Cdc42 is required for cytoskeletal support of endothelial cell adhesion during blood vessel formation in mice

Blood vessel formation is essential for embryonic development, growth and viability, as endothelial tubes allow exchange of nutrients and waste. The mammalian vasculature takes shape in the extraembryonic yolk sac and shortly thereafter in the embryo proper. Vessels form either de novo (vasculogenesis) or by sprouting from pre-existing vessels (angiogenesis). Endothelial cell (EC) progenitors, or angioblasts, emerge in the mesoderm at embryonic day 8 (E8), assembling into vascular cords. Angioblasts then differentiate, becoming ECs upon lumen formation or tubulogenesis. Cord ECs form a central lumen allowing passage of blood. The primary vascular system is then extended via angiogenesis (Risau and Flamme, 1995). Thus, formation of blood vessels is a complex multistep process. Elucidating the molecular bases of how ECs dynamically coordinate cell shape and adhesion to drive blood vessel morphogenesis is a central question in vascular biology and is essential to development of pro- and anti-angiogenic therapies.

The Rho GTPase cell division control protein 42 (Cdc42) has emerged as a crucial regulator of blood vessel formation and tubulogenesis. Over a decade ago, studies showed that it was essential for in vitro EC lumen formation (Bayless and Davis, 2002; Davis et al., 2011). Cdc42 was then shown to be similarly required for lumen formation in epithelial systems, both in vitro (Bray et al., 2011; Bryant et al., 2010; Martin-Belmonte and Mostov, 2007) and in vivo (Kesavan et al., 2009; Melendez et al., 2011). Many functions have since been ascribed to Cdc42, from regulation of exocytosis and apical membrane biogenesis during tubulogenesis (Bryant et al., 2010) to ADAM17-mediated VEGFR2 (Kdr – Mouse Genome Informatics) shedding (Jin et al., 2013). During mammalian tissue development, Cdc42 is ubiquitously expressed at the transcriptional level (see genepaint.org for a digital atlas of gene expression patterns in the mouse) and has been shown to be required for the development of many tissues, including the embryonic vasculature (Bray et al., 2011; Garvalov et al., 2007; Jin et al., 2013; Reginensi et al., 2013; Yang et al., 2007a). However, it remains unclear which cellular events Cdc42 controls and how it supports blood vessel morphogenesis.

An important role for Cdc42 is its control of the cell cytoskeleton. Cdc42 was discovered in yeast decades ago and was shown to be required for localization of budding sites, due to its influence on the actin cytoskeleton (Johnson and Pringle, 1990). Indeed, many subsequent studies have underscored how Cdc42 control of the cytoskeleton and actomyosin contractility is essential to proper organogenesis (Melendez et al., 2011). Deletion of Cdc42 in embryonic stem cells, for instance, results in disorganized filamentous actin (F-actin) and failed PIP2-induced actin polymerization (Chen et al., 2000). Actin, in turn, is essential to numerous cellular processes, including cell adhesion, migration, filopodia formation, endocytic trafficking and more (Adams et al., 1990; Johnson and Pringle, 1990; Yang et al., 2007b). Cdc42 also regulates a multitude of proteins known to affect actin organization, including Pak2, Pak4, Par6 (Pard6a), MLCK (Mylk2), MRCK (Cdc42bpa), N-WASP (Wasl), IRSp53 (Baiap2), IQGAP (Iqgap1), mDia2 (Diap3) and cofilin (Cfl1) (Downs et al., 2001; Fukata et al., 2002; Garvalov et al., 2007; Gomes et al., 2005; Kučera et al., 2009; Meadows et al., 2012; Rohatgi et al., 1999; Stratman and Davis, 2012). The question arises as to whether and how Cdc42 regulates actin organization in ECs during vascular development.

Cdc42 also influences cell adhesion. Cell junctions, both cell-cell and cell-extracellular matrix (ECM), are anchored to the cytoskeleton and are impaired in the absence of Cdc42 (Allen et al., 1997; Fukuhara et al., 2003). Loss of Cdc42 in adult hematopoietic stem cells (HSCs) results in cytoskeletal and adhesion defects that increase bone marrow niche egress (Yang et al., 2007a). Cdc42 was also found to control intercellular gaps between ECs, regulating vessel permeability (Broman et al., 2007). One study showed that activation of Cdc42 could restore blood vessel barrier function by re-establishing EC-EC adherens junctions (AJs) following thrombin disruption (Kouklis et al., 2004). Importantly, these studies raise the possibility that Cdc42 regulates the cytoskeleton, and in turn cell adhesion, in embryonic vessels.

Here, we show how endothelial Cdc42 regulates cell adhesion, cell shape and polarity via control of cytoskeletal organization during vessel morphogenesis. We genetically ablated Cdc42 in vessels during embryonic and postnatal development, and find that Cdc42 is required in different vascular beds, in different ways, for their formation and maintenance. Early ablation of Cdc42 in yolk sac vasculature blocks both angioblast coalescence and lumen formation, whereas later deletion impairs vessel integrity. Upon loss of Cdc42, F-actin becomes disorganized, resulting in failure of endothelial cell-cell and cell-ECM adhesion. We model these findings in cultured ECs and observe that, similar to in vivo observations, Cdc42 regulates EC actin organization and adhesion. By contrast, deletion of Cdc42 in postnatal vessels impairs filopodia formation and plexus remodeling. We propose that Cdc42 controls these processes, at least partially, through N-WASP, a regulator of actin polymerization, and Pak2/4-mediated NMIIA signaling, which controls actin contractility and crosslinking. Together, this work elucidates how Cdc42 regulates the ability of ECs to interact with each other to drive morphogenesis.

To determine its role in early vessels, Cdc42 was deleted in ECs by crossing a conditional allele of Cdc42 (Cdc42^f/f) with the Tie2-Cre driver line (or Cdc42^Tie2KO), which is reported to recombine in ECs as they emerge from the mesoderm (Sato et al., 1993; Yang et al., 2006). All Cdc42^Tie2KO embryos died in utero, whereas Cdc42^f/+;Tie2-Cre (Cdc42^Tie2Het) littermates were viable and born at normal Mendelian ratios. Analysis of mid-gestation embryos revealed that Cdc42^Tie2KO embryos died of blood vessel defects between E9 and E10.5 (Fig. 1A,A′), as previously reported (Jin et al., 2013). Mutant embryos varied in size at E9 (supplementary material Fig. S1A) but were markedly smaller by E10, displaying hemorrhages and focal blood pooling throughout both the embryo proper and extraembryonic yolk sac (Fig. 1B,B′; supplementary material Fig. S1B). Blood vessel remodeling in the yolk sac failed, resulting in large ‘bags' of blood (Fig. 1B,B′).

To visualize embryonic blood vessels, anti-PECAM immunofluorescent staining was carried out in E8.25 (six somites) (Fig. 1C-C″) and E9.0 (12 somites) embryos (Fig. 1D-E″). The paired dorsal aortae, which form at E8.0 as angioblasts coalesce in parallel linear aggregates, are the first embryonic vessels to develop. All Cdc42^Tie2KO embryos exhibited defective aortae, albeit of variable severity, between E8.25 and 9.5 when compared with Cdc42^Tie2Het controls. At this stage, 30% of the mutants exhibited a severe phenotype with fragmented junctions (Fig. 1C′,D′,E′,F′) characterized by focal aggregates of PECAM at cell-cell boundaries (Fig. 1E′), whereas 70% exhibited relatively intact junctions (Fig. 1C″,D″,E″,F″). Overall, however, aortae of both phenotypes displayed a significant number of lumen occlusions (25%), as seen with PECAM or an Flk1(VEGFR2)-eGFP reporter (Fig. 1E″,F″).

Because yolk sac vessels failed to form proper lumens and were discontinuous, we assayed blood circulation upon loss of Cdc42. Indeed, circulation in the early mouse embryo occurs as a loop between the aortae and yolk sac vessels, as no veins are present in embryonic tissues at this stage (Chong et al., 2011; Jones, 2011). To assess blood circulation, India ink was microinjected into the inflow tracts leading into the heart of live E9.5 Cdc42^Tie2Het and Cdc42^Tie2KO embryos. The heart of both controls and mutant embryos beat at similar intervals and with similar strength (5-6 beats per 10 s). In control embryos, ink traveled rapidly through the pumping heart into the aortae (supplementary material Movie 1), while it failed to do so in Cdc42^Tie2KO embryos (supplementary material Movie 2), suggesting that embryos died early due to impaired blood circulation. These findings suggest that Cdc42 is crucial to embryonic circulation via its requirement for vessel continuity.

Occasional constricted aortic lumens were not likely to account for the complete loss of embryonic circulation. We hypothesized that failed Cdc42^Tie2KO yolk sac vessels might in turn impact embryonic circulation. At these early stages, blood flow passes from heart to aortae, completing the circulatory loop with return of blood to the heart via yolk sac vessels (Chong et al., 2011; Jones, 2011; Sacharidou et al., 2012). We assessed this vascular bed in Cdc42^Tie2Het and Cdc42^Tie2KO yolk sacs. Control yolk sacs developed characteristic blood islands in distal regions (nearest maternal tissues), which were continuous with a honeycomb-like plexus of vascular cords that extended to the embryo (proximal region, in vicinity to embryo) (Fig. 2A). Of note, lumens opened first distally and slightly later proximally (Fig. 2A,B,C,D). By contrast, Cdc42^Tie2KO embryos were unable to form a proper vascular plexus in the proximal yolk sac (Fig. 2A′) and displayed disorganized angioblasts (Fig. 2B′) and fewer patent lumens (Fig. 2C′). In the distal yolk sac, large disorganized blood-filled vascular structures were observed, expanding by E9-10 (Fig. 2D′).

To establish whether yolk sac vessels experienced blood circulation, we examined expression of Connexin40 (CX40; Gja5 – Mouse Genome Informatics), an arterial marker that depends on blood flow and can be used to assess vessel functionality (Stalmans et al., 2002). In situ hybridization showed that control yolk sac vessels had normal CX40 expression, indicating normal blood flow, and vascular remodeling (presence of large and small vessels) (Fig. 2E). By contrast, Cdc42^Tie2KO yolk sacs showed no CX40 expression (Fig. 2E′), suggesting failure of blood flow.

To determine whether yolk sac vessel defects in Cdc42^Tie2Het and Cdc42^Tie2KO were cell autonomous or primarily due to blocked circulation, we examined angioblast assembly in allantois explant cultures, in which vessels experience no blood flow (Downs et al., 2001). Control allantoises stained with PECAM developed characteristic net-like plexuses (Fig. 2F). By contrast, allantoises from Cdc42^Tie2KO embryos exhibited discontinuous cords and aggregated ECs (Fig. 2F′). This ex vivo phenotype resembled vessels in the yolk sac and suggested that Cdc42 is necessary for fundamental EC characteristics that are EC-intrinsic and separate from blood flow.

Given previous work demonstrating the role of Cdc42 in EC lumen formation (Sacharidou et al., 2012), the presence of aortic lumens following Tie2-Cre deletion of Cdc42 was unexpected. Cdc42 is required for tubulogenesis in human umbilical vein endothelial cells (HUVECs) cultured in three-dimensional (3D) matrices, yet loss of lumens in vivo was primarily restricted to the proximal yolk sac. We hypothesized that timing of Cdc42 deletion relative to EC lumen formation might account for the presence of lumens and variability of phenotypes. We therefore characterized Cdc42 deletion efficiency by assessing Cdc42 protein in Cdc42^Tie2KO;Rosa26-YFP and found that Cdc42 was deleted earlier in yolk sac than in embryonic vessels (Fig. 3A). At E8.0, Cdc42 protein was present in aortic ECs (Fig. 3B-C′), whereas expression was significantly decreased in ECs of both distal (Fig. 3D-E′) and proximal yolk sac (Fig. 3F-G′). These data probably explain why lumen defects are observed primarily in this tissue, where Cdc42 is deleted first.

We further evaluated loss of Cdc42 in different vascular beds by characterizing Cre deletion driven by Tie2. At E8.25, YFP was expressed in 89% of proximal yolk sac ECs and 81% of aortic ECs in both Cdc42^Tie2Het and Cdc42^Tie2KO embryos (Fig. 3H). Cdc42 immunostaining quantification showed that Cdc42 was absent from 54% of proximal yolk sac ECs (Fig. 3F-G′,I), 40% of distal yolk sac ECs and 20% of blood cells (Fig. 3D-E′,I), but was not significantly decreased in aortic ECs at E8.25 (5 somites) (Fig. 3B-C′,I). To confirm differential initiation of Cre expression in embryonic versus yolk sac ECs, Rosa26-lacZ reporter activity was assessed. Blood and angioblasts began to express Cre in the yolk sac at E7.75, but not in embryonic angioblasts until E8.0 (1-3 somites) (Fig. 3J-M). These results confirm that Cdc42^Tie2KO yolk sac ECs (and blood) express Cre and delete Cdc42 earlier than embryonic ECs; this timing of deletion probably explains why lumen formation is affected in yolk sac but not in aortic ECs.

We next assayed whether failed vessels in Cdc42^Tie2KO embryos were due to defects in EC proliferation or survival. We assessed phospho-histone H3 (pHH3) and cleaved caspase-3 expression and found that proliferation was decreased in the absence of Cdc42. Control vessels displayed 59% pHH3⁺ cells, whereas Cdc42^Tie2KO vessels displayed 42%, demonstrating a modest 20% proliferation reduction (supplementary material Fig. S2A-B). Reduction of Cdc42 by siRNA (siCdc42) (supplementary material Fig. S2C) in MS1 ECs, however, did not alter total levels of pHH3 (supplementary material Fig. S2D). Apoptosis, by contrast, was unchanged in either yolk sac or dorsal aortae of E8.25 Cdc42^Tie2KO (supplementary material Fig. S2E-F; and data not shown), suggesting that cell death did not cause vessel failure. By E9.5, however, hours after initiation of vascular defects, ∼14% of Cdc42^Tie2KO aortic ECs were apoptotic (supplementary material Fig. S2G-J), along with non-vascular cells that flanked the aortae. Given the timing of appearance of apoptotic ECs in Cdc42^Tie2KO embryos, we propose that cells undergo apoptosis as secondary effects from failed circulation. Similarly, siCdc42-treated MS1s showed no increase in levels of cleaved caspase-3 (supplementary material Fig. S2K). In contrast to previous findings (Jin et al., 2013), our results suggest that loss of Cdc42 has a modest effect on endothelial proliferation, but not directly on survival.

To determine whether Cdc42 was necessary for blood vessel growth in later embryonic development, we assayed E14.5 embryos after inducible deletion of Cdc42, using the tamoxifen (TM)-inducible vascular driver line Cad5-Cre^ERT2 (Wang et al., 2010). Cdc42^Cad5KO;R26R-YFP mice were TM-induced to delete Cdc42 for 24, 48 or 72 h at different embryonic stages. Mid-gestation deletion of Cdc42 for 72 h resulted in 100% lethality (resorbing embryos); deletion for 48 h produced viable embryos with widespread hemorrhages (Fig. 4A-B′); and deletion for 24 h displayed no evident vascular defects (Cdc42 protein was not deleted in this timeframe). To examine EC lumens and polarity in developing capillaries, dermis of 48 h-induced embryos were sectioned and stained for YFP, moesin (an ERM family protein) and podocalyxin (PODXL, a sialomucin), the latter two previously shown to be polarized apically in capillaries (Strilić et al., 2009). Control vessels showed patent lumens and apical PODXL localization, whereas half of the Cdc42^Cad5KO;R26R-YFP capillaries failed to open lumens and possessed intracellular PODXL aggregates (Fig. 4C-H). These results suggest that Cdc42 is necessary for vessel lumen formation and EC polarity during angiogenesis.

To examine further the role of Cdc42 during maintenance and expansion of vascular beds after birth, we induced Cdc42 deletion in pups from postnatal days 0 to 4 (P0-4). Surprisingly, Cdc42^Cad5KO pups survived and maintained normal weights after TM induction, with no evident hemorrhages. This observation suggested that, although vascular Cdc42 is essential for embryonic vessel growth, postnatal blood vessels did not require Cdc42 for integrity during this timeframe.

By contrast, Cdc42 was required in rapidly growing vessels of the retina. Cdc42^Cad5KO retinas displayed markedly aberrant blood vessel formation, with defects in both sprouting angiogenesis and remodeling angiogenesis. Control retinal blood vessels grew normally from the center of the retina to the periphery, with EC sprouts around the leading edge of the vasculature (Fig. 5A,B). In Cdc42^Cad5KO;R26R-YFP retinas, although vessels extended approximately the same distance (radius) from the center (Fig. 5C), ECs along the growing front (periphery) were constricted (Fig. 5A′,B′) and central vessels exhibited patches of swollen and blood-filled vessels, as well as high vessel density (Fig. 5D-G) that consistently correlated with YFP expression (Fig. 5F,F′). Of note, lumens were present in these vessels.

In addition, we assessed vascular remodeling (transformation into tree-like large and small vessels) by quantifying large vessels extending from the retina center. During remodeling, vessels emanating from the center of the retina are normally pruned (Stalmans et al., 2002). In Cdc42^Cad5Het embryos, we routinely observed between 13 and 15 large vessels, alternating between arteries and veins (Fig. 5H; and data not shown). By contrast, in Cdc42^Cad5KOs, the number of large vessels was significantly increased, with between 16 and 23 vessels, suggesting reduction of vascular remodeling (Fig. 5H-I). Cdc42 is thus necessary for proper outgrowth and remodeling of the vasculature during retinal angiogenesis.

Because filopodia are thought to be important for angiogenesis (Gerhardt et al., 2003) and Cdc42 has been shown to be necessary for filopodial development (Chauhan et al., 2009; Mattila and Lappalainen, 2008), we assessed sprouting EC filopodia in Cdc42^Cad5KO retinas. Whereas Cdc42^Cad5Het retinal vessels produced many filopodia extending into the avascular periphery (Fig. 5J,K), tip cells in Cdc42^Cad5KO retinas produced markedly fewer filopodia (Fig. 5J′,K′,L) and narrower sprouts (Fig. 5M; supplementary material Fig. S3) that displayed normal length and sprouting angles. In vitro, filopodia formation and EC sprouting into 3D collagen matrices were also significantly reduced upon loss of Cdc42 (Fig. 5N-O′). When a monolayer of HUVECs was plated atop collagen matrices and allowed to invade the underlying gel, both the number of invading sprouts and their depth of penetration were reduced (supplementary material Fig. S4). Cdc42 is thus necessary for both sprouting of new vessels and filopodia development.

To elucidate how loss of Cdc42 might impact developing vessels at the cellular level, PECAM1 expression was examined using high magnification confocal microscopy on whole-mount E8.25 Cdc42^Tie2Het and Cdc42^Tie2KO dorsal aortae. In heterozygotes, PECAM⁺ adhesions lined EC-EC interfaces, while Cdc42^Tie2KO ECs exhibited disorganized focal aggregates (Fig. 6A,A′). ECs were scored for the presence of aggregates and whether ECs were completely surrounded by PECAM or displayed irregular boundaries. In Cdc42^Tie2KO embryos, 43% of ECs were not completely surrounded by PECAM (Fig. 6B), and ∼77% possessed aggregates (Fig. 6C), while in Cdc42^Tie2Het, 100% were normally surrounded by PECAM and only 21% displayed aggregates. Similarly, ECs in Cdc42^Tie2KO allantois cultures displayed adhesion aggregates, as seen with the tight-junction marker ZO-1 (Tjp1 – Mouse Genome Informatics; supplementary material Fig. S5). Cultured MS1s treated with siCdc42 displayed similar aggregates and severely disrupted cell-cell junctions, as visualized with the AJ marker VE-cadherin (VEcad; Cdh5 – Mouse Genome Informatics) (Fig. 6D-G). Interestingly, junction aggregates colocalized with F-actin (phalloidin). These data underscored how cell-cell junctions are highly sensitive to loss of Cdc42.

To test adhesion of ECs to ECM after in vivo Cdc42 depletion, we assessed the focal adhesion marker phosphorylated Paxillin (pPax, Y118; pPxn – Mouse Genome Informatics) (Sachdev et al., 2009). Cdc42^Tie2KO vessels displayed reduced pPax staining at the basal membrane of aortic ECs at E9.5 compared with Cdc42^Tie2Het embryos (Fig. 6H-I′). Basal collagen IV was also reduced upon Cdc42 loss (Fig. 6J-K′). Similarly, cultured MS1s treated with siCdc42 exhibited reduced pPax staining (Fig. 6L-N′). Whereas control MS1s showed many focal adhesions anchored to actin fibers, MS1s treated with siCdc42 displayed significantly fewer focal adhesions, and pPax colocalized with globular aggregates of F-actin instead. Immunoblot analysis revealed an overall reduction of pPax following siCdc42 treatment (Fig. 6O). These results show that aberrant EC adhesion develops after Cdc42 reduction, both in vitro and in vivo.

Because Cdc42 deletion from ECs disrupted both cell shape and adhesion, we hypothesized that Cdc42 regulates the actin cytoskeleton in ECs, as reported for other cell types, such as macrophages (Allen et al., 1997) and neural progenitor cells (Katayama et al., 2013). We therefore examined F-actin in Cdc42^Tie2KO ECs. We found that F-actin was normally enriched basolaterally in aortic ECs (Fig. 7A,B), whereas Cdc42^Tie2KOs displayed punctate F-actin, scattered throughout the cytoplasm (Fig. 7A′,B′). F-actin staining in mutant ECs was also reduced in overall intensity, suggesting reduced and/or improperly organized F-actin (Fig. 7C). Similarly, HUVECs grown in 3D ECM matrices treated with siCdc42 (Fig. 7D) exhibited punctate and aggregated F-actin compared with controls, which displayed smooth and basally polarized F-actin (Fig. 7E,E′). This cytoskeletal disruption was always coincident with tubulogenesis failure, as previously described (Bayless and Davis, 2002).

We also characterized F-actin anchoring of EC-EC junctions. Cell interfaces between control confluent MS1s displayed three types of morphology: either broad lamellipodia-like junctions with low levels of phalloidin staining (18%) (Fig. 7F-F″), junctions with prominent actin fibers or junctional cortex protrusions that were tethered to VEcad junctions (47%) (Fig. 7G-G″), or smoother junctions with cortical F-actin flanking VEcad junctions (35%) (Fig. 7H,H′). Junctional cortex protrusions (jagged interface) have been reported to be crucial in establishment and maturation of EC-EC junctions (Bentley et al., 2014). Cells treated with siCdc42 strikingly displayed none of these characteristics, but instead produced broad, disorganized and fragmented junctions with only punctate F-actin, suggesting aberrant F-actin-mediated junction formation (Fig. 7I,I′). To examine further how F-actin is regulated by Cdc42 activity, we treated MS1s with constitutively active Cdc42 (V12) adenovirus or GFP control adenovirus. In contrast to GFP-infected ECs, Cdc42V12-infected ECs exhibited increased actin fiber development and elevated overall fluorescence intensity (Fig. 7J-L). This result further suggested that endothelial Cdc42 promotes F-actin development and organization.

To model disruption of the EC cytoskeleton seen upon loss of Cdc42, we used a pharmacological approach whereby confluent cells were treated with Cytochalasin D, preventing actin polymerization (Casella et al., 1981). After treatment, EC-EC adhesions marked by VEcad became extensively fragmented (supplementary material Fig. S6). This phenotype mimicked the EC-EC junction disruption that we observed, both in vitro and in vivo following Cdc42 reduction.

To examine the mechanism by which Cdc42 might regulate actin organization in ECs, we evaluated the actomyosin machinery, known to crosslink and organize actin (Wirtz and Khatau, 2010). We found that aortic ECs displayed reduced phosphorylation of myosin light chain (pMLC) at serine 19 when Cdc42 expression was lost (Fig. 8A-C). To examine molecular cascades between Cdc42 and actin in ECs, we carried out experiments in cultured ECs using siRNA. As expected, loss of Cdc42 led to decreased phosphorylation of its downstream effectors Pak2 and Pak4 (Fig. 8D-G). Of interest, these kinases are known to phosphorylate and activate MLC (Zeng et al., 2000), and knockdown of either kinase alone or in combination led to loss of pMLC in ECs (Fig. 8H-J). In addition, loss of either NMHCIIA or MLCK in turn resulted in significant disruption of the actin cytoskeleton (Fig. 8K-L). Loss of Cdc42 did not result in reduced phosphorylation of the Cdc42 effector and known actin nucleating factor N-WASP, but did disrupt its recruitment to sites of VEcad⁺ EC-EC adhesion (Fig. 8M-P). Knockdown of N-WASP by siRNA also resulted in disruption of actin anchoring of EC-EC adhesions (Fig. 8Q-R″). Whereas siRNA knockdown of MRCK, mDia2, IQGAP, Par6 or IRSp53 by contrast did not affect the EC cytoskeleton or EC-EC adhesions (supplementary material Fig. S7), it is possible that these act in parallel or via additional unknown effectors. Together, these findings suggest that an important function of endothelial Cdc42 and its effectors is to regulate the cytoskeleton, which in turn supports cell adhesion.

In this article, we tested when, where and why Cdc42 is required during blood vessel formation. We find that Cdc42 is essential in ECs both during vessel development and maintenance, throughout mouse gestational development, as well as postnatal retinal growth. Defects upon Cdc42 deletion with Tie2-Cre initiate in yolk sac vessels, where blood circulation is blocked and growing vessels fail to open lumens or rapidly tear, resulting in embryonic lethality. By contrast, loss of Cdc42 during postnatal stages suppresses angiogenic remodeling, filopodia formation and sprouting. Specifically, we find that loss of Cdc42 in ECs, both in vitro and in vivo, leads to rapid loss of actin organization. As the cytoskeleton anchors junctions, cell adhesion is disrupted both between cells and between ECs and ECM. These findings support Cdc42 as a key molecular player in blood vessel morphogenesis and delineate its spatiotemporal requirement for basic EC functions.

A key question from this study is how does loss of Cdc42 impact EC lumen formation? Cdc42 is essential for lumen formation and cell polarity in cultured ECs (Bayless and Davis, 2002) and epithelial tubule-forming tissues (Kesavan et al., 2009). Here, two different Cre lines are used to delete Cdc42 at distinct stages of vessel formation. Deletion with the widely used Tie2-Cre line (Cdc42^Tie2KO) leads to defects that initiate in the yolk sac, where Tie2-Cre deletes early and efficiently. Interestingly, whereas proximal yolk sac vessels fail to open lumens, vessels in the distal yolk sac expand into large bag-like structures. Interestingly, work by Ferkowicz and Yoder shows that a vascular plexus forms first in the proximal yolk sac, and then extends towards the ectoplacental cone to enclose blood and form blood islands (Ferkowicz and Yoder, 2005). Proximal yolk sac cords open lumens at their centers, while distal yolk sac ECs surround and encapsulate blood. We speculate that the disparate failures observed following Cdc42 deletion reflect these morphogenetic differences.

The distinct Cdc42-dependent tubulogenesis defects that we observed are notable in that they contrast with previous descriptions of endothelial Cdc42 ablation. Tie2-Cre deletion of Cdc42 by one group was reported to result in embryos with fewer vessels, but that developed relatively normally to E12.5 prior to lethality (Hu et al., 2011). We never recovered mutant embryos past stages E9-10. Another study reported that EC deletion of Cdc42 resulted in absence of vessels in the yolk sac as a result of EC apoptosis via an ADAM17/VEGFR2-dependent mechanism (Jin et al., 2013). We never observed significant EC apoptosis, instead finding later non-vascular cell death, probably due to secondary effects resulting from circulation failure. Differences in Cdc42^ECKO studies might reflect different genetic backgrounds, or, alternatively, the different conditional alleles used.

Another cause of heterogeneous vascular failures in Cdc42^Tie2KO is the timing of Cre-mediated deletion. We find that ablation of Cdc42 using most available Cre driver lines results in progressive and mosaic depletion. We propose that deletion dynamics using EC-Cre lines impact interpretation of phenotypes and must therefore be carefully characterized. Here, using R26R-lacZ and R26R-YFP, as well as staining for Cdc42 protein, we track the onset and spatiotemporal dynamics of Cdc42 deletion with widely used endothelial Cre lines. Notably, gene deletion does not occur rapidly or homogenously in emerging angioblasts and cords, and hence produces varying phenotypes. These findings pose a cautionary note, as genes deleted with reportedly early blood vessel-specific Cre lines probably fail to efficiently deplete protein products prior to lumen formation, precluding their analysis in that process.

Our study suggests that a major Cdc42 function in embryonic endothelium is regulation of vessel morphogenesis via the cytoskeleton. Cdc42 regulates the actin cytoskeleton in many ways, including actin polymerization, depolymerization, branching, bundling and contractility, in many cell types (Derivery and Gautreau, 2010; Peng et al., 2003; Sumi et al., 1999; Wilkinson et al., 2005). We show that loss of EC Cdc42 results in dramatic disruption of actin organization, both in vivo and in vitro. When constitutively activated Cdc42 is expressed in MS1s, an excess of actin fibers develops within single ECs, underscoring its profound influence on actin. Moreover, we identify the Cdc42 effectors Pak2 and Pak4 as key regulators of actin via their control of the actomyosin machinery. Specifically, we show that pMLC requires both Cdc42 and Pak2/4; we demonstrate that NMHCIIA and MLCK are in turn required for the actin cytoskeleton in ECs; and we also place N-WASP as important in anchoring of EC-EC junctions by actin (Fig. 9A). Therefore, the Cdc42 molecular signaling pathway is essential to basic EC biology.

We propose that Cdc42 support of the cytoskeleton in turn impacts establishment and maturation of EC junctions (Fig. 9B,C). EC junctions arise via highly regulated interactions with the cytoskeleton (Hoelzle and Svitkina, 2012). Treatment of ECs with the drug Cytochalasin D, which blocks actin polymerization, causes EC-EC junction fragmentation. Cell-cell associations are known to undergo a step-wise strengthening process, with ECs first overlapping, then retracting lamellipodia. At early stages of EC-EC adhesion, boundaries between cells are highly jagged and have been referred to as ‘junctional cortex protrusions' (Bentley et al., 2014). Following retraction, ECs maintain contact through bridges formed by filopodia-like protrusions joined by VEcad-rich junctions, which are eventually strengthened by underlying cortical actin. EC-EC junctions then become smoother and stronger. Cdc42 ablation from embryoid body-derived ECs produced junctions that lacked filopodial-like structures and instead were rich in lamellipodia (Qi et al., 2011). These findings, combined with those from this study, suggest that Cdc42 supports crucial filopodial and other cortex protrusion structures necessary for EC junction dynamics. It still remains to be discovered how vascular Cdc42 regulates the organization of F-actin in different areas of the cell, as well as how exactly Cdc42 and actin support cell-cell and cell-ECM junction remodeling during blood vessel growth.

Cdc42 is fundamental for several vascular processes throughout development, and here we demonstrate a requirement for Cdc42 in EC polarity, adhesion and cytoskeletal organization. Small GTPases such as Cdc42 serve as a point of convergence for many signaling pathways, making GTPases potentially ideal targets manipulating the vasculature to ameliorate human diseases in which abnormal vessels are a primary pathogenic feature. Future studies will be necessary to uncover pathways regulated during blood vessel development downstream of Cdc42 in order to understand the function of Cdc42 in normal and pathological settings.

All animal husbandry was performed in accordance with protocols approved by the UT Southwestern Medical Center IACUC. Embryos were dissected and fixed in 4% PFA/PBS for 40 min at 4°C, then dehydrated to 75% ethanol for storage at −20°C.

For deletion of Cdc42, Cad5-Cre^ERT2 mothers were gavaged with 2 mg Tamoxifen (TM) per 40 g mouse for 72 h, 48 h or 24 h to assess vascular defects. Three litters were used to assess each time point. Epidermis from three E14.5 embryo littermates per genotype were sectioned (sampled every 100 μm). Five 160-μm² fields-of-view (FOV) were imaged on sections. For the 48 h TM-induced embryos, 146 ECs were counted in Cdc42^Tie2Het images and 142 ECs were counted in Cdc42^Tie2KO images. To induce retinal vessel Cdc42 deletion, mothers were gavaged with TM (3 mg TM/40 g mouse) twice daily at the onset of pup birth from P0 to P3. Retinas (n=4) were dissected at P4. Four litters were used to obtain pups and only littermates were compared. Specific numbers assessed are detailed in the figure legends.

Embryos were fixed using gluteraldehyde for 15 min, rinsed in PBS and stained for β-galactosidase (β-gal) overnight (O/N) as previously described (Villasenor et al., 2008). Images were taken with a NeoLumar stereomicroscope (Zeiss) using a DP-70 camera (Olympus).

Fixed tissues were washed (PBS), cryoprotected in 30% sucrose O/N, embedded in Tissue-Tek O.C.T. and sectioned. Sections were washed (PBS) and blocked (1 h RT 5% serum). Primary antibody incubations were performed at 4°C O/N (for dilutions, see supplementary material Table S1). Slides were washed (PBS), incubated in secondary antibody (4 h, RT). Slides were washed (PBS) and mounted using Prolong Gold Mounting Medium with DAPI. Images were obtained using a LSM510 or LSM710 Meta Zeiss confocal. TSA immunofluorescent staining (PerkinElmer; individual fluorescein tyramide reagent pack, cat# SAT701) was used to stain Cdc42.

Whole-mount staining was performed as previously described (Tang et al., 2011). Briefly, fixed embryos were stained using an ABC Elite reagent (Vector; VECTASTAIN Elite ABC Kit, cat# PK-6100) fluorescein tyramide reagent (PerkinElmer; individual fluorescein tyramide reagent pack, cat# SAT701) prepared in amplification diluent (see reagent pack instructions). Tissues were visualized after clearing in BABB using a Zeiss AxioObserver epifluorescence microscope.

Retinas were fixed (4% PFA/PBS, 1 h at RT), then dehydrated to 75% ethanol for storage at −20°C. Retinas were rehydrated in PBST (1% Triton X-100/PBS) and incubated in Isolectin Alexa Fluor 568 (Invitrogen, 1:100) (O/N at 4°C). The following day, retinas were incubated in 5% serum/PBST (1 h at RT), then with anti-GFP antibody (PBST O/N at 4°C). Retinas were washed (PBS), then treated with anti-chicken secondary antibody (3 h at 37°C). Finally, retinas were washed (PBS) and mounted onto coverslips with Prolong Gold Mounting Medium.

Mouse embryo staining was performed as previously described (Neumann et al., 2009; Xu et al., 2009). Connexin40 (Gja5) clone was obtained from Open Biosystems (MMM1013-9202306).

siGENOME siRNAs obtained from GE Dharmacon were transfected into cultured MS1 or HUVECs using standard protocols for transfection and western analysis blot, as previously described (Koh et al., 2008) (antibodies used for western blots are detailed in supplementary material Table S1 and siRNA sequences are detailed in supplementary material Table S2).

MS1s (American Type Culture Collection, ATCC) or HUVECs (ATCC) were fixed with 4% PFA, then rinsed with PBSN (0.1% NP40/PBS). Cells were permeabilized in PBSN, then blocked in CAS block (Invitrogen, 008120). Primary antibody dissolved in blocking solution (1:60) was added for 1 h. Afterwards, the coverslip was washed (PBSN), incubated with secondary antibody (1:200), washed (PBSN), then mounted with Prolong Gold anti-fade with DAPI (Invitrogen, P36931).

Allantois assay was performed as described (Crosby et al., 2005). Allantoises from E8.5 embryos were cultured ex vivo for 24 h, then fixed (4% PFA for 1 h) and stained with anti-PECAM antibodies. Stained allantoises were analyzed with a NeoLumar stereomicroscope (Zeiss).

ECs were seeded atop 2.5 mg/ml collagen type I containing 200 ng/ml SCF, IL-3, SDF-1α and FGF-2, according to Stratman et al. (2011). FGF-2 was included in medium (40 ng/ml). Cultures were fixed with 3% gluteraldehyde for 30 min, stained with 0.1% Toluidine Blue in 30% methanol and de-stained before photography of sprouting, depth of invasion and EC tip cell morphology.

All datasets were taken from n≥3 biological replicates (embryos or retinas), with n≥5-10 FOVs analyzed. Data are presented as mean±s.e.m. Quantification of cellular parameters, such as adhesion aggregates or actin bundles, was carried our using the cell counter plugin tool in ImageJ by defining structures using morphological parameters (i.e. actin bundles were defined as identifiable elongated rod-shaped phalloidin+structures). All statistical analysis was performed using two-tailed, unpaired Student's t-test in Graphpad Prism software. P-values lower than 0.05 were considered statistically significant.

We thank Ralf Adams for the Cad5-CreERT2 line, Tom Sato for the Tie2-Cre line, Janet Rossant for the Flk1-eGFP line and Thomas Carroll for the R26R-YFP line. We thank the Olson, MacDonald, Carroll and Cleaver labs for invaluable discussions and assistance. We thank Hiromi Yanagisawa for use of cell culture equipment and Stephen Fu for skilled technical support.