with shRNA, and forHeLa cells, for which the effect was evident even 6 hours afterinitiation of the wound-healing assay (Fig. 2E for SaOS2 andsupplementary material Fig. S2 for HeLa).

Golgi disassembly and reassembly are important for therepositioning of the organelle in wound-healing assays; the ERKpathway is believed to be important for the disassembly step (Biselet al., 2008). As CCM3 has been reported to activate ERK in othercellular systems (Ma et al., 2007), we assessed ERK activity inCCM3-depleted cells. Unexpectedly, CCM3 depletion did notaffect ERK activation by serum (Fig. 3A), suggesting that CCM3regulates Golgi repositioning by means of an ERK-independentmechanism. Furthermore, in both serum-deprived and serum-stimulated cells, CCM3 knockdown increased the number of cellswith a dispersed Golgi. This was irrespective of cell type or methodof CCM3 knockdown employed (Fig. 3B,C), and was even apparentin normal proliferating cells (Fig. 4A,B). Furthermore, the cells withdispersed Golgi were the same as those that had undetectable CCM3by immunofluorescence (supplementary material Fig. S3). The sameGolgi phenotype was seen in HeLa cells (supplementary materialFig. S2), confirming that this is not a cell-specific phenomenon.

The dispersion seen after CCM3 knockdown could be due to theinability of the cell to reassemble the organelle after its disassembly.To test this hypothesis, we induced Golgi dispersion with the fungalmetabolite Brefeldin A (BFA). BFA induced comparable dispersalof the Golgi apparatus in both control cells and cells depleted ofCCM3. However, whereas a compact Golgi was discernible in mostcontrol cells within 90 minutes of washout of BFA, a high

Fig. 1. CCM3 localizes to the Golgi apparatus. (A)SaOS2 cells were transfected with control or CCM3 siRNA. CCM3 (red) and Golgi 58K (a Golgi marker,green) were stained 96 hours later by immunofluorescence and visualized by confocal microscopic analysis. Nuclei were identified by Hoechst 33342 staining(blue). Scale bar: 10m. (B)Rat liver homogenate and rat liver Golgi membranes were analyzed by immunoblotting using anti-CCM3, anti-GM130 and anti-Golgi-58K (as Golgi markers), or anti-tubulin (as a cytosolic marker).

Jour

nal o

f Cel

l Sci

ence

1276

percentage of cells with reduced CCM3 showed a dispersedtubuloreticular pattern even 180 minutes after washout of BFA (Fig.4A, quantified in Fig. 4B and in supplementary material Fig. S2D

Journal of Cell Science 123 (8)

for HeLa cells). Again, impaired Golgi reassembly was observedin all cell types tested and with both strategies of CCM3 knockdown.Furthermore, the effects of RNAi on Golgi reassembly were

Fig. 2. CCM3 is important for Golgi and centrosome repositioning, and cell migration. (A)SaOS2 cells were transiently transfected with control or CCM3siRNA (left panel). Stable cell populations with control or CCM3 shRNA in SaOS2 cells (middle panel) and HeLa cells (right panel) were obtained by lentiviraltransduction followed by puromycin selection. Cell extracts were analysed by immunoblotting using anti-CCM3 antibody. Anti-tubulin was used as a loadingcontrol. (B)SaOS2 and HeLa cells stably transduced with control or CCM3 shRNA were subject to wound-healing assays. Representative images at 4×magnification for HeLa cells are shown 0 and 24 hours after wound healing. (C)Migration in wound-healing assays was quantified by measuring the number ofpixels in the denuded area, and expressed as the denuded area at time 0 hours minus the denuded area at time 24 hours (for HeLa cells) or 48 hours (for SaOS2cells). Three wells were counted per condition in each experiment. The average and standard deviation of three independent experiments are shown. *P<0.01.(D)SaOS2 confluent monolayers were serum-deprived overnight, wounded and treated with 10% serum for 90 minutes (serum) or left untreated (SD). The Golgiapparatus was labeled with anti-GM130 antibody and DNA with Hoechst 33342. Golgi was scored as oriented (+) if more than half of the GM130 staining was laidwithin a 90° angle facing the wound edge (at the upper part of the micrograph). Scale bar: 20m. (E)Quantification of Golgi orientation. An average of 100 cellswas counted per condition in each experiment. Discontinuous lines mark basal levels expected for a random orientation of 25%. The average and standarddeviation of three independent experiments are shown. *P<0.01. (F,G)Centrosome orientation was assessed under the same conditions and with the same criteriaas for Golgi (see D,E), using an anti--tubulin antibody for staining. *P<0.01.

Jour

nal o

f Cel

l Sci

ence

1277CCM3, Golgi and cell orientation

dependent on CCM3 knockdown, because re-expressing CCM3 bytransfection (Fig. 4C) rescued the effects of CCM3 shRNA on Golgireassembly in both control cells and after BFA treatment (Fig. 4D).Likewise, expression of CCM3 also rescued both Golgi andcentrosome orientation in the wound-healing assay (Fig. 4E forGolgi and data not shown for centrosome), and cell migration inthe same assay (Fig. 4F).

CCM3 forms a complex with the cis-Golgi GCKIII proteinsand GM130CCM3 has been found bound to the other CCM proteins, CCM2and CCM1, at least upon overexpression (Hilder et al., 2007; Stahlet al., 2008; Voss et al., 2007), and depletion of CCM2 has beenshown to affect cytoskeletal organization, probably throughmodulation of the small G proteins cdc42 and RhoA (Crose et al.,

2009; Whitehead et al., 2009). Because small G proteins have beenimplicated in the regulation of centrosome relocation and directedcell migration (Nobes and Hall, 1999), the CCM3 phenotype couldbe mediated by their modulation. However, our results show thatstress fiber formation was not affected in CCM3-depleted cells,either when serum deprived or when stimulated with serum.Moreover, when we measured the activity of cdc42 and RhoA inserum-deprived and serum-stimulated cells, we found no differencebetween control and CCM3-depleted cells (supplementary materialFig. S4). We concluded that the effect of CCM3 depletion wasprobably mediated by a mechanism other than modulation of smallG proteins.

The localization of CCM3 on the Golgi is probably caused byits binding to a protein or proteins present there. CCM3 has beenreported to interact with several different proteins and protein

Fig. 3. Lack of CCM3 results in Golgi disassembly. In all cases,SaOS2 cells were serum starved overnight, and then treated with10% serum for 90 minutes (serum) or left untreated (SD).(A)Extracts from cells stably transduced with control or CCM3shRNA were analyzed by immunoblotting using anti-p-ERK. TotalERK1,2 (ERK) was used as a loading control. (B)SaOS2 cellswere analyzed by immunofluorescence confocal microscopy. TheGolgi apparatus was visualized with antibodies against GM130 andGolgi 58K, and DNA was stained with Hoechst 33342. The arrowspoint to cells with dispersed Golgi. Scale bar: 10m. The lowerpanels show the dispersed Golgi area of a single cell (boxedregion) at a higher magnification. (C)Quantification of Golgidispersion of SaOS2 cells after transient transfection of control orCCM3 siRNA, and in stable cell populations with control orCCM3 shRNA. An average of 300 cells was counted per conditionin each experiment. The average and standard deviation of threeindependent experiments are shown. *P<0.01.

Jour

nal o

f Cel

l Sci

ence

1278

complexes, among them the GCKIII family of kinases (Mst4, Mst3and SOK1) (Goudreault et al., 2008; Ma et al., 2007; Rual et al.,2005; Voss et al., 2007). Two GCKIII family members (Mst4 andSOK1, but not Mst3) have been found in the cis-Golgi network,and SOK1 has been shown to be involved in Golgi repositioningand cell migration (Preisinger et al., 2004). We found that all threeGCKIII proteins are present in purified Golgi membranes.Surprisingly, Mst4 was less enriched than the other two proteins;this correlated with reduced binding of Mst4 to the Golgi proteinGM130 (supplementary material Fig. S5). We confirmed thatCCM3 and the GCKIII kinase SOK1 interact in mammalian cellswhen overexpressed (Fig. 5A). Moreover, Mst3 and Mst4 also bindto CCM3 (Fig. 5B). The interaction of CCM3 with SOK1, Mst3or Mst4 is not an artifact of overexpression, because we detectedendogenous CCM3 in immunoprecipitates of endogenous SOK1and Mst4 (Fig. 5C). Furthermore, although endogenous CCM3could not be detected in immunoprecipitates of Mst3, Mst3 was

Journal of Cell Science 123 (8)

present in CCM3 immunoprecipitates (Fig. 5D). To identify thedomain of SOK1 important for interaction with CCM3, we co-transfected CCM3 with full-length SOK1, the N-terminal kinasedomain (SOK1C) or the C-terminal regulatory domain (SOK1RD).CCM3 could be immunoprecipitated with full-length SOK1 andSOK1RD, but not with SOK1C (Fig. 5E). Thus, the interactionbetween CCM3 and SOK1 is mediated by the C-terminal domainof the latter.

The domain of SOK1 that binds to CCM3 (amino acids 333 to426) is C terminal with respect to its binding site for GM130 (aminoacids 270 to 302) (Nogueira et al., 2008; Preisinger et al., 2004)and does not overlap with it. This is compatible with SOK1 bindingto GM130 and CCM3 simultaneously. To determine if such atrimeric complex can exist in cells, we transfected epitope-taggedversions of SOK1 and CCM3 in HEK293 cells (Fig. 5F). SOK1was recovered in immunoprecipitates of endogenous GM130 withthe same efficiency as when transfected alone or with CCM3. By

Fig. 4. CCM3 is important for Golgi reassembly afterdispersion. (A)SaOS2 cells stably transduced with control orCCM3 shRNA were treated with BFA (10g/l) for 90 minutesat 37°C. BFA was then washed out and cells were incubated at37°C for 45 minutes, 90 minutes or 180 minutes to allow Golgireassembly. Cells were fixed and stained for GM130 byimmunofluorescence and analyzed by confocal microscopy. Scalebar: 10m. (B)Quantification of Golgi dispersion of SaOS2 cellswith transiently transfected control or CCM3 siRNA, and instable cell populations with control or CCM3 shRNA, when leftuntreated (control), and when treated with BFA and recovered forthe specified periods of time. An average of 300 cells wascounted per condition in each experiment. The average andstandard deviation of three independent experiments are shown.P<0.01 for CCM3 versus control siRNA, and CCM3 versuscontrol shRNA, in the control and at times 45, 90 and 180minutes. (C)SaOS2 cells transduced with control shRNA (firstlane) or CCM3 shRNA (second and third lanes) were transfectedwith empty vector or HA-CCM3 expression vector. Extracts wereprepared and a western blot analysis was performed using anti-HA, anti-CCM3 or anti-tubulin (as a loading control) antibodies.(D)Stable SaOS2 cell populations with CCM3 shRNA weretransfected with HA-CCM3 or empty vector (HA). Cells werethen subjected to the same treatment and analysis as in B.*P<0.01. (E)Quantification of Golgi orientation in GFP-positiveSaOS2 cells stably expressing control or CCM3 shRNA, andtransfected with GFP and either HA empty vector or HA-CCM3.Confluent monolayers of cells were serum deprived overnight,wounded, and treated with 10% serum for 90 minutes (serum) orleft untreated (SD). After staining with anti-GM130, cells wereanalyzed by immunofluorescence and scored for Golgiorientation. An average of 80 GFP-positive cells was counted percondition in each experiment. Discontinuous lines mark basallevels expected for a random orientation of 25%. The average andstandard deviation of three independent experiments are shown.N.S., P>0.05. (F)Quantification of cell migration in GFP-positiveSaOS2 cells stably expressing control or CCM3 shRNA, andtransfected with GFP and either HA empty vector or HA-CCM3.Confluent monolayers of cells were subject to wound-healingassays. Migration in wound-healing assays was quantified bymeasuring the number of pixels in the denuded area, andexpressed as the denuded area at time 0 hours minus the denudedarea at 48 hours. Three wells were counted per condition in eachexperiment. The average and standard deviation of threeindependent experiments are shown. N.S., P>0.05.

Jour

nal o

f Cel

l Sci

ence

1279CCM3, Golgi and cell orientation

contrast, whereas CCM3 could be detected in the GM130immunoprecipitate, its recovery was greatly increased when SOK1was co-transfected. Furthermore, both SOK1 and GM130 werereadily detected in immunoprecipitates of endogenous CCM3 inSaOS2 cells (Fig. 5G). These results demonstrate that CCM3 formsa complex with GM130 and strongly suggest that SOK1 (andprobably other GCKIII proteins) bridge this interaction.

Effects of CCM3 depletion on GCKIII proteinsThe CCM3-interacting proteins SOK1 and Mst4 are important forGolgi assembly and cell orientation. Thus, we asked whether CCM3could modulate their activities. Depletion of CCM3 inducedinhibition of SOK1 kinase activity, both in HeLa and in SaOS2cells (Fig. 6A). This inhibition correlated with downregulation ofthe levels of SOK1 and the other two GCKIII proteins, Mst3 andMst4. This was not a non-specific effect of RNAi, because it wasreversed by the expression of CCM3 (Fig. 6B). We studied furtherthe mechanism of this downregulation of GCKIII proteins and found

that depletion of CCM3 did not induce changes in SOK1 mRNA(Fig. 6C). We then asked whether downregulation could be due todestabilization of the proteins. Thus, we assessed the decay of Mst3after protein synthesis inhibition with cycloheximide (CHX) (asMst3 is the only GCKIII protein we could consistently detect bywestern blot in shCCM3 cells). Depletion of CCM3 induced amarked shortening of the half-life of the Mst3 protein (13.9 hoursin control cells versus 5.5 hours in CCM3-depleted cells; Fig. 6D).To gain further insight into the mechanism of destabilization ofGCKIII proteins in the absence of CCM3, we treated cells with theproteasome inhibitor MG132. Mst3 accumulated significantly incells depleted of CCM3 but not in control cells, confirming thatMst3 degradation in the absence of CCM3 is mediated through theproteasome (Fig. 6E). Consistently, when we assessed the amountof ubiquitylated SOK1 in CCM3-depleted cells, we found that itwas at least as much as in cells with normal amounts of CCM3,despite the reduced amount of SOK1. This was the case both inuntreated cells and in cells treated with the proteasome inhibitor

Fig. 5. CCM3 forms a complex with the GCKIIIproteins and cis-Golgi GM130 protein.(A)HEK293 cells were transfected with empty orHA-CCM3 expression vectors with or withoutFLAG-SOK1. Western blots of FLAG-SOK1, HA-CCM3 and tubulin were performed with total celllysates or after anti-FLAG or anti-HAimmunoprecipitations (IP). (B)HEK293 cells weretransfected with GFP-CCM3 together with emptyvector, HA-SOK1, HA-Mst3 or HA-Mst4. Westernblots for GFP, HA-GCKIII and tubulin wereperformed with total cell lysates or after anti-HAimmunoprecipitations. (C)Extracts were preparedfrom SaOS2 cells, and western blots for CCM3,SOK1, Mst3, Mst4 or tubulin were performed withtotal cell lysates or after the specifiedimmunoprecipitations. (D)Extracts were preparedfrom SaOS2 cells, and western blots for Mst3 andCCM3 were performed after immunoprecipitationwith non-specific or anti-CCM3 antibody.(E)Western blots for FLAG, HA and tubulin wereperformed of total cell lysates orimmunoprecipitations of HEK293 cells transfectedwith empty or HA-CCM3 expression vectors with orwithout full-length FLAG-SOK1 (FL), the Cterminus of FLAG-SOK1 (RD) or the catalyticdomain of FLAG-SOK1 (C). A schematicrepresentation of the SOK1 fragments used is shown.(F)HEK293 cells were transfected with emptyvectors, HA-CCM3, FLAG-SOK1 or both. Westernblots were performed for GM130, FLAG-SOK1 andHA-CCM3 in total cell lysates or after anti-GM130immunoprecipitations. (G)Extracts were preparedfrom SaOS2 cells, and western blots for CCM3,SOK1, GM130 and tubulin were performed with totalcell lysates or after nonspecific, anti-CCM3 or anti-SOK1 immunoprecipitations.

Jour

nal o

f Cel

l Sci

ence

1280

(Fig. 6F). We concluded that degradation of GCKIII proteins ismediated at least in part by ubiquitylation and degradation throughthe proteasome, and that this is enhanced in the absence of CCM3.

To assess whether the stabilization of the kinase component ofthe GM130-GCKIII complex by CCM3 plays a crucial role in Golgiassembly and cell migration, we expressed all possible combinationsof GCKIII proteins in CCM3-depleted cells. All combinations testedresulted in an increase in Golgi assembly, with the expression ofthe three proteins inducing levels of Golgi assemblyindistinguishable from that of control cells (Fig. 7A).

Journal of Cell Science 123 (8)

Fig. 6. Depletion of CCM3 induces enhanced ubiquitylation andproteasome-dependent degradation of GCKIII proteins. (A)The kinaseactivity of SOK1 was determined after immunoprecipitation with the SOK1antibody. Quantification of a representative experiment is shown. (B)SaOS2cells stably transduced with control or CCM3 shRNA constructs weretransfected with empty plasmid or HA-CCM3 expression vector. Western blotsfor GCKIII family kinases, GM130, HA-CCM3, total CCM3 and tubulin wereperformed. (C)Total RNA was prepared from SaOS2 and HeLa stable cellpopulations transduced with control or CCM3 shRNA, and the levels ofSOK1, CCM3 and GAPDH mRNAs were determined by RT-PCR. (D)SaOS2cells stably transduced with control or CCM3 shRNA were treated with20g/ml CHX. Extracts were prepared at the indicated times of treatment, andwestern blots for total Mst3 and tubulin were performed. (E)SaOS2 cellsstably transduced with control or CCM3 shRNA were treated with 10MMG132 for the indicated times. Mst3 and tubulin were analyzed with westernblotting. (F)SaOS2 cells stably transduced with control or CCM3 shRNAwere transiently transfected with HA-tagged ubiquitin (HA-Ubi) and incubatedin the presence or absence of 10M MG132 for 9 hours. Extracts wereimmunoprecipitated with non-specific or anti-SOK1 antibody, and ubiquitinconjugates were detected with anti-HA antibody. Control western blots forSOK1 and tubulin were performed with total cell lysates.

Fig. 7. Golgi disassembly in the absence of CCM3 is the result of GCKIIIdownregulation and disruption of the CCM3-GCKIII complex. (A)SaOS2cells stably transduced with CCM3 or control shRNA were transfected withempty vector and, for CCM3 shRNA, with HA-SOK1, HA-Mst3 and HA-Mst4 in the indicated combinations. After staining with anti-GM130, cellswere analyzed by immunofluorescence and scored for Golgi dispersion. Anaverage of 400 cells was counted per condition in each experiment. Theaverage and standard deviation of three independent experiments are shown.N.S., P>0.05. (B)SaOS2 cells were transfected with empty vector or GFP-SOK1-RD plasmid. Western blots for GM130, SOK1, CCM3 and GFP wereperformed for whole cell lysates (upper panels); for SOK1 and CCM3 afterimmunoprecipitation with an anti-SOK1 antibody or an unrelated antibody(middle panels); and for SOK1, CCM3 and GFP after immunoprecipitationwith an anti-CCM3 antibody or an unrelated antibody (lower panels).(C)SaOS2 cells stably transduced with CCM3 shRNA were transfected withGFP or GFP-SOK1-RD. Cells were either untreated (control) or incubatedwith BFA for 90 minutes, then washed and recovered for 180 minutes. Afterstaining with anti-GM130, cells were analyzed by immunofluorescence andGFP-positive cells scored for Golgi dispersion. An average of 80 GFP-positivecells was counted per condition in each experiment. The average and standarddeviation of three independent experiments are shown. *P<0.01.

Jour

nal o

f Cel

l Sci

ence

1281CCM3, Golgi and cell orientation

The above results show that GCKIII proteins can rescue the Golgiphenotype induced by CCM3 depletion and suggest that CCM3induces its effects through binding to this family of proteins on thecis face of the Golgi. We reasoned that, if this were the case,dissociation of CCM3 from GCKIII proteins should have the sameeffect as CCM3 depletion. Thus, we disrupted the GCKIII-CCM3interaction by overexpressing SOK1RD, the C-terminal domain ofSOK1 that binds CCM3. As expected, SOK1RD could bind toendogenous CCM3 (Fig. 7B, lower panels). Importantly, expressionof SOK1RD clearly reduced the amount of CCM3 bound toendogenous SOK1 (Fig. 7B, middle and lower panels). Consistentwith the binding of CCM3 to GCKIII proteins being important forits effects, levels of SOK1 were also greatly reduced (Fig. 7B, upperpanels). Furthermore, the percentage of cells with disassembledGolgi was significantly higher in cells expressing SOK1RD than incontrol cells (Fig. 7C).

Having established that the effects of CCM3 on Golgi assemblyare mediated through its binding and modulation of GCKIIIproteins, we asked what event downstream of these kinases couldbe important for this effect. 14-3-3 has been shown to bephosphorylated at serine 58 by SOK1, both in untreated cells andafter oxidative stress (Preisinger et al., 2004; Zhou et al., 2009).Consistent with 14-3-3 being a target of GCKIII proteins, itsphosphorylation at serine 58 was diminished in CCM3-depletedcells (Fig. 8A). More significantly, transfection of the 14-3-3phosphorylation-mimicking mutant S58D (but not wild-type14-3-3 or the non-phosphorylatable S58A mutant) rescued the

phenotype of CCM3 knockdown (Fig. 8B). Thus, the effects ofCCM3 on Golgi assembly are mediated at least in part through14-3-3 phosphorylation at serine 58.

Fig. 8. Depletion of CCM3 induces Golgi dispersion through inhibition of14-3-3 phosphorylation. (A)Extracts from SaOS2 and HeLa cells stablytransduced with control or CCM3 shRNA were analysed by immunoblottingusing anti-phosphoSer58-14-3-3 and anti-tubulin antibodies. As a positivecontrol for serine 58 phosphorylation, SaOS2 control shRNA was treated with500M H2O2 for 40 minutes. (B)SaOS2 stable cell populations with CCM3shRNA were transiently transfected with 14-3-3-WT (wild type), 14-3-3-S58D, 14-3-3-S58A or empty vector. Cells were fixed and stained for GM130by immunofluorescence and analyzed by confocal microscopy. An average of300 cells was counted per condition in each experiment. The average andstandard deviation of three independent experiments are shown. *P<0.01.

Fig. 9. Clinically relevant mutants of CCM3 are unable to bind to GCKIIIproteins, upregulate GCKIII levels or rescue the Golgi phenotype ofCCM3-depleted cells. (A)HA empty vector, HA-CCM3-WT, HA-CCM3-196or HA-CCM3-203 vectors were transfected in SaOS2 cells stably transducedwith CCM3 shRNA. Cell extracts were analyzed by immunoblotting usinganti-HA, anti-CCM3 and anti-tubulin antibodies. A schematic representationof wild-type CCM3 (WT) and the mutants (196 and 303) used in this study isshown. (B)Cells from A were treated with BFA for 90 minutes and thenrecovered after 45 minutes, 90 minutes or 180 minutes. After staining forGM130, cells were scored for Golgi dispersion. An average of 300 cells wascounted per condition in each experiment. The average and standard deviationof three independent experiments are shown. P<0.01 for HA-empty, HA-CCM3-196 and HA-CCM3-203 versus HA-CCM3-WT for the control and attimes 45 minutes, 90 minutes and 180 minutes. (C)HEK293 cells weretransfected with HA empty vector, HA-CCM3-WT, HA-CCM3-196 or HA-CCM3-203. Western blots were performed for HA, SOK1 and tubulin in totalcell lysates or after anti-SOK1 immunoprecipitations.

Jour

nal o

f Cel

l Sci

ence

1282

Clinically relevant CCM3 mutants cannot rescue the CCM3depletion phenotypeIf the cellular phenotype of CCM3 depletion we have describedis relevant to CCM pathogenesis, then clinically relevant CCM3mutants should not be able to rescue it. Most CCM3 mutationsare deletions of the whole gene or premature stop codons thatresult in highly defective truncated proteins. However, twodifferent mutations in codons 196 and 203 have been described.Although both result in premature stop codons, the mutated genesencode proteins that are close in length to the wild-type protein(212 residues) (Fig. 9A) (Bergametti et al., 2005; Liquori et al.,2006). When we expressed these mutants, we found that theirlevels of expression were similar to that of wild-type CCM3 (Fig.9A). Strikingly, the mutants were unable to rescue the defect inGolgi assembly induced by CCM3 depletion (Fig. 9B); thiscorrelated with the inability of both mutants to bind to SOK1efficiently (Fig. 9C) or to restore SOK1 expression to normal levels(Fig. 9A). The fact that mutants of CCM3 that lead to CCMformation were unable to rescue the Golgi phenotype induced bythe depletion of wild-type CCM3 suggests that this phenotype iscentral to the pathophysiology of cavernous malformations.Finally, the inability of the CCM3 mutants to bind SOK1, to restoreSOK1 levels to normal or to rescue the Golgi phenotype isconsistent with the function of CCM3 in Golgi assembly beingmediated by its binding to, and stabilization of, SOK1 and otherGCKIII proteins.

DiscussionCCMs are vascular lesions in which endothelial cells form largecaverns instead of normal capillaries. Because of the nature andlocation of the lesion, direct biochemical and cellular study of theseabnormalities has been difficult. However, the realization that someCCMs have a genetic basis, and the identification of three CCMloci, has allowed the modeling of the disease at both the whole-animal and cellular level by inactivation of the relevant genes.Studies inactivating CCM1 and CCM2 have shown that these genesare important for cytoskeletal regulation. Accompanying evidenceindicates that their products interact with endothelial transmembranereceptors, which suggests that endothelial cell shape and signalingmight underlie at least some forms of the disease. However, nostudy showing a similar function for CCM3 has been reported sofar.

In this paper, we show that CCM3 is important for Golgi assemblyand for centrosome and Golgi orientation in wound-healing assays.Independently of its role in directed cell migration, the adequatepositioning of centrosome and Golgi has been proposed to reflectthe capability of the cell to polarize (Yadav et al., 2009); therefore,lack of polarizing ability is likely to be an important consequenceof CCM3 inactivation. The two functions, cell orientation and Golgiassembly, are likely to be related, as the Golgi apparatus needs todisassemble and reassemble to change its position within the cellwhen the cell undergoes reorientation, and the integrity and positionof the Golgi complex are known to be important for this orientationand directed migration (Vinogradova et al., 2009; Yadav et al.,2009). The effects of CCM3 on the Golgi apparatus are also likelyto be linked to CCM pathogenesis, as two different clinicallyimportant mutants cannot rescue the effects of CCM3 knockdownon the Golgi.

Strong circumstantial evidence points to the Golgi apparatus asthe crucial site of action of the GCKIII-CCM3 complex. The effectscaused by CCM3 depletion are compatible with Golgi disassembly

Journal of Cell Science 123 (8)

being the primary effect. In addition, GCKIII proteins and 14-3-3phosphorylation are important for CCM3 functions, and theseproteins have been localized to the Golgi (Preisinger et al., 2004;this article). Although there might be a pool of CCM3 that is notassociated with the Golgi apparatus, it is clear that, independentlyof its location, CCM3 exerts the effects we report here through thebinding and modulation of GCKIII proteins and the phosphorylationof 14-3-3. Importantly, CCM3 does not overly affect the activityof the small G proteins Rho and cdc42, or the actin cytoskeleton,which suggests that it is not acting in concert with CCM1 andCCM2.

The relevant biochemical effect of CCM3 with respect to GCKIIIproteins seems to be to stabilize them, because, in the absence ofCCM3, both the levels and kinase activity of all three GCKIIIproteins are downregulated due to destabilization of the proteins.Given that this facilitated degradation occurs through a proteasome-and ubiquitin-dependent pathway, CCM3 is likely to exert itsfunction by protecting GCKIII proteins from ubiquitin ligation.Whether it directly impairs the action of an E3 ligase or inhibits amodification that marks GCKIII proteins for ubiquitylation will bethe subject of future studies that will require the identification ofthe relevant E3 ligase.

The identification of the specific GCKIII protein or proteins thatare important for CCM3 action will also require future research.We show here that only the expression of all three family members(SOK1, Mst3 and Mst4) fully restores normal Golgi assembly, butdissection of the contribution of each GCKIII protein to thephenotype induced by CCM3 depletion will require inhibition, notoverexpression, of each individual kinase. However, SOK1 is themost likely GCKIII protein to be important in this respect. First,SOK1, but not the other two family members (Mst3 and Mst4), hasbeen shown to be involved in Golgi assembly and cell orientation(Preisinger et al., 2004). Second, our results show thatphosphorylation of 14-3-3 at serine 58 is an important downstreamevent for the CCM3 phenotype and SOK1 is the only GCKIIIprotein that has been shown to phosphorylate 14-3-3.

Phosphorylation of 14-3-3 at serine 58 is an importantconsequence of GCKIII activity. Because of the pleiotropic natureof 14-3-3 proteins, it is difficult to pinpoint the mechanism throughwhich this modification can be linked to Golgi assembly and cellorientation. However, it should be noted that 14-3-3 has beenshown to bind to proteins important for cell polarity, such as par3(Hurd et al., 2003). Intriguingly, besides its role in cell polarity,par3 is associated with tight junctions and is important for theirassembly (Anderson and Van Itallie, 2008). Importantly, one of thecharacteristic features of the cells that form cerebral cavernomas istheir aberrant tight junctions, which result in a leaky blood-brainbarrier (Clatterbuck et al., 2001).

It is possible that a defect in cell polarity might also underlie thepathogenesis of other CCM syndromes. For instance, cells deficientin CCM2 have low activity of cdc42, a master regulator of cellpolarity (Whitehead et al., 2009). Because depletion of CCM3affects cell orientation without having a detectable effect on theactivity of cdc42, different CCM proteins might be inducing thesame or a very similar cellular effect through different mechanisms.

In summary, our findings demonstrate that clinically relevantmutations of CCM3 are unable to rescue the phenotypes that resultfrom knockdown of CCM3. These abnormalities are likely to playa role in the pathogenesis of cerebral cavernomas and, if confirmed,this could set the stage for corrective therapeutic strategies aimedat CCM3.

Jour

nal o

f Cel

l Sci

ence

1283CCM3, Golgi and cell orientation

Materials and MethodsAntibodies, plasmids and drugsThe antibodies used in this study are: rabbit polyclonal CCM3 (Proteintech Group);rabbit polyclonal GM130 CB-1008 (Calbiochem); mouse monoclonal FLAG M2,mouse monoclonal Golgi 58K clone 58K-9, mouse monoclonal -tubulin clone B-5-1-2 and mouse monoclonal -tubulin clone GTU-88 (Sigma-Aldrich); goatpolyclonal CCM3 (sc-67908), goat polyclonal SOK1 (sc-6865), goat polyclonal MST3(sc-21400), goat polyclonal MST4 (sc-7150), goat polyclonal GM130 (sc-16268),mouse monoclonal anti-phospho-ERK (sc-7383) and mouse monoclonal HA probe(sc-7392) (SantaCruz Biotechnology); rabbit monoclonal ERK1-2 4695, rabbitpolyclonal MST3 3723 and rabbit polyclonal MST4 3822 (Cell Signaling Technology);mouse monoclonal SOK1 clone 1G6 (Abnova); mouse monoclonal cytochrome cclone 7H8.2C12 (Becton Dickinson Biosciences); rabbit polyclonal GFP (ab290)(Abcam); and rabbit polyclonal 14-3-3-phospho-Ser58 (PA1-4612) (AffinityBioReagents). The secondary antibodies used are: donkey anti-goat AlexaFluor488,goat anti-mouse AlexaFluor488 (Molecular Probes); goat anti-rabbit Cy3 (JacksonImmunoResearch Laboratories); goat anti-rabbit horseradish peroxidase (HRP), goatanti-mouse HRP (Pierce).

All plasmids were constructed using standard molecular biology techniques. Theoligonucleotides designed for the different mutants were purchased from ThermoScientific (sequence available upon request).

BFA was from Sigma-Aldrich. Cells were treated with BFA (10 g/l for SaOS2cells or 5 g/l for HeLa cells) for 90 minutes (SaOS2 cells) or 30 minutes (HeLacells) at 37°C to disperse Golgi protein. BFA was washed out and cells were incubatedat 37°C for 45 minutes, 90 minutes or 180 minutes in a complete culture medium toallow the Golgi reassembly process. Cells were treated with 20 g/ml CHX fromSigma-Aldrich. Cells were treated with MG132 (10 M) from Calbiochem.

siRNA- and shRNA-mediated knockdownAll chemically synthesized siRNA was purchased from Dharmacon (siRNA CCM3ON-TARGETplus and siRNA control ON-TARGETplus) and transfected using a cellline nucleofector kit with Nucleofector II from Amaxa Biosystems. Stable cellpopulations with silenced CCM3 or control were obtained through selection afterlentiviral transduction using MISSION lentiviral Non-Target shRNA ControlTransduction Particles or MISSION lentiviral shRNA CCM3 Transduction Particlesagainst the 3�-UTR of human CCM3 from Sigma-Aldrich (TRC libraries).

RT-PCR analysisCells were harvested and RNA extracted using TRIzol reagent (Invitrogen). Reversetranscriptase (RT)-PCR was performed by standard procedures using adequate PCRamplification primers (sequence available upon request). PCR products were separatedusing 1.0% agarose gel electrophoresis and visualized by ethidium bromide staining.

Cell culture and wound-healing assay for cell migrationHEK293 cells (European Collection of Cell Cultures), SaOS2 cells and HeLa cells(American Type Culture Collection) were grown as recommended.

For wound-healing assays, HeLa and SaOS2 cells were grown to confluence, andserum starved in 0.1% FBS overnight. Monolayer wounds were produced using a200 l pipette tip scratched through the center of the well. Photomicrographs weretaken of the initial wound for comparison. Cells were then treated with either 0.1%FBS (SD) or 10% FBS (serum), and allowed to migrate into denuded areas for 24hours in HeLa cells or 48 hours in SaOS2 cells. Cell migration was monitored usingan inverted microscope (Olympus IX70) with a 4�/0.10 objective using a Hoffmanmodulation contrast system. Photoimages were acquired using an Olympus DP digitalcamera. The distance of migration was measured, using ImageJ software, as thenumber of pixels within the denuded area at time 0 hours minus the number of pixelswithin the denuded area at time 24 hours for HeLa and 48 hours for SaOS2 cells.

Immunofluorescence and image analysisCells were cultured on coverslips and fixed for 15 minutes in paraformaldehyde (PFA)4%, permeabilized for 10 minutes in PBS with 0.25% Triton X-100, and incubatedwith the indicated antibodies followed by fluorescent secondary antibodies. DNAwas stained with Hoechst33342. The coverslips were mounted in aqueous mediumwith anti-fading agents (Gel-Mount). Confocal images were collected using a Leicaconfocal microscope equipped with an HCX PL APO CS 63�/1.32 objective. LCSsoftware was used for acquisition and analysis. Images are combinations of opticalsections taken in the z-axis at 0.5 m intervals. For all microscope photographs, AdobePhotoshop software was used to cut, resize and mount the photographs in the figures.

Fluorescent staining of actinFor actin staining, cells grown on glass slides were fixed for 15 minutes in PFA 2%,permeabilized for 10 minutes in PBS with 0.5% Triton X-100 and then incubatedfor 45 minutes with 10 g/l of phalloidin-TRITC (Sigma). The cells were thenmounted in aqueous medium with anti-fading agents (Gel-Mount) and examined byfluorescent microscopy.

Quantification of Golgi and centrosome polarization, and Golgi dispersionThe Golgi was stained with GM130 antibody and centrosomes with -tubulin antibody.Golgi and centrosome orientation was determined for the first row of cells facing the

wound, as described previously (Etienne-Manneville and Hall, 2001), and was countedas oriented if more than half was located in a 90° sector emerging from the centerof the nucleus and facing the wound edge. Basal levels of expected randomorientation of 25% are marked by discontinuous lines in the graphs. The percentageof Golgi or centrosome orientation was calculated by dividing the number of orientedcells by the number of total cells for each condition. A minimum of 100 cells wascounted for each condition and experiment, unless otherwise specified. The meanpercentage from four independent experiments was averaged, and presented as a meanand standard deviation.

Golgi was considered dispersed when the GM130 staining or Golgi 58K stainingspanned more than 5 m at its longest diameter. A minimum of 300 cells was countedfor each condition and experiment, unless otherwise specified. The mean percentagefrom four independent experiments was averaged, and presented as a mean andstandard deviation.

Transfections, immunoprecipitation and immunoblottingHEK293 cells were transfected using the calcium phosphate protocol. SaOS2 cellswere transfected using the cell line nucleofector kit with Nucleofector II from AmaxaBiosystems following manufacturer’s instructions.

Western blotting and immunoprecipitations were performed by standard procedures,after preparation of extracts in lysis buffer [20 mM HEPES pH 7.4, 2 mM EGTA,50 mM -glycerophosphate, 1 mM sodium orthovanadate, 1% Triton X-100, 10%glycerol, 1 mM dithiothreitol (DTT), 400 M PMSF, 2 M pepstatin, 2.3 g/mlaprotinin, 2 M leupeptin].

Kinase assaysExtracts were immunoprecipitated with anti-SOK1 antibody and kinase assays wereperformed as described (Pombo et al., 1996). Briefly, immune complexes werecollected with protein G-sepharose beads. Beads were washed three times in lysisbuffer, three times in LiCl buffer (500 mM LiCl, 2 mM DTT, 100 mM Tris-HCl pH7.6) and three times in assay buffer (20 mM MOPS pH 7.2, 2 mM EGTA, 10 mMMgCl2, 0.1% Triton X-100, 1 mM DTT). Kinase assays were started by the additionof myelin basic protein (MBP) and [-32P]ATP (100 M, 3000-9000 c.p.m./pmol).After 5 minutes at 30°C, the kinase reactions were stopped with Laemmli samplebuffer. Following SDS-PAGE and autoradiography, the bands corresponding to thesubstrate were cut out of the gel and radioactivity was determined by liquid scintillationcounting in a 1414 liquid scintillation counter (Wallac-Winspectral).

Isolation of Golgi membranesLiver from Sprague-Dawley rats was used to obtain enriched Golgi fractions usingdiscontinuous sucrose gradient centrifugation (Bonifacino et al., 2004). Briefly, 10g of liver was minced and suspended in 40 ml of homogenization medium (HM:0.25 mM sucrose, 10 mM Tris-Cl pH 7.4). The sample was homogenized using aPotter-Elvehjem homogenizer with five up-and-down strokes of the pestle, rotatingat 500 rpm. The homogenate was centrifuged at low speed (1000 g for 10 minutesat 4°C). The nuclear pellet was transferred in 20 ml of HM to a Dounce homogenizerand then centrifuged at low speed. Previous and last supernatants were combinedand centrifuged (3000 g for 10 minutes at 4°C). The resulting supernatant wascentrifuged in a high-speed centrifuge (17,000 g for 20 minutes at 4°C). The pelletwas resuspended in 2 ml of HM, with 8 ml sucrose (2.0 M) added, and homogenizedwith a Dounce homogenizer. 5 ml of homogenate was loaded in a discontinuoussucrose gradient (4 ml 1.33 M, 2 ml 1.2 M, 2 ml 1.1 M, 2 ml 0.77 M and 0.25 M tofill the tube) and ultracentrifuged (100,000 g for 1 hour at 4°C). The enriched Golgimembrane fractions were collected at the 0.77 M-1.1 M and 1.1 M-1.2 M interfaces.Analysis was carried out by SDS-PAGE and immunoblotting.

Determination of RhoA and Cdc42 activityActivation assays for the small GTPases were performed using the G-LISA RhoAor Cdc42 luminescence-based Biochem kits (Cytoskeleton). Briefly, SaOS2 cellsstably transduced with control or CCM3 shRNA were grown in a six-well clusterplate, serum starved for 24 hours, then treated with 10% serum for 3 minutes (serum)or left untreated, lysed in the provided lysis buffer (including protease inhibitors) andcentrifuged to obtain the soluble extract. After correction of the total proteinconcentration according to the manufacturer’s instructions, 0.25 mg/ml of each samplewas added to the ELISA plates. Wells were washed and active GTPase was detectedby specific antibodies according to the manufacturer’s instructions in a luminescentreaction and measured in a Sunrise microplate reader (Tecan).

Statistical analysisThe statistical significance of all data obtained was assessed by Mann-Whitney tests.All data were analyzed using SPSS software version 12.0.0.

We thank Anxo Vidal and Gonzalo Egea for helpful discussions, andCarme Caelles (Instituto de Recerca Biomédica, Barcelona, Spain) forproviding the pCDNA3-HA-ubiquitin construct. This work wassupported by grants from the Instituto de Salud Carlos III of Spain(PI080655, co-financed with Regional Development European Funds)and from the Xunta de Galicia (PGIDT06PXIB208061PR to J.Z. and

Jour

nal o

f Cel

l Sci

ence

1284 Journal of Cell Science 123 (8)

INCITE09 208 110 PR to C.P.). M.F. is a predoctoral fellow from theMinisterio de Educación y Ciencia of Spain.

Supplementary material available online athttp://jcs.biologists.org/cgi/content/full/123/8/1274/DC1

ReferencesAkers, A. L., Johnson, E., Steinberg, G. K., Zabramski, J. M. and Marchuk, D. A.

(2009). Biallelic somatic and germline mutations in cerebral cavernous malformations(CCMs): Evidence for a two-hit mechanism of CCM pathogenesis. Hum. Mol. Genet.18, 919-930.

Anderson, J. M. and Van Itallie, C. M. (2008). Tight junctions. Curr. Biol. 18, R941-R943.

Bergametti, F., Denier, C., Labauge, P., Arnoult, M., Boetto, S., Clanet, M., Coubes,P., Echenne, B., Ibrahim, R., Irthum, B. et al. (2005). Mutations within theprogrammed cell death 10 gene cause cerebral cavernous malformations. Am. J. Hum.Genet. 76, 42-51.

Bisel, B., Wang, Y., Wei, J. H., Xiang, Y., Tang, D., Miron-Mendoza, M., Yoshimura,S., Nakamura, N. and Seemann, J. (2008). ERK regulates Golgi and centrosomeorientation towards the leading edge through GRASP65. J. Cell Biol. 182, 837-843.

Bonifacino, J. S., Dasso, M., Harford, J. B., Lippincott-Schwartz, J. and Yamada, K.M. (2004). Isolation of Golgi membranes from a rat liver light mitochondrial fractionby flotation through a discontinuous sucrose gradient. In Short Protocols in Cell Biology:A Compendium of Methods from Current Protocols in Cell Biology (ed. J. S. Bonifacino),pp. 3. Hoboken, NJ: John Wiley and Sons, Inc.

Clatterbuck, R. E., Eberhart, C. G., Crain, B. J. and Rigamonti, D. (2001).Ultrastructural and immunocytochemical evidence that an incompetent blood-brain barrieris related to the pathophysiology of cavernous malformations. J. Neurol. Neurosurg.Psychiatry. 71, 188-192.

Crose, L. E., Hilder, T. L., Sciaky, N. and Johnson, G. L. (2009). Cerebral cavernousmalformation 2 protein promotes smad ubiquitin regulatory factor 1-mediated RhoAdegradation in endothelial cells. J. Biol. Chem. 284, 13301-13305.

Dan, I., Watanabe, N. M. and Kusumi, A. (2001). The Ste20 group kinases as regulatorsof MAP kinase cascades. Trends Cell Biol. 11, 220-230.

Dan, I., Ong, S. E., Watanabe, N. M., Blagoev, B., Nielsen, M. M., Kajikawa, E.,Kristiansen, T. Z., Mann, M. and Pandey, A. (2002). Cloning of MASK, a novelmember of the mammalian germinal center kinase III subfamily, with apoptosis-inducingproperties. J. Biol. Chem. 277, 5929-5939.

Etienne-Manneville, S. and Hall, A. (2001). Integrin-mediated activation of Cdc42 controlscell polarity in migrating astrocytes through PKCzeta. Cell 106, 489-498.

Goudreault, M., D’Ambrosio, L. M., Kean, M. J., Mullin, M., Larsen, B. G., Sanchez,A., Chaudhry, S., Chen, G. I., Sicheri, F., Nesvizhskii, A. I. et al. (2008). A PP2Aphosphatase high-density interaction network identifies a novel striatin-interactingphosphatase and kinase complex linked to the cerebral cavernous malformation 3 (CCM3)protein. Mol. Cell. Proteomics 8, 157-171.

Hilder, T. L., Malone, M. H., Bencharit, S., Colicelli, J., Haystead, T. A., Johnson, G.L. and Wu, C. C. (2007). Proteomic identification of the cerebral cavernous malformationsignaling complex. J. Proteome Res. 6, 4343-4355.

Hogan, B. M., Bussmann, J., Wolburg, H. and Schulte-Merker, S. (2008). Ccm1 cellautonomously regulates endothelial cellular morphogenesis and vascular tubulogenesisin zebrafish. Hum. Mol. Genet. 17, 2424-2432.

Huang, C. Y., Wu, Y. M., Hsu, C. Y., Lee, W. S., Lai, M. D., Lu, T. J., Huang, C. L.,Leu, T. H., Shih, H. M., Fang, H. I. et al. (2002). Caspase activation of mammaliansterile 20-like kinase 3 (Mst3). nuclear translocation and induction of apoptosis. J. Biol.Chem. 277, 34367-34374.

Hurd, T. W., Fan, S., Liu, C. J., Kweon, H. K., Hakansson, K. and Margolis, B. (2003).Phosphorylation-dependent binding of 14-3-3 to the polarity protein Par3 regulates cellpolarity in mammalian epithelia. Curr. Biol. 13, 2082-2090.

Kleaveland, B., Zheng, X., Liu, J. J., Blum, Y., Tung, J. J., Zou, Z., Chen, M., Guo,L., Lu, M. M., Zhou, D. et al. (2009). Regulation of cardiovascular development andintegrity by the heart of glass-cerebral cavernous malformation protein pathway. Nat.Med. 15, 169-176.

Labauge, P., Denier, C., Bergametti, F. and Tournier-Lasserve, E. (2007). Genetics ofcavernous angiomas. Lancet Neurol. 6, 237-244.

Laberge-le Couteulx, S., Jung, H. H., Labauge, P., Houtteville, J. P., Lescoat, C.,Cecillon, M., Marechal, E., Joutel, A., Bach, J. F. and Tournier-Lasserve, E. (1999).

Truncating mutations in CCM1, encoding KRIT1, cause hereditary cavernous angiomas.Nat. Genet. 23, 189-193.

Ling, P., Lu, T. J., Yuan, C. J. and Lai, M. D. (2008). Biosignaling of mammalian Ste20-related kinases. Cell. Signal. 20, 1237-1247.

Liquori, C. L., Berg, M. J., Siegel, A. M., Huang, E., Zawistowski, J. S., Stoffer, T.,Verlaan, D., Balogun, F., Hughes, L., Leedom, T. P. et al. (2003). Mutations in a geneencoding a novel protein containing a phosphotyrosine-binding domain cause type 2cerebral cavernous malformations. Am. J. Hum. Genet. 73, 1459-1464.

Liquori, C. L., Berg, M. J., Squitieri, F., Ottenbacher, M., Sorlie, M., Leedom, T. P.,Cannella, M., Maglione, V., Ptacek, L., Johnson, E. W. et al. (2006). Low frequencyof PDCD10 mutations in a panel of CCM3 probands: Potential for a fourth CCM locus.Hum. Mutat. 27, 118.

Lu, T. J., Lai, W. Y., Huang, C. Y., Hsieh, W. J., Yu, J. S., Hsieh, Y. J., Chang, W. T.,Leu, T. H., Chang, W. C., Chuang, W. J. et al. (2006). Inhibition of cell migration byautophosphorylated mammalian sterile 20-like kinase 3 (MST3) involves paxillin andprotein-tyrosine phosphatase-PEST. J. Biol. Chem. 281, 38405-38417.

Ma, X., Zhao, H., Shan, J., Long, F., Chen, Y., Chen, Y., Zhang, Y., Han, X. and Ma,D. (2007). PDCD10 interacts with Ste20-related kinase MST4 to promote cell growthand transformation via modulation of the ERK pathway. Mol. Biol. Cell 18, 1965-1978.

Moriarity, J. L., Clatterbuck, R. E. and Rigamonti, D. (1999). The natural history ofcavernous malformations. Neurosurg. Clin. N. Am. 10, 411-417.

Nobes, C. D. and Hall, A. (1999). Rho GTPases control polarity, protrusion, and adhesionduring cell movement. J. Cell Biol. 144, 1235-1244.

Nogueira, E., Fidalgo, M., Molnar, A., Kyriakis, J., Force, T., Zalvide, J. and Pombo,C. M. (2008). SOK1 translocates from the Golgi to the nucleus upon chemical anoxiaand induces apoptotic cell death. J. Biol. Chem. 283, 16248-16258.

Pagenstecher, A., Stahl, S., Sure, U. and Felbor, U. (2009). A two-hit mechanism causescerebral cavernous malformations: Complete inactivation of CCM1, CCM2 or CCM3in affected endothelial cells. Hum. Mol. Genet. 18, 911-918.

Pombo, C. M., Bonventre, J. V., Molnar, A., Kyriakis, J. and Force, T. (1996). Activationof a human Ste20-like kinase by oxidant stress defines a novel stress response pathway.EMBO J. 15, 4537-4546.

Pombo, C. M., Force, T., Kyriakis, J., Nogueira, E., Fidalgo, M. and Zalvide, J. (2007).The GCK II and III subfamilies of the STE20 group kinases. Front. Biosci. 12, 850-859.

Preisinger, C., Short, B., De Corte,V., Bruyneel, E., Haas, A., Kopajtich, R., Gettemans,J. and Barr, F. A. (2004). YSK1 is activated by the Golgi matrix protein GM130 andplays a role in cell migration through its substrate 14-3-3{zeta}. J. Cell Biol. 164, 1009-1020.

Rual, J. F., Venkatesan, K., Hao, T., Hirozane-Kishikawa, T., Dricot, A., Li, N., Berriz,G. F., Gibbons, F. D., Dreze, M., Ayivi-Guedehoussou, N. et al. (2005). Towards aproteome-scale map of the human protein-protein interaction network. Nature 437, 1173-1178.

Stahl, S., Gaetzner, S., Voss, K., Brackertz, B., Schleider, E., Surucu, O., Kunze, E.,Netzer, C., Korenke, C., Finckh, U. et al. (2008). Novel CCM1, CCM2, and CCM3mutations in patients with cerebral cavernous malformations: In-frame deletion in CCM2prevents formation of a CCM1/CCM2/CCM3 protein complex. Hum. Mutat. 29, 709-717.

Vinogradova, T., Miller, P. M. and Kaverina, I. (2009). Microtubule network asymmetryin motile cells: Role of Golgi-derived array. Cell Cycle 8, 2168-2174.

Voss, K., Stahl, S., Schleider, E., Ullrich, S., Nickel, J., Mueller, T. D. and Felbor, U.(2007). CCM3 interacts with CCM2 indicating common pathogenesis for cerebralcavernous malformations. Neurogenetics 8, 249-256.

Whitehead, K. J., Plummer, N. W., Adams, J. A., Marchuk, D. A. and Li, D. Y. (2004).Ccm1 is required for arterial morphogenesis: Implications for the etiology of humancavernous malformations. Development 131, 1437-1448.

Whitehead, K. J., Chan, A. C., Navankasattusas, S., Koh, W., London, N. R., Ling,J., Mayo, A. H., Drakos, S. G., Marchuk, D. A., Davis, G. E. et al. (2009). Thecerebral cavernous malformation signaling pathway promotes vascular integrity via rhoGTPases. Nat. Med. 15, 177-184.

Yadav, S., Puri, S. and Linstedt, A. D. (2009). A primary role for Golgi positioning indirected secretion, cell polarity and wound healing. Mol. Biol. Cell 20, 1728-1736.

Zabramski, J. M., Henn, J. S. and Coons, S. (1999). Pathology of cerebral vascularmalformations. Neurosurg. Clin. N. Am. 10, 395-410.

Zhou, J., Shao, Z., Kerkela, R., Ichijo, H., Muslin, A., Pombo, C. and Force, T. (2009).Serine 58 of 14-3-3{zeta} is a molecular switch regulating ASK1 and oxidant stress-induced cell death. Mol. Cell. Biol. 29, 4167-4176.

Jour

nal o

f Cel

l Sci

ence