Defective endothelial cell migration in the absence of Cdc42 leads to capillary-venous malformations

The small GTPase cell division cycle 42 (Cdc42) is present in most eukaryotes, where it exerts basic functions in the regulation of actin-based morphogenesis and cell polarity (Boureux et al., 2007; Jaffe and Hall, 2005; Melendez et al., 2011). Cdc42 has also been implicated in a variety of other cellular processes in many different cell types (Etienne-Manneville, 2004; Hall, 1998; Heasman and Ridley, 2008; Melendez et al., 2011). The current comprehensive picture of Cdc42 function is inferred from an assemblage of data obtained from diverse cell types, species, experimental models and tools, which in most cases cannot be applied to any single cell type or stage of development. An example is angiogenesis – the process of new blood vessel formation from existing ones – and the role of Cdc42 in endothelial cells (ECs). Work with cultured ECs shows that Cdc42 regulates the actin cytoskeleton (Barry et al., 2015; Hall, 1998; Wójciak-Stothard et al., 1998), membrane protrusions (Kouklis et al., 2003), macro-pinocytosis (Tkachenko et al., 2004), adherence junctions and EC barriers (Broman et al., 2006), cell migration (Dormond et al., 2001; Sakabe et al., 2017; Vitorino and Meyer, 2008) polarization (Etienne-Manneville, 2004; Vitorino and Meyer, 2008) and tube formation (Bayless and Davis, 2002; Davis et al., 2011). It is unclear, however, if all of these cellular functions regulated by Cdc42 in vitro contribute to vascular morphogenesis in vivo.

Experiments in zebrafish and mice have provided some insights: in zebrafish, Cdc42 has been suggested to regulate filopodia formation and EC guidance mechanisms (Fantin et al., 2015; Kaur et al., 2006; Wakayama et al., 2015). In mice, constitutive [Tie2 (Tek – Mouse Genome Informatics)-Cre-mediated] EC-specific Cdc42 deletion leads to abnormal vascular morphogenesis, disrupted blood circulation and early embryonic death (Barry et al., 2015; Jin et al., 2013), making it difficult to discriminate between primary cellular consequences of Cdc42 deletion and secondary effects of a disrupted circulation. When Cdc42 deletion was instead induced at an early postnatal time point (Cdh5-Cre-mediated) and analyzed during retinal sprouting angiogenesis, it caused reduced EC filopodia formation (Barry et al., 2015; Sakabe et al., 2017), an effect also observed following pharmacological Cdc42 inhibition of Cdc42 (Fantin et al., 2015). Based on these observations, it has been suggested that Cdc42 plays a role in F-actin organization, with consequences for vascular apical basal polarity (Barry et al., 2015) and EC migration (Sakabe et al., 2017).

Here, using conditional and mosaic gene ablation, together with computational modeling, we aimed to further clarify which of the Cdc42-dependent cellular functions are relevant for vascular morphogenesis in the postnatal retina. Our work confirms some of the previously suggested EC roles of Cdc42 and adds a new role in axial polarization. We also show that loss of Cdc42 leads to vascular malformations specifically in capillaries and veins, locations in which EC proliferation is normally high. We suggest that these malformations arise as a consequence of the inability of Cdc42-depleted ECs to migrate away from these areas and disperse within the developing vascular network.

In order to avoid the secondary systemic consequences of embryonic vascular disruption, we induced EC-specific Cdc42 deletion in newborn mice and focused our analysis at the developing retina, an organ that allows simultaneous imaging of sprouting and remodeling of vessels with readily distinguishable arterio-venous hierarchy. The mouse retina is vascularized by angiogenic sprouting (Fruttiger, 2002), which starts at postnatal day (P) 0 and is finished around P25 (Stahl et al., 2010). Early postnatal gene deletion therefore targets the developing retinal vasculature at a time when many other vascular beds are already well developed and generally less susceptible to the consequences of gene deletion. We established mice in which EC-specific deletion of Cdc42 was tamoxifen-inducible [Cdh5(PAC)-Cre^ERT2;Cdc42^flox/flox] (Pitulescu et al., 2010; Wu et al., 2006) (hereafter referred to as Cdc42^iΔEC) and also introduced Rosa26-EYFP (Srinivas et al., 2001) or Rosa26-tdTomato(Ai14) (Madisen et al., 2010) reporter alleles to allow visualization of Cre-mediated recombination at cellular resolution. Daily administration of tamoxifen for 3 days (P0-P2) caused a robust, albeit not complete, loss of endothelial Cdc42 mRNA and protein at P7 and was chosen as the standard approach for our analysis (Fig. 1A-C). As shown below, the incomplete knockout (KO) of Cdc42 mRNA and protein was a result of cellular mosaicism, i.e. a vasculature composed of a mixture of normal and Cdc42-depleted ECs [hereafter referred to as wild-type (WT)-EC and Cdc42-KO-EC, respectively] at variable proportions – an inherent but often neglected consequence of inducible gene deletion. Thus, Cdc42^iΔEC retinas might contain both WT-EC (Rosa26-YFP negative) and Cdc42-KO-EC (Rosa-26-YFP positive) (see Materials and Methods for a technical note on this approach).

At P7, Cdc42^iΔEC retinas displayed reduced vascular outgrowth and a significant shortening of arteries and veins (Fig. 1D,E). In addition, conspicuous vascular malformations were observed in veins and their neighboring capillary network. The malformed vessels were focally dilated and commonly formed vascular outpouchings (Fig. 1D,F). These malformations progressively expanded with time and assumed mulberry-like shapes at later developmental stages (Fig. 1G). Interestingly, we observed vascular malformations only in retinas where the proportion of Cdc42-depleted cells was ≥60% (i.e. ≥60% Cdc42-KO-EC and ≤40% WT-EC). This strongly suggests that the vascular malformations do not simply arise by local clonal expansion of mutant cells in a tumor formation-like process. Rather, a threshold proportion of Cdc42 mutant cells is required to initiate lesion formation, which subsequently also engages the WT cells. Notably, we never observed vascular malformations in arteries. Although they occasionally appeared constricted, they were not changed significantly in overall width (Fig. 1D,F).

We next analyzed whether the capillary-venous vascular malformations were blood-perfused hollow structures that retained apical-basal (A-B) polarity. Because several studies have reported a role for Cdc42 in A-B polarization of epithelia (Martín-Belmonte et al., 2008; Wan et al., 2013; Elias et al., 2015; Kesavan et al., 2009) and ECs (Barry et al., 2015), and because A-B polarity is considered a prerequisite for lumen formation in epithelial tissues and blood vessels (Martín-Belmonte et al., 2007; Davis et al., 2011), we expected to find A-B polarity defects in Cdc42^iΔEC retinal vessels. Surprisingly, however, we failed to find convincing defects in A-B polarization and vascular lumen formation. The apical markers ICAM2 (Bradley et al., 1995) and podocalyxin/GP135 (Podxl) (Horvat et al., 1986; Meder et al., 2005) were both consistently localized at the luminal side of ECs, whereas the basement membrane protein collagen IV (Laurie et al., 1983) delineated the abluminal surface also at vessel stretches composed mostly of Cdc42-KO-EC, including the capillary-venous malformations (Fig. 1I, Fig. S1A, Movies 1-8). Blood vessels in Cdc42^iΔEC retinas were generally lumenized, with the exception of remodeling or regressing capillaries. The hollow cavernous outpouchings of the capillary-venous malformations were connected by continuous lumens and were delineated by a single layer of ECs that exhibited seemingly normal A-B polarization (Fig. 1H,I, Movies 1-11). These results suggest that during postnatal angiogenesis of the retina, Cdc42 might not strictly be required for A-B polarization of ECs, nor for vascular lumen formation. Our data instead suggest an important role of Cdc42 in restricting vessel diameter and in prohibiting the formation of capillary-venous malformations. We further investigated whether the loss of Cdc42 directly affected endothelial cell survival, as suggested by studies of Cdc42 KO embryos (Jin et al., 2013). Using the early apoptosis marker cleaved caspase 3, we found no evidence for increased EC apoptosis in Cdc42^iΔEC retinas (Fig. S1B,C). Our data are in agreement with observations suggesting that EC apoptosis in the embryo occurs secondary to disrupted circulation in Cdc42 KO embryos (Barry et al., 2015).

Because the delayed radial expansion of the Cdc42^iΔEC retinal vasculature (Fig. 1D,E) suggests problems with angiogenic sprouting, we studied this process in more detail. We took advantage of the individually variable degrees of Cdc42 deletion in Cdc42^iΔEC retinas to analyze the behavior of Cdc42-KO-EC in mosaic vessels with different proportions of WT-EC. We first characterized the mosaicism with regard to Cre-activated Rosa26-EYFP and found that recombination was highly variable (11-95%) between individual retinas. YFP⁺ WT-EC distributed randomly throughout the retinal vasculature (Fig. 2A,B), as shown by the equal proportions of recombined cells at tip (TR) and stalk (SR) positions (TR/SR≈1) (Fig. 2C-E). In contrast, Cdc42-KO-EC were significantly underrepresented at tip positions (TR/SR≈0.3) (Fig. 2A,B,D,E). In retinas with low proportions of Cdc42-KO-EC, no mutant cells occurred within several hundred micrometers from the vascular sprouting front (Fig. 2A,B). This distance decreased with increasing recombination rate, but even in highly recombined retinas where mutant cells were forced to assume tip cell positions, their disadvantage at this location was evident (Fig. 2B, bottom row). A reduced TR/SR ratio was also observed in Cdc42^iΔEC aortic ring explant cultures (Fig. S2A,B) and was obvious even at the high recombination rate (90±0.04%) that we achieved in those experiments (see Materials and Methods for details). These results demonstrate that Cdc42-KO-EC have a competitive disadvantage to assume tip positions in angiogenic sprouts, suggesting a crucial and cell-autonomous requirement for Cdc42 in tip cell selection.

Endothelial tip cells protrude actin-rich filopodia (Gerhardt et al., 2003) (Figs 2C and 3A), which have been proposed to sense the environment and direct vascular growth towards sources of VEGF proteins, anchor tip cells to extracellular substrates, pull the angiogenic sprout forward and facilitate the formation of anastomoses with other ECs (De Smet et al., 2009; Gerhardt et al., 2003). Cdc42-KO-EC at tip positions lacked filopodia and instead protruded a few short studded processes from their surfaces (Fig. 3A-D). Neighboring WT-EC tip cells, on the other hand, displayed normal filopodial extensions, demonstrating that Cdc42 is cell-autonomously required for filopodia formation (Fig. 3A). The morphologies of Cdc42-KO-EC at tip cell positions ranged between two extremes (Fig. 3E), a rounded lumenized cell (Fig. 3E, middle row, Movies 9 and 10) or a nonlumenized cell with abnormally elongated shafts (Fig. 3E, bottom row). Irrespective of their morphology, Cdc42-KO-EC at the tip cell position consistently lacked filopodia.

We next tested whether Cdc42-KO-EC could be forced to adopt tip cell characteristics and extend filopodia through inhibition of Notch signaling. The pharmacological gamma secretase inhibitor N-[N-(3,5-difluorophenacetyl)-L-alanyl]-S-phenylglycerin t-butyl ester (DAPT) normally increases tip cell and filopodia numbers in the developing retina (Hellström et al., 2007), but failed to do so in Cdc42^iΔEC retinas (Fig. S3A,B). Interestingly, however, DAPT treatment greatly enhanced the severity of the capillary-venous malformations in Cdc42^iΔEC retinas (Fig. S3C). This aspect is further discussed below.

Because in vitro studies have reported rapid and dramatic cellular changes upon Cdc42 disruption (Kroschewski et al., 1999; Nobes and Hall, 1995), the question arose as to what extent the phenotypes observed in Cdc42^iΔEC retinas at P7 could reflect vascular adaptations. We therefore studied the vascular phenotypes in Cdc42^iΔEC retinas 2 days earlier and found that all key abnormalities observed at P7 were already present at P5, including reduced vascular extension, formation of capillary-venous malformation, inability of Cdc42-KO-EC to form filopodia and assume tip cell positions (Fig. S4A-F). Similar to the situation at P7, we did not find evidence for an increase in EC apoptosis in Cdc42^iΔEC retinas at P5 (Fig. S4G,H).

Differential vascular endothelial cadherin [VE-cadherin (Cdh5)] dynamics at EC junctions has been proposed to drive cell re-arrangements during angiogenic sprouting (Bentley et al., 2014a). A role for Cdc42 in EC junction formation has been proposed in vitro (Broman et al., 2006) and during embryonic development (Barry et al., 2015). In agreement with these observations, we found that VE-cadherin localization was altered in Cdc42^iΔEC retinal vessels, irrespective of vessel type, but to a variable extent (Fig. 4A). Importantly, these defects were not strictly cell-autonomous, because WT-EC in chimeric vessels also showed abnormal junctional VE-cadherin staining (see Fig. 4A for examples). In contrast, junctional localization of another EC adhesion molecule, cluster of differentiation 31 (CD31/PECAM1), was unaffected in Cdc42^iΔEC retinas (Fig. 4B). Immunolabeling of the junction-associated proteins zonula occludens-1 (ZO-1) and β-catenin was technically problematic in whole-mounted retinas, and we therefore analyzed these proteins only in confluent primary cultures of Cdc42-KO-EC. In these conditions, VE-cadherin, β-catenin and ZO-1 were all localized to cellular junctions in Cdc42-KO-EC, but, in contrast to WT-EC, their junctions appeared thinner, while remaining continuous (Fig. 4C). The wide junction labeling observed in WT-EC under these culture conditions might reflect ongoing junctional remodeling (Abu Taha and Schnittler, 2014). Thus, Cdc42-KO-EC junctions appear less active or less able to remodel, albeit not necessarily unstable. We also analyzed mosaic cultures of Cdc42-KO-EC and WT-EC and found that individual WT-EC formed morphologically different contacts with neighboring WT-EC or Cdc42-KO-EC, respectively. Whereas the contacts between WT-ECs were normally wide, the contacts between WT-EC and adjacent Cdc42-KO-EC were thin and indistinguishable from the junctions formed between two adjacent Cdc42-KO-EC (Fig. 4D). This suggests that junctional defects in Cdc42-KO-EC may have dominant non-cell-autonomous effects on the behavior of neighboring WT-EC. The observed defects in VE-cadherin localization are, however, unlikely to have a significant impact on the formation of vascular malformations, because arteries did not dilate or develop malformed vascular structures, in spite of abnormal VE-cadherin localization (Fig. S5).

Because we failed to find clear evidence for A-B polarization defects in Cdc42^iΔEC retinas, we asked whether other types of polarization could be affected. ECs have also been reported to polarize perpendicular to their A-B axis, referred to as axial (Franco et al., 2015), front-rear (Dubrac et al., 2016) or planar cell polarity (PCP) (Devenport, 2014; Lee and Bautch, 2011). We first assessed EC axial polarization in conjunction with EC migration in scratch wound assays in vitro. The hallmark of this process is re-orientation of the microtubule organizing centers (MTOCs) and Golgi complex in the direction of cell migration (Kupfer et al., 1982) (Fig. 5A,B), which is considered a prerequisite for subsequent collective cell migration (Vitorino and Meyer, 2008). In the scratch wound assay, axial polarization first occurs in pioneer cells that emerge at the scratch border. Neighboring cells subsequently re-orient and follow the pioneers through drag forces (Kupfer et al., 1982). A crucial role for Cdc42 in this type of polarity was earlier suggested by Vitorino and colleagues using small interfering RNA-mediated knockdown of Cdc42 in human umbilical vein endothelial cells (Vitorino and Meyer, 2008). In order to stay as close as possible to the in vivo situation, we assessed axial polarity in scratch wound assays performed on primary brain ECs isolated from Cdc42^iΔEC mice. We determined EC polarization towards the wound, considering the nucleus-Golgi axis angle (Fig. 5A,B). Five hours after scratching, we found that the majority of WT cells in the first to third rows polarized within ±45° perpendicular to the wound border, with a slight lag in the third row. In contrast, the majority of Cdc42-KO-EC failed to polarize towards the wound (Fig. 5B,C,D, Table S1A). At 24 h postscratching, polarization of WT cells had further improved in the third row, whereas polarization of Cdc42-KO-EC did not improve with time in any row (Fig. 5B,C,D, Table S1A).

EC axial polarization is also observed in vivo (schematic representation in Fig. 6A,B) and has been reported to correlate with the direction of blood flow (Franco et al., 2015). To investigate whether the loss of Cdc42 influences EC axial polarity in vivo, we analyzed the Golgi orientation in arteries, veins and capillaries of WT and Cdc42^iΔEC retinas (Movie 12 shows how polarity was determined). This analysis confirmed EC polarization against the predicted blood flow direction in WT retinas and showed that EC axial polarity was specifically disturbed in veins and capillaries, but not in arteries of Cdc42^iΔEC retinas (Fig. 6C-G, Table S1B). We used the central-peripheral axis as a proxy for flow direction in arteries and veins (Fig. 6A). In the capillary plexus at the sprouting front, flow has been predicted to be dynamically changing direction (Franco et al., 2015). When analyzing the three most distal EC rows (i.e. tip cells and the two proximal stalk cells rows), we observed a dual-oriented polarity pattern in WT retinas, in which approximately half of the ECs oriented towards the retinal periphery and the other half towards the center. This pattern was lost in Cdc42^iΔEC retinas, where EC polarization instead appeared random (Fig. 6C,F,G, Table S1B).

To obtain further molecular insight into how Cdc42 regulated EC axial polarity, we analyzed the subcellular localization of members of the Par3/Par6 (Pard6a)/aPKC complex. In epithelial cells, Par complex proteins become enriched in areas in which polarity is subsequently established (Etienne-Manneville, 2004); inhibition of Cdc42 has been shown to prevent this enrichment in Caenorhabditis elegans (De Smet et al., 2009). When analyzing the subcellular localization of Par3 and aPKCζ in WT-EC in the context of the scratch wound assays, we observed that both Par3 and aPKCζ were localized to cellular junctions and membrane protrusions. In contrast, this localization was lost in Cdc42-KO-EC (Fig. S6), suggesting that Cdc42 controls EC axial polarity, at least in part, through localization of the Par complex.

Sprout extension, shuffling of tip cells, branch anastomosis and regression collectively shape the growing vascular network during angiogenesis. These processes require cytoskeletal rearrangement and EC migration (Franco et al., 2015; Geudens and Gerhardt, 2011), functions known to be regulated by Cdc42 in various cell types (Hall, 1998) including ECs (Sakabe et al., 2017; Vitorino and Meyer, 2008). Indeed, the delayed vascular outgrowth observed in Cdc42^iΔEC retinas (Fig. 1D,E) suggested impaired endothelial motility. Because real-time analysis of angiogenic sprouting in the postnatal mouse retina is technically challenging, we used three different in vitro assays to analyze EC motility and directed migration. To allow unambiguous identification of ECs in those assays, we bred Cldn5-GFP mice [which express green fluorescent protein (GFP) specifically in ECs] to Cdc42^iΔEC and control mice and subsequently isolated aortas and primary brain ECs. We further used the Rosa26-tdTomato(Ai14) reporter (Madisen et al., 2010) as a marker for recombination efficiency. Based on the Rosa26-tdTomato(Ai14) reporter, we obtained a high degree of recombination in both aortic rings (90±0.04%) and isolated ECs (80.8±12.02%). Time-lapse movies of aortic ring cultures showed that sprouting from Cdc42^iΔEC explants was significantly altered, including reduced sprout number and length, as well as an increased numbers of branch points (Fig. 7A,C, Movies 13 and 14). Cell tracking showed that overall sprout elongation was slower in Cdc42^iΔEC explants and that sprouts displayed slow movements with reduced directionality compared to controls (Fig. 7A,C, Movies 13 and 14). In order to understand whether Cdc42 was specifically required for the coordinated cell movements in angiogenic sprouts, or generally needed for EC motility, we tracked individual cell movements in 2D cultures of primary brain ECs isolated from Cdc42^iΔEC mice or controls. Whereas WT-EC in subconfluent cultures displayed a scattering behavior, Cdc42-KO-EC stayed attached in groups that displayed slow, wiggly and seemingly randomized movements (Fig. 7B,D, Movies 15 and 16), indicating a general requirement of Cdc42 for EC motility. EC migration speed was also analyzed in scratch wound assays, which revealed an impaired ability of Cdc42-KO-EC to migrate (Fig. 7E,F, Movies 17 and 18). Together, these analyses show that general cell motility, directed cell migration and the ability to uphold migratory directionality are all significantly compromised in Cdc42-KO-EC.

Cdc42 is a key regulator of the actin cytoskeleton (Hall, 1998), and it is generally assumed that this function accounts for its role in cell motility and migration (Jaffe and Hall, 2005). As expected, we found several signs of cytoskeletal abnormality in Cdc42-KO-EC. Stress fibers were scarce and discontinuous (Fig. 7G, green arrowhead) and actin accumulations at the leading edge of migrating cells, as well as actin structures in the few remaining cellular protrusions, were substantially decreased (Fig. 7G, yellow arrowhead). Interestingly, junctional actin cables were less affected; they appeared thinner but were continuous (Fig. 7G, red arrowhead). In addition, paxillin, an intracellular adaptor protein propagating signals from integrins and growth factor receptors in the regulation of cell motility (López-Colomé et al., 2017), was diffusely distributed in the cytoplasm of Cdc42-KO-EC, as opposed to its concentration at focal adhesions and along filamentous actin structures in WT-EC (Fig. 7H).

To investigate the degree to which individual Cdc42-regulated cellular processes contribute to angiogenic sprouting alone or in combination, we exploited predictive computational modelling (Fig. 8A). We utilized the memAgent-Spring model (MSM) in original and extended forms, previously shown to be highly predictive in vivo (Bentley et al., 2014b; Bentley and Chakravartula, 2017) (see Materials and Methods for details).

Calibrating the model to the reduction in filopodia (Cdc42_FIL) observed in vivo (Fig. S7A,B) was sufficient to reproduce the tip cell disadvantage and horizontal (rear-front) segregation of Cdc42-KO-EC in silico, with striking agreement (Fig. 8B,D, Movie 20). Surprisingly, the simulations also predicted unequal vertical (top-bottom) segregation of WT and Cdc42-KO-EC in the vessel, as the mutant cells also lose the competition for the VEGF protein source provided by the astrocyte network, which is much stronger below the vessel (Fig. 8G,H, Movie 20). We subsequently confirmed vertical cell segregation in vivo (Fig. 8K,L, Movie 21), a phenotype that would otherwise have escaped notice.

Both horizontal and vertical segregation occurred when the EC axial polarity-sensing ability was reduced on its own (Cdc42_POL) by 75% (Fig. 8E,I, Fig. S7C, Movie 22). When filopodia and EC axial polarity were disrupted simultaneously (Cdc42_FIL/POL) (Fig. 8F,J, Movie 23) a stronger segregation was observed than with Cdc42_FIL and Cdc42_POL alone, requiring only a 30% reduction in polarity (Fig. S7D) and suggesting that the processes might synergize.

As our experiments revealed a moderate reduction of junctional actin (Fig. 7G,H) and abnormal EC junctions in Cdc42-KO-EC (Fig. 4), we also modeled the effects of changes in the junctional actin cortex (Cdc42_COR) and cell adhesion (Cdc42_ADH). We found that as little as 25% reduction in Cdc42_COR sufficed to segregate WT-EC and Cdc42-KO-EC horizontally (Fig. S7E, Movie 24); however, Cdc42_COR defects failed to reproduce the vertical segregation, and co-modulation with filopodia (Cdc42_FIL/COR) did not decrease the threshold of Cdc42_COR required to match the in vivo data (Fig. S7F). Cdc42_ADH simulations predicted that a substantially increased adhesion (2- to 2.75-fold) would be required to match the in vivo horizontal segregation data (Fig. S7G, Movie 25). However, simultaneous modulation of adhesion and filopodia (Cdc42_FIL/ADH) failed to show an additive effect (Fig. S7H), overall indicating that adhesion changes might well contribute to the horizontal segregation of Cdc42 KO cells but cannot account for all aspects of the sprouting phenotype in Cdc42^iΔE retinas. We finally modeled all four entities in combination (Cdc42_{FIL/POL/ADH/COR}) (Fig. S7I, Movie 26) and were able to find parameters that could match the in vivo data; interestingly now, a reduction rather than increase in intercellular adhesion was required to match the in vivo data, and both the extension of cortical actin-dependent junctional protrusions and polarity sensing ability had to be only moderately reduced (10% and 30%, respectively).

Taken together, in silico modeling results therefore indicate that filopodia formation and EC axial polarity exert the major influences on angiogenic sprouting, with regulation of junctional actin and cell adhesion playing a lesser role.

Vascular malformations are considered a subgroup of vascular anomalies that differ from vascular tumors by their nonproliferative nature (Wassef et al., 2015). In order to clarify whether the vascular lesions in Cdc42^iΔE retinas would classify as vascular malformations, we analyzed EC proliferation in detail. We used 5-ethynyl-2′-deoxyuridine (EdU) incorporation to visualize active DNA synthesis in combination with a marker for endothelial cell nuclei (ERG) in order to identify proliferating ECs unambiguously, and subsequently quantified the EdU/ERG double-positive cells. Consistent with previously published results (Ehling et al., 2013), our analysis revealed that in control mouse retinas, EC proliferation occurs predominantly in veins and capillaries (Fig. 9A-C). Importantly, we were unable to detect any increase in EC proliferation in Cdc42^iΔEC retinas, irrespective of vessel type (artery, vein, capillary) or quantification method. Neither the amount of EdU/ERG double-positive cells, nor the amount of EdU-positive cells per vascular area, significantly changed in Cdc42^iΔEC retinas (Fig. 9A-C). Because proliferation in the mouse retina is reported to peak at P5 (Franco et al., 2015), which is 2 days prior to our analysis, we wanted to ensure that we had not missed an earlier, transient expansion of Cdc42-KO-EC. However, at P5, we also failed to observe any significant increase in EC proliferation (Fig. S8A,B). In addition, the total amount of ECs (quantified as ERG-positive nuclei) per retina was not significantly increased in Cdc42^iΔEC retinas at P5 or at P7 (Fig. 9D, Fig. S8C). In spite of this, we observed a threefold increase in EC density in Cdc42^iΔEC veins (Fig. 9F), in accordance with the dense concentration of ECs in the vascular lesions (Figs 9E and 10A, white boxes). The most plausible explanation for the crowding of ECs in the capillary-venous area is therefore that the proliferating ECs fail to migrate and re-distribute in the vascular tree. In fact, the inability of Cdc42-KO-EC to redistribute within the retinal vessel network was strikingly obvious in low degree chimeras, where Cdc42-KO-EC cells adopt a rounded shape and accumulate in veins and capillaries but fail to distribute into the arterial network (Fig. 10B). Quantifications of YFP⁺ cells in arteries and veins revealed that in Cdc42^iΔEC retinas the probability of a recombinant cell to be found in arteries is below 20% as compared to 50% in controls (see Materials and Methods for details) (Fig. 10C). We next quantified YFP⁺ cell distribution in the retinal vasculature by subdividing the vascular network of adjacent artery-vein pairs and their intervening capillary plexus in four equal quadrants, quantifying the relative amount of YFP⁺ recombinant cells in each quadrant and plotting the data in rose plots (Fig. 10D; see Materials and Methods for details). This analysis demonstrates that Cdc42-KO-EC accumulate in the proximity of veins and the adjacent vascular plexus and fail to distribute away from these sites (Fig. 10E,F). Based on these observations, we propose that the vascular malformations in Cdc42^iΔEC retinas arise as a consequence of defective cell migration, when ECs, proliferating at normal rates, are unable to re-distribute within the vascular network. Malformations hence remain restricted to capillaries and veins, vessel types in which proliferation is naturally high (but not increased by the loss of Cdc42) and are absent from arteries, where proliferation is naturally low. The defective cell migration leads, in the presence of naturally high proliferation rates, to dilations of veins and capillaries that ultimately progress into mulberry-like vascular malformations (Fig. 10G).

Cdc42 is known to regulate a multitude of cellular processes, but although some of its functions appear conserved in all cell types studied, others are cell-type specific. In this study, we aimed to clarify which cellular processes are regulated by Cdc42 in ECs in vivo and how these processes contribute to angiogenic sprouting and blood vessel morphogenesis. To avoid secondary systemic effects on vascular development, a notorious problem with embryonic gene deletion, we inactivated Cdc42 postnatally and studied its effects in the developing retinal vasculature. At this location, we found that Cdc42 regulates only a subset of its reported functions in ECs, namely filopodia formation (Barry et al., 2015; Fantin et al., 2015; Sakabe et al., 2017; Wakayama et al., 2015), adherence junction stability (Barry et al., 2015; Broman et al., 2006) and migration (Jin et al., 2013; Sakabe et al., 2017; Vitorino and Meyer, 2008; Wakayama et al., 2015). However, other known functions, including A-B polarization, lumen formation and proliferation, did not appear to be significantly changed by the loss of Cdc42. That vascular lumens can form in Cdc42^iΔEC retinas is surprising because Cdc42 has previously been implicated in vessel lumen formation both in vitro (Bayless and Davis, 2002; Davis et al., 2011) and during embryonic development (Barry et al., 2015). It is presently unknown whether the differences between embryonic and postnatal Cdc42 deletion reflect different requirements for Cdc42 in vascular lumen formation in embryonic and postnatal vessels. Because vascular lumens appear variably collapsed and dilated in Cdc42 mutant embryos (Barry et al., 2015), it is possible that these defects represent secondary consequences of cardiovascular dysfunction rather than primary defects in lumen formation. Differences in blood flow might also contribute to the observed differences. The mouse retinal vasculature develops by angiogenic sprouting from pre-existing lumenized vessels. The process of extension of the endothelial lumen during angiogenic sprouting involving inverse membrane blebbing dependent on blood pressure (Gebala et al., 2016). This differs from the proposed mechanism of endothelial lumen formation during vasculogenic vessel assembly in the aorta and yolk sac (Strilić et al., 2009, 2010), and in vitro, which takes place in the absence of blood flow. Consistent with our results, Barry and colleagues also observed that retinal blood vessels were lumenized following Cdc42 deletion and suggested that the timing of deletion relative to EC lumen formation might account for the presence of lumens and variability of phenotypes (Barry et al., 2015). Further studies are required to clarify the role of Cdc42 in vascular lumen formation in more detail.

Cdc42 is considered a key regulator of cell polarity (Etienne-Manneville, 2004). It is crucial for A-B polarization of epithelial cells in vitro (Bryant et al., 2010; Martín-Belmonte et al., 2007) and its deletion causes defects in lung (Wan et al., 2013), kidney (Elias et al., 2015) and pancreatic (Kesavan et al., 2009) tubular patterning owing to defective A-B polarization of the respective epithelia. Cdc42 has also been suggested to control A-B polarity in the endothelium (Barry et al., 2015; Lizama and Zovein, 2013), and, in addition, Cdc42 has been shown to control yet another aspect of cell polarity, PCP, both in the neuroepithelium of the inner ear (Kirjavainen et al., 2015) and during convergent extension in Xenopus laevis (Choi and Han, 2002). Surprisingly, our data indicate that in the postnatal retina, Cdc42 is dispensable for A-B polarization of blood vessels, but is essential for EC axial polarity in veins and capillaries. Our modeling data suggest that defective EC axial polarity affects the directional migration of tip cells at the sprouting front, as well as on stalk cells during sprout elongation and vascular regression. In agreement with a role of EC axial polarity in angiogenic sprouting, Dubrac and colleagues recently reported similar polarity defects upon EC-specific deletion of NCK1/2 (Dubrac et al., 2016). However, those polarity defects were only observed at the sprouting front and were not associated with vascular malformations. In Cdc42^iΔEC retinas, the loss of axial polarity correlates with the formation of vascular malformations, but whether these phenomena are causally linked remains to be established.

We propose that defective EC migration is causal in the development of vascular malformations in Cdc42^iΔEC retinas. Analysis of mosaic Cdc42^iΔEC retinas showed that Cdc42-KO-EC were unable to disperse through the retinal vasculature and instead accumulated in veins and their adjacent capillaries. During retinal development, the naturally high EC proliferation in veins and capillaries generates new ECs, which, under normal conditions, get incorporated into the expanding vascular network. Because EC proliferation was unaffected in Cdc42^iΔEC retinas, we conclude that the malformations build up when Cdc42-KO-EC are unable to migrate away from the location in which they are born. The inability of Cdc42-KO-EC to disperse within the vascular network explains their underrepresentation in arteries, and explains why arteries appear morphologically normal in Cdc42^iΔEC retinas. Whereas defective cell migration alone might be predicted to cause local crowding of ECs, the formation of large, hollow and mulberry-shaped malformations likely depends on the naturally high EC proliferation at these sites. In keeping with this scenario, the vascular malformations in Cdc42^iΔEC retinas were aggravated by pharmacological inhibition of Notch protein signaling, which is known to enhance EC proliferation. In conclusion, our results suggest that a balance between EC proliferation and migration is crucial for normal vascular morphogenesis and that a disturbance in this balance might provoke the formation of vascular malformations. Our data are also consistent with the idea that arteries form from vein-derived tip cells (Xu et al., 2014).

An interesting link between Cdc42 and the Hippo signaling pathway has recently been described in the context of tip cell migration and angiogenic sprouting (Sakabe et al., 2017; Kim et al., 2017). Whether Hippo signaling is of relevance for vascular malformations remains to be elucidated, however. It is further worth noting that the vascular malformations in Cdc42^iΔEC retinas resemble those observed in animal models of cerebral cavernous malformations (CCMs), both in terms of lesion morphology and location (Boulday et al., 2011). Further studies should therefore investigate putative molecular similarities between Cdc42^iΔEC-induced malformations and CCMs.

Animal housing, as well as the experiments performed, were in accordance with Swedish legislation and were approved by the local animal ethics committees prior to experimentation. The protocols included in this study were approved by the Stockholm's North Committee on the Ethics of Animal Experiments (permit number N33/10) and by the Uppsala Committee (permit numbers C224/12, C115/15, C111515/16). All efforts were made to minimize animal suffering. Mice were used regardless of their sex, because no difference in the phenotypes was detected.

Endothelium-specific tamoxifen-inducible Cdc42 KO mice (referred to as Cdc42^iΔEC) were obtained by crossing Cdc42^flox/flox mice (Wu et al., 2006) with endothelial-specific Cdh5(PAC)-Cre^ERT2 mice (Pitulescu et al., 2010). To evaluate Cre-mediated recombination efficiency, Cdc42^iΔEC mice were bred to Rosa26-EYFP (Srinivas et al., 2001) or, in a few cases, to Rosa26-tdTomato(Ai14) (Madisen et al., 2010) reporter mice. For all relevant experiments, Cdh5(PAC)-Cre^ERT2:R26R-EYFP and Cdh5(PAC)-Cre^ERT2:R26-tdTomato mice were used as controls. Postnatal gene deletion/reporter gene activation was induced by daily tamoxifen (T5648, Sigma-Aldrich) administration [100 µl of a 20 mg/ml solution in 90% corn oil (C8267, Sigma-Aldrich) and 10% ethanol], which was gavaged to lactating females from P0-P2 and transferred to the offspring through the milk. Mice were sacrificed for analysis at P7, unless otherwise noted. A technical note on the mosaic analysis: it should be highlighted that this approach relies on equal recombination efficiencies at the reporter locus and the Cdc42^flox locus and that, in principle, both constructs can recombine independently of one another. What we observe in practice is a high degree of correlation between reporter gene expression and a cell-autonomous phenotype (the loss of tip cell filopodia), which strongly argues for similar recombination rates. We analyzed a total of 239 YFP⁺ 238 YFP^– ECs at the tip cell position in ten highly recombinant Cdc42^iΔEC retinas and found that only 2% of YFP⁺ cells were able to extend filopodia, while 98% failed to do so. In contrast, 94% YFP^– ECs at the tip cell position extended filopodia, whereas in the remaining 6% of YFP^– ECs, filopodia formation appeared affected.

The transgenic mouse Cldn5(BAC)-eGFP (referred to as Cldn5:GFP), expressing eGFP under the control of the Cldn5 promoter, was generated using bacterial artificial chromosome (BAC) technology. Further characterization of this transgenic line will be presented in a dedicated independent publication. Cldn5:GFP mice were bred into the Cdc42^iΔEC or C57BL6J control WT backgrounds and used to unambiguously identify ECs in live imaging experiments.

For immunofluorescence in retinas and cells, the following antibodies were used (species, antibody, catalogue number, batch number, supplier, dilution): rat anti-VE-cadherin (14144181, E04063-1267, e-bioscience, 1:100); rat anti-ICAM2 (553326, 3308761, BD Biosciences, 1:200); goat anti-podocalyxin (AF15556, R&D Systems, 1:200); rabbit anti-collagen IV (2150-1470, 161115, AbD Serotec, 1:200); goat anti-CD31 (AF3628, R&D Systems, 1:500); rabbit anti-Golph4 (Golim4) (Ab28049, GR244016-7, Abcam, 1:400); rabbit anti- ERG (92513, GR219881-25, Abcam, 1:200); chicken anti-GFP (13970, GR236651-21, Abcam, 1:500); rabbit anti-cleaved caspase 3 (9661, 9661S, Cell Signaling Technology, 1:400); mouse anti-GM130 (Golga2) (610823, Transduction, 1:400); goat anti-VE-cadherin (sc-6458, A1912, Santa Cruz Biotechnology, 1:400); mouse anti-β-catenin (610154, 64266, BD Pharmingen, 1:400); mouse anti-paxillin (610051, 3053935, BD Biosciences, 1:400); rabbit anti-ZO-1 (40-2300, 977576A, Invitrogen, 1:400), rabbit anti-aPKCζ (sc-216, K0113, Santa Cruz Biotechnology, 1:400), rabbit anti-Par3 (07-330, DAM1772423, Millipore, 1:400). Phalloidin-488 (PK-PF488P-7-01, 357P039, PromoKine, 1:200) and phalloidin-555 (PK-PF555-7-01, 370P054, PromoKine 1:200) were also used, and, in some cases, nuclei were labeled with Hoechst 33342 (Invitrogen) diluted 1:10,000 in phosphate-buffered saline (PBS).

Whole-mount staining of retinas was performed as previously described (Gerhardt et al., 2003). In brief, eyes were fixed in 4% paraformaldehyde (PFA) in PBS at 4°C (1-2 h), retinas were dissected and permeabilized in PBS, 1% bovine serum albumin (BSA), and 0.5% Triton X-100 at 4°C overnight. For isolectin B4 (IB4) staining, retinas were permeabilized in PBlec (PBS, pH 6.8, 1% Triton-X100, 0.1 mM CaCl₂, 0.1 mM MgCl₂, 0.1 mM MnCl₂) and incubated with IB4, Alexa Fluor 647 conjugate (I32450, Life Technologies) in PBlec at 4°C overnight. Primary antibodies were diluted in PBS, 0.5% BSA and 0.25% Triton X-100 and incubated at 4°C from 4 h to overnight depending on the sample, followed by detection with secondary antibodies (Alexa Fluor 405, 488, 568, Cy3 or 647, Life Technologies). Retinas were mounted using ProLong Gold Antifade (Life Technologies).

Images were acquired using a Leica SP8 confocal microscope. Maximum intensity projections and 3D reconstructions of serial Z-sections were generated using LAS AF software (Leica Microsystems). Image processing and quantification was performed with Volocity 6.3 (Improvision) or Fiji (NIH). Figures were assembled using Adobe Photoshop and Illustrator CS6. Genotypes and treatments were blinded for all analyses.

Confocal images of whole-mount retinas, were taken using different objectives. For filopodia quantification, images were acquired with an HC PL APO 63×/1.30 GLYC objective and quantified in six fields/retina (two arterial front fields, two venous front fields and two front fields in between the artery and vein). Total number and filopodia length from WT and Cdc42 KO endothelial tip cells were quantified. For apoptosis analysis, cleaved caspase 3 staining from nine fields per animal was acquired using an HC PL APO 40×/1.10 W objective (three arterial fields, three venous fields and three fields between the artery and vein). The total number of double-positive cells (cleaved caspase 3⁺ and IB4⁺ or CD31⁺) per field was analyzed.

The overall recombination rate in P5 and P7 retinas, referred to as SR, was calculated as the percentage of YFP⁺ vascular area (YFP⁺/IB4 or CD31 area ×100) The maximum intensity projected confocal images of the whole retina wedges were acquired with an HC PL APO 10×/0.4 CS objective. TR was calculated as the percentage of recombined tip cells (number YFP⁺/total number of tip cells ×100) using sprouting front pictures taken with HC PL APO 40×/1.10 W and HC PL APO 63×/1.30 GLYC objectives. TR/SR ratios were quantified in the vascular front of retinas, with six fields/retina.

For aortic rings, maximum intensity projected confocal images of whole aortic explants from Cldn5:GFP⁺ controls and Cdc42^iΔEC were acquired with an HCX IRAPO L 25×/0.95 W 0.17 objective. SR and TR were quantified in six fields, considering tdTomato+area as the Cre recombination marker, and the Cldn5 GFP⁺ area as the vessel marker.

Tile scans of flat-mounted retinas from control and Cdc42^iΔEC mice were acquired using an HC PL APO 10×/0.4 CS objective on a Leica SP8 confocal microscope. Vascular extension, artery and vein length, as well as the widths of arteries, veins, and capillaries, were measured in at least three individual animals.

To determine the distribution of ECs within the retinal vasculature at P7, the location of YFP⁺ cells was analyzed in low degree chimeras (between 7% and 52% recombination rate). The probability of ECs in Cdc42^iΔEC localizing to an artery or a vein was analyzed and compared with control retinas [Cdh5(PAC)-Cre^ERT2; Rosa26-EYFP]. Retinas were stained with IB4 to visualize the entire vascular network and, in addition, with an antibody against GFP to visualize YFP⁺ ECs. Four incisions were made in each retina in order to obtain four wedges each containing an artery-vein pair.

χ² test was used to compare the distribution of YFP⁺ signal among the quadrants of the two groups (two artery-vein pairs and their adjacent capillary vasculature were analyzed per animal in three controls and three Cdc42^iΔEC animals).

Single injection of 0.1 mg/g gamma secretase inhibitor DAPT (565770, Calbiochem) dissolved in 10% ethanol and 90% corn oil (injection volume 10μl/g body weight) was administered subcutaneously to P7 controls and Cdc42^iΔEC mice for 12 h. Untreated Cdc42^iΔEC and control mice injected with vehicle were used as experimental controls. Confocal images of three fields/retina were acquired using an HC PL APO 40×/1.10 W objective (one arterial front field, one venous front field and one front field between the artery and vein). Total number of filopodia per 100 µm of vascular front length was normalized to vehicle-treated control retinas.

Proliferation of ECs was examined using a Click-iT EdU Alexa Fluor 555 Imaging Kit (C10638, Invitrogen). EdU was prepared (5 mg/ml in 10% dimethyl sulfoxide) and 10 µl/g body weight was injected intraperitoneally into pups 4 h before sacrifice. Retinas were stained for ERG, GFP, and IB4 or CD31, and images were acquired with an HCX IRAPO L 25×/0.95 W 0.17 25× objective. Quantification of EC proliferation, displayed as percentage of proliferating (EdU⁺) cells/per EC nuclei (ERG⁺) and proliferating (EdU⁺) cells per vascular area (scored as IB4⁺ or CD31⁺ area), was measured in whole retinas or in four pictures of sprouting front, four veins, and four arteries per retina.

Tile scan confocal images of whole-mount retinas were taken using an HCX IRAPO L 25×/0.95 W 0.17 objective. Retinas were stained for ERG, GFP and IB4 or CD31. Total number of ECs per retina (ERG⁺ nuclei) was quantified in P5 (control n=3 and Cdc42^iΔEC n=4) P7 retinas. Moreover, in P7 retinas, total number of ECs (ERG⁺ nuclei) per vein length (mm) and EC density (ERG⁺ nuclei per frame) was quantified in nine pictures (300 µm×300 µm) from malformed areas versus nonmalformed areas in Cdc42^iΔEC retinas.

Briefly, thoracic aortas were dissected from P7 Cldn5:GFP WT and Cldn5:GFP Cdc42^iΔEC mice immediately after sacrifice. The fibro adipose tissue was removed and the aortas were serially cross-sectioned into 1- to 2-mm-thick rings. The aortic rings were embedded between two layers of collagen type I (1.5 mg/ml) (354236, BD Biosciences). The mouse aortic explants were then cultivated at 37°C, 5% CO₂ with Dulbecco's modified Eagle medium (DMEM) without Phenol Red (31053-028, Life Technologies) and supplemented with 15% fetal calf serum (Sigma-Aldrich), 30 ng/ml VEGF-A165 (450-32, PeproTech), 25 mM HEPES, 1.2 mM sodium pyruvate, penicillin (100 µg/µl) and streptomycin (100 µg/µl) using 24 glass bottom wells (P24G-1.5-13-F, MatTek). After 3-4 days of culture, when the aortic rings were sprouting, they were imaged over 2-3 days. After live imaging experiments, the aortic rings were fixed in 4% PFA. Confocal tile images of whole aortic explants were taken using an HC PL APO 10×/0.4 CS or HCX IRAPO L 25×/0.95 W 0.17 objective. Sprout number, length and branches per sprout in aortic explants from WT controls and Cdc42^iΔEC were quantified.

EC isolation was based on the method described in Niaudet et al. (2015). P7 brains from control and Cdc42^iΔEC mice were minced and digested with 1% collagenase (C6885, Sigma-Aldrich) in HEPES-buffered DMEM (Life Technologies) for 10 min under stirring at 400 rpm and 37°C. The cell suspension was filtered through a 70-mm nylon strainer (BD Falcon) and collected by centrifugation at 300 g for 10 min. Cell pellets were resuspended in DMEM supplemented with 10% FBS and incubated for 45 min RT with Dynabeads (Invitrogen) previously coated with rat-anti mouse PECAM1 antibody (553370, BD Pharmingen). ECs were separated using a magnet bar and dissociation from the beads was carried out with Tryp-LE 10× (Life Technologies). Brain ECs were cultured at 37°C, 5% CO₂, in endothelial cell growth medium 2 (C22011, Promo Cell). For immunofluorescence, brain ECs in culture were fixed for 5 min in 3% PFA and permeabilized with 0.2% BSA, 0.1% Triton-X100 in PBS for 1 h. Primary antibodies were diluted in 0.1% BSA, 0.05% Triton X-100 in PBS and incubated for 2 h at room temperature, followed by 1 h incubation with secondary antibodies (Alexa Fluor 405, 488, 568, Cy3 or 647, Life Technologies). The effects of Cdc42 deletion on the actin cytoskeleton, EC junctions and mislocalization of paxillin, Par3 and aPKC, was consistently observed in at least three independent primary brain EC cultures and analyzed in at least 50 cells, in which ∼100% of cells were consistent with the reported phenotype.

Live imaging experiments were performed with primary brain ECs and aortic rings isolated from P7 Cldn5:GFP control and Cldn5:GFP, Cdc42^iΔEC mice (tdTomato was used to report recombination in these experiments). Confocal microscope with motorized stage and incubation equipment (brick gas mixer, 5% CO₂ and temperature control at 37°C) was used to image the samples over 24-97 h using an HC PL APO 10x/0.4 CS or HCX IRAPO L 25×/0.95 W 0.17 objective. Tracking analysis was performed using the MTrackJ plugin from Fiji and expressed as migration speed (µm/h) for ECs and sprout elongation speed (µm/h) for aortic rings. Aortic ring and EC movies were created from the captured images using Fiji software.

Primary brain ECs isolated from P7 Cldn5:GFP controls and Cldn5:GFP, Cdc42^iΔEC animals (tdTomato was used to report recombination in these experiments) were seeded in ImageLock 96-well plates (Essen Bioscience). Confluent cell monolayers were scratched using a 96-pin mechanical device (WoundMaker, Essen Bioscience), producing a 700-800 µm wound. Cell migration was recorded for 72 h using a 10× objective with IncuCyte ZOOM (Essen Bioscience). The integrated CellPlayer Cell Migration Software Module (Essen Bioscience) enabled automated analysis of IncuCyte cell migration, and relative wound density analysis was performed in multiple scratches from control and Cdc42^iΔEC cultures. Two-way analysis of variance (ANOVA) repeated measures test was applied for statistical analysis. The overall recombination rate (80.8±12.02%) in scratch assay was calculated in three animals from three independent experiments as the percentage of Ai14 (tdTomato)-positive vascular area (Ai14⁺/Cldn5⁺ area ×100).

Following the scratch assay, brain ECs were stained for EC nuclei ERG, VE-cadherin, and the Golgi marker GM130. EC polarization towards the wound was determined by the nucleus-Golgi axis angle as previously described (Etienne-Manneville and Hall, 2001). A cell was considered correctly polarized when the Golgi apparatus was oriented within ±45° perpendicular to the wound border. To evaluate Golgi orientation in blood vessel in vivo, P7 retinas were stained with CD31, Golph4, and ERG or Hoechst. Confocal images from arteries, veins and capillaries at the sprouting front region were taken, and the orientation of EC nucleus-Golgi pairs along blood vessels was analyzed by optical sectioning (see Movie 12 for how polarity was determined and Table S1 for quantification details). The percentage of ECs with the Golgi polarized within ±45° against the predicted flow direction was determined in arteries and veins. For the sprouting front, the percentage of ECs in the first (tip), second and third (stalk) rows with the Golgi polarized ±45° towards the center and retinal periphery was determined. Rose plots showing the angular distribution of individual nucleus-Golgi orientation were generated in Oriana4 software (Kovach Computing).

Freshly isolated brain ECs from P7 control and Cdc42^iΔEC animals were snap-frozen, lysed and used for western blotting and qRT-PCR. Proteins from frozen cells were extracted in phospho-lysis buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.5% deoxycholate, 0.5% SDS, 0.1% NP-40, 0.1% Triton X-100, supplemented with PhosSTOP phosphatase inhibitor and Complete protease inhibitor cocktail (both Roche Diagnostics).

Ten micrograms of total proteins were separated by sodium dodecyl sulfate polyacrylamide gel electrophoresis on 4-12% gradient gels (Bio-Rad) and transferred to polyvinylidene fluoride membranes (Immobilon-P, Millipore). Membranes were incubated with primary antibodies rabbit anti-Cdc42 (2466, Cell Signaling Technology) and anti-calnexin (ab22595, Abcam), followed by horseradish peroxidase-linked secondary antibodies, and signals were detected by a chemiluminescence system (GE Healthcare).

Total RNA was isolated using an RNeasy micro kit (Qiagen). Reverse transcription was performed using iScript Reverse Transcription Supermix (Bio-Rad). cDNA was amplified in a CFX96 Real-Time system (Bio-Rad) using Taqman probe Cdc42 (Mm01194005). Expression levels were normalized to the housekeeping gene encoding for Gapdh. Fold change differences were calculated based on the relative quantification C_T method, also referred as 2^−ΔΔCT.

The memAgent-Spring model (MSM) of tip/stalk dynamics in angiogenesis was used (see Bentley et al., 2008, 2009 for full details of methods). In the MSM, the EC membrane is represented by a collection of computational agents (‘memAgents’), permitting localized responses of the cell to its environment. These memAgents are connected by springs following Hooke's law, capturing the tension in the actin cortex beneath. The model is well calibrated to match in vivo and in vitro VEGF/Dll4/Notch protein signaling and cell shape changes during filopodia-led cell migration and has been experimentally verified (Bentley and Chakravartula, 2017; Kur et al., 2016; Ubezio et al., 2016).

For simulations involving cell rearrangement, the in vivo-validated MSM-Cellular Potts model (CPM) model was utilized (see Bentley et al., 2014a for full methods), which was previously calibrated to in vitro sprouting data. This model integrates a Cellular Potts module to simulate local junctional dynamics as cells rearrange in the adhered tube, encompassing differential adhesion-based mechanisms.

We evaluated each Cdc42 cellular process alone (filopodia, adhesion, actin cortex and polarity) or in combination (Fig. 8A) in simulated vessels consisting of randomly positioned Cdc42-KO-ECs and WT-ECs in a (6:4) ratio matching our in vivo ratio, and compared those with simulations of WT:WT. ECs (Fig. 8C, Movie 19). To model each simulated cellular effector of Cdc42 alone (filopodia, polarity, junctional actin cortex and intercellular adhesion), an existing parameter of the MSM-CPM was changed with a constant k as described below (Fig. 8A). We conducted a sensitivity analysis over a range of values for k (for all processes except filopodia, where the reduction had been quantified in vivo, in which case k was calibrated to match this reduction) to find how much of a reduction, if any, is required in that particular process to match the in vivo contribution of KO cells to the tip and what other phenotypic effects would be predicted.

(2) Cdc42_POL: modulating polarized formation of junctional cortex protrusions.

To simulate a polarity loss mutant, a new threshold k₂ was added to the polarity-sensing module of the MSM-CPM. By stochastic generation of a random number ρ, we allowed any copyflip attempt, by which an in silico EC protrudes its junctional cortex into a neighboring EC's space, to take place if ρ≥k₂. However, if ρ<k₂, copyflips were only allowed to occur when they were polarized towards the tip of the sprout (as described in Bentley et al., 2014a). For example, a k2-value of 0.50 increases the chance of accepting a nonpolarized copyflip with 50% (whereas these are not accepted at all in the original MSM-CPM).

P7 Cdc42^iΔEC retinas were stained with CD31 and GFP, to visualized recombined cells. 3D images of confocal stacks from veins were acquired, and quantification of the percentage of Cdc42-KO-EC (YFP⁺) cells situated in the upper (towards the vitreous) and bottom side (towards astrocytes) of Cdc42^iΔEC veins was analyzed (n=2 animals, 370 cells analyzed).

Statistical significance was determined by performing a number of analyses; unless otherwise stated, unpaired two-tailed Student's t-test was used. One-way ANOVA, two-way ANOVA and χ² test were performed where appropriate (see each experimental method for details). GraphPad Prism version 7.02 for Windows (GraphPad Software) was used for the analyses. Differences were considered significant with a P-value <0.05. The variance was similar between the groups that were statistically compared and their distribution was normal. Quantified data are presented as bar graphs of mean±s.d., shown as error bars or as rose diagrams of the mean.

We thank Pia Peterson, Jana Chmielniakova, Cecilia Olsson, Helen Leksell for technical assistance; Thommie Karlsson for help with microscopy; and Jennifer Hofmann and Lena Claesson-Welsh for critical reading of the manuscript.