The GEF Trio controls endothelial cell size and arterial remodeling downstream of Vegf signaling in both zebrafish and cell models

Arterial networks enlarge in response to increase in tissue metabolism to facilitate flow and nutrient delivery. Typically, the transition of a growing artery with a small diameter into a large caliber artery with a sizeable diameter occurs upon the blood flow driven change in number and shape of endothelial cells lining the arterial lumen. Here, using zebrafish embryos and endothelial cell models, we describe an alternative, flow independent model, involving enlargement of arterial endothelial cells, which results in the formation of large diameter arteries. Endothelial enlargement requires the GEF1 domain of the guanine nucleotide exchange factor Trio and activation of Rho-GTPases Rac1 and RhoG in the cell periphery, inducing F-actin cytoskeleton remodeling, myosin based tension at junction regions and focal adhesions. Activation of Trio in developing arteries in vivo involves precise titration of the Vegf signaling strength in the arterial wall, which is controlled by the soluble Vegf receptor Flt1.

T he arterial vascular network distributes blood flow through the body, a process crucial for sustaining organ function and homeostasis. During embryonic development arterial networks enlarge and expand to meet the increasing metabolic demand of the growing and differentiating tissue. Also in ischemic cardiovascular diseases, revascularization, and regeneration of hypoperfused organs involve adaptations of the arterial system 1,2 . Selective targeting of the arterial growth process, and in particular creating arteries with a structurally large diameter to enhance flow conductance and relief hypoxia, is considered therapeutically relevant for treating patients with ischemic cardiovascular diseases 1,2 . The molecular and cellular mechanisms controlling arterial diameter are only partly understood but involve interactions between blood flow and endothelial cells, and BMP-Smad signaling controlling distinct endothelial cell behaviors during the vascular remodeling process [3][4][5] .
An increase in flow promotes endothelial cell proliferation and migration collectively facilitating the transition of a small caliber vessel segment, with only few endothelial cells, into a larger caliber arterial segment, with many endothelial cells [6][7][8][9] . Flow activates the Akt-PI3-kinase signaling pathway responsible for endothelial proliferation. In addition, flow activates the BMP9-Alk1-Eng-Smad4 signaling pathway, which restricts flow-induced Akt activation and promotes endothelial quiescence 5,6,[10][11][12][13][14] . The extent of diameter remodeling of arteries is determined by the amount of flowing blood and the shear stress setpoint 15 . Blood flow creates frictional forces exerted on endothelial cells, termed shear stress. Shear stress level has to be kept within strict boundaries to maintain endothelial quiescence and vessel stability 3 . A sustained increase in blood flow and shear stress promotes outward remodeling of the arterial lumen, resulting in a return of shear stress levels to the original setpoint. Vegf receptor-3/Flt4 is a component of the endothelial junctional mechanosensory complex, and acts as a determinant of the shear stress setpoint 16,17 . Lowering Flt4 expression and increasing the shear stress setpoint causes arteries to narrow thereby allowing shear stress to return to the setpoint 16 . In addition, increases in flow, after causing an initial diameter increase, have been shown to induce vessel contraction mediated via endothelial cell shape changes 10 . The transforming growth factor beta co-receptor Endoglin regulates vessel contraction, and loss of endoglin results in enlarged endothelial cells and blood vessel diameter in response to flow increases 10 . In adult pre-existing collateral arteries, flow-dependent arterial outward remodeling also involves activation of NFκB and inflammatory pathways in endothelial cells, leading to the recruitment of monocytes/macrophages that assist with remodeling by secretion of matrix metalloproteinases, and growth factors 18,19 . In gridlock mutant zebrafish embryos, macrophages contribute to flow driven outward remodeling of aortic collaterals, suggesting a developmentally conserved role for macrophages in arterial remodeling 20 .
Cell shape is determined by the activity of small Rho GTPases. Their activity contributes to maintaining an equilibrium between the forces providing centripetal tension and forces ensuring cell spreading and avoidance of cell collapse [21][22][23] . Here we show that in endothelial cells, the guanine nucleotide exchange factor Trio, and activation of the small GTPases Rac1 and RhoG, trigger Factin remodeling events in the endothelial cell periphery, increasing endothelial cell size. Arterial-specific expression of Trio in vivo augments endothelial cell size, resulting in functional arteries with a structurally larger lumen diameter, without change in endothelial cell numbers. Activation of Trio in vivo requires subtle fine-tuning of local arterial Vegf-Kdrl signaling levels, which is achieved by arterial Flt1 acting as a Vegf trap. Genetic targeting of the local arterial Flt1-Vegf balance results in endothelial cell enlargement, and significant outward arterial diameter remodeling, even during low flow conditions. Increases in vessel diameter reduce the resistance to flow. Trio-induced endothelial shape changes, and diameter remodeling in response to Vegf, may therefore aid to fine-tune local flow distribution in response to changes in tissue metabolism or hypoxia.

Results
Arterial Flt1 determines arterial diameter. Vegf is an attractive candidate for targeting arterial caliber as it controls key aspects of arterial development 1,24,25 , and is capable of activating small Rho-GTPases. Yet, beyond a narrow therapeutic window Vegf may induce adverse side-effects such as vessel overgrowth, e.g., hemangioma formation, and increased vessel permeability. How to deliver Vegf without deleterious side-effects is still an outstanding issue 26 . To fine-tune the spatio-temporal delivery of Vegf needed to obtain large arteries, we examined formation of arterial networks in the trunk of developing zebrafish embryos using different vegfaa gain of function scenarios.
We first employed transgenic embryos with constitutive or inducible expression of vegfaa under control of the somite muscle-specific 503unc promoter 27 (here termed vegfaa musc ; Fig. 1a, b; Supplementary Fig. 1a-d). Constitutive vegfaa overexpression increased endothelial cell number and disrupted the arterial vasculature, abrupting blood flow perfusion when compared to wild-type (WT) (Fig. 1a, b). Similar results were obtained upon inducible vegfaa overexpression ( Supplementary  Fig. 1b, d, e-w). Because such vegfaa transgenic approaches resulted in Vegfaa overdose inappropriate for targeting the diameter of arteries, we next aimed at obtaining more subtle changes by manipulating Vegf protein bioavailability, which is determined by the Vegf decoy receptor Flt1.
Flt1 has a very high affinity for Vegf and mainly acts as a Vegf trapping receptor, limiting Vegf bioavailability and signaling through Kdrl 28,29 . Conversely, ablating flt1 or displacing Flt1trapped Vegf produces a Vegf gain-of-function scenario, with endogenous production of Vegf. We hypothesized that Flt1 can be used as a vehicle to deliver Vegf directly into growing arteries. For this approach to work efficiently, Flt1 protein must be expressed in close proximity to the Vegf signaling receptor Kdrl on arterial endothelium. We found that this is indeed the case ( Fig. 1c-g). The developing arterial intersegmental vessels (aISV) displayed flt1 promoter activity ( Supplementary Fig. 2a-c) and expressed both flt1 isoforms; the membrane bound flt1 (mflt1) and soluble flt1 (sflt1) (Supplementary Fig. 2d, e). To determine Flt1 protein distribution, we generated a knockin line, harboring two HA tags in exon 3 of the endogenous flt1 locus, which labeled both mFlt1 and sFlt1 ( Supplementary Fig. 2f). Immune staining of the transgene demonstrated Flt1 protein expression in the developing aISVs (Fig. 1c, d). To substantiate the distribution of the secreted, soluble Flt1 isoform around developing arteries, we furthermore examined protein distribution in a transgenic line expressing an engineered HA-tagged sflt1 transgene with an inactive Vegfaa binding domain (Tg(flt1 enh :sflt1_Δ7-HAHA) ka612 , Supplementary Fig. 2g-n). Immune staining showed sFlt1 protein co-localizing with kdrl in arterial endothelium ( Fig. 1e-g;  Supplementary Fig. 2l-n). Transgenics with tagged wild-type sflt1-HAHA could not be used, because gain of wt-sflt1-HAHA decreased local Vegf 29 and inhibited aISV formation, thus rendering this experiment uninterpretable ( Supplementary  Fig. 2i-k).
Arterial ISVs, which express the Flt1 protein, are juxtaposed to Vegf-producing cells in developing somites 24 . Ablation of flt1, thereby creating a Vegfaa gain of function scenario 29 , resulted in a significant increase of aISV diameter (Fig. 1h-l). A similar result was obtained with a flt1 targeting morpholino 29 (Fig. 1m, n).
While both loss of flt1 and plgf gain of function induce a Vegf gain of function scenario, one component that may contribute to the difference in arterial diameter between plgf musc embryos and flt1 −/− mutants involves the spatial distribution of arterial Flt1 and Kdrl receptors. In plgf musc , Flt1, and Kdrl are both expressed in the same arterial ECs (Fig. 1g, Supplement Fig. 2u accounts for the observed difference, the model predicts that reducing flt1 expression in plgf musc , should annihilate the difference between plgf musc and flt1 −/− mutants. Nevertheless, arteries in plgf musc transgenics injected with a flt1 targeting morpholino should still show some degree of arterial diameter growth as the knockdown of flt1 itself induces a vegfaa gain-offunction scenario, similar as in flt1 −/− mutants. We tested these assumptions, and in-line with our hypothesis we indeed found that knockdown of flt1 in plgf musc embryos reduced arterial diameter growth, to the levels observed in flt1 −/− mutants ( Fig. 2e-h).
Our data support the model in which excessive Plgf displaces trapped Vegfaa from arterial Flt1, thereby allowing the released free Vegfaa to bind arterial Kdrl receptors to induce signaling. Inline with Vegfaa-Kdrl signaling in plgf musc embryos, we found that inhibiting Kdrl signaling or reducing vegfaa using a low dose vegfaa ATG targeting morpholino reduced lumen diameter growth in plgf musc (Fig. 3a-f). Arterial ISVs express kdrl and the Vegf receptor-3, flt4, which can regulate Vegf receptor-2 signaling output 30 . Genetically ablating flt4 or morpholino-mediated knockdown of flt4 did not affect arterial diameter growth in plgf musc or flt1 loss of function embryos (Fig. 3g-k; Supplementary Fig. 3a-g). As Plgf may bind to neuropilin-1 receptors we examined nrp1 mutants. Plgf increased aISV diameter in nrp1a −/− mutants ( Supplementary Fig. 3h). Taken together these data indicate that Vegf-Kdrl is the main signaling pathway accounting for diameter growth in plgf musc . Developing aISVs in plgf musc were enlarged prior to formation of a functional perfused trunk vascular network (Fig. 3l-n). To rule out increased flow as a trigger for Plgf-induced arterial remodeling, we next exposed aISVs of plgf musc embryos to low flow conditions. To reduce heart rate and trunk perfusion, we treated WT and plgf musc embryos with the L-type calcium channel blocker nifedipine 10,16 (Fig. 3o-r). WT embryos exposed to nifedipine showed reduced aISV and aorta diameter (Fig. 3o-r; Supplementary Fig. 4a-d). However, plgf musc embryos exposed to nifedipine still showed a significant increase in aISV diameter when compared to WT with normal flow, or WT with nifedipine ( Fig. 3r) Fig. 4h-l). Tnnt2 morphants showed a reduced aISV size and aortic diameter (Fig. 3s-v; Supplementary  Fig. 4h-k). However, despite loss of flow, aISV size was still larger in plgf musc embryos injected with tnnt2 targeting morpholino, when compared to WT with flow, or WT injected with tnnt2 morpholino (Fig. 3s-v); while tnnt2 morphants showed deficits in lumen formation (Fig. 3t), plgf musc -tnnt2 morphants showed a clear lumen on confocal sections of 3D stacks (Fig. 3u).
As mouse studies attribute the proarteriogenic effect of Plgf to activation of mFlt1 signaling in macrophages, we decided to investigate Plgf in a macrophage loss-of-function model 31 . Reducing macrophage differentiation (using a triple morpholino knockdown of pu.1, gcsfr, and irf8) had no effect on aISV diameters in plgf musc gain of function embryos ( Supplementary  Fig. 5a-l). Taken together, our data suggest that Plgf-induced aISV remodeling is neither triggered by increased shear stress nor macrophages.
developing aISV can be achieved in three ways: by having more endothelial cells, larger endothelial cells or a combination of both. We detected the combination of both: more and enlarged endothelial cells in aISVs upon plgf gain-of-function when compared to WT ( . We reasoned that RhoGTPase activity and changes in F-actin remodeling could contribute to the in vivo phenotypic changes in endothelial cell shape. To block F-actin polymerization, we administered the actin polymerization inhibitor Latrunculin B (LatB) 32 in vivo. LatB significantly reduced aISV diameter growth in plgf musc embryos when compared to vehicle treated plgf musc embryos ( Fig. 4f; Supplementary Fig. 6h-p). Major aspects of intracellular F-actin dynamics are regulated through the small Rho GTPases family, including Rac1 and RhoG 33 . The Rac1 inhibitor CAS 1177865-17-6 significantly inhibited aISV diameter expansion in plgf musc embryos ( Fig. 4f; Supplementary Fig. 6h-    Endothelial enlargement requires the GEF1 domain of Trio. Rho GTPases control actin reorganization, and GEFs like Trio activate Rho GTPases by promoting their exchange of GDP for GTP. Trio is a unique Rho GEF, because it has two separate GEF domains, GEF1 and GEF2, that control the GTPases RhoG/Rac1 and RhoA, respectively 36 . To substantiate which GEF domain of Trio is functionally relevant for cell enlargement in our setting, we performed gain-of-function experiments with Trio deletion constructs 37 . In vitro overexpression of a truncated Trio form containing the spectrin domain and GEF1 domain (termed TrioN), but lacking the GEF2 domain, resulted in significantly larger endothelial cells compared to control-transfected cells ( Fig. 5a-d, quantification in Fig. 5g). In further support of functional requirement of the GEF1 domain we found that overexpression of the GEF1 domain only also induced endothelial cell enlargement (Fig. 5e,f quantification in Fig. 5g). Besides full length Trio, 4 additional Trio-GEF1 containing isoforms (Trio-A, B, C, D) arising from alternative splicing have been described 38 .
In HUVEC, all GEF1 containing isoforms were detectable but full length Trio was the most abundant Trio form ( Supplementary  Fig. 8a). The Trio-B isoform is truncated at the C-terminus and similar to TrioN only contains the GEF1 domain. In support of the GEF1 domain, endothelial overexpression of Trio-B increased endothelial size (Supplementary Fig. 8b). The GEF1 domain of Trio can activate both Rac1 and RhoG. Endothelial cells transfected with constitutively active Rac1, or a photo-activatable Rac1 probe 39 (Rac1-PA-WT) that was activated by light showed a significantly larger size when compared to mCherry-expressing control cells (Fig. 5h-l), whereas a catalyticdead mutant of the photo-activatable Rac1 probe (Rac1-PA-C450A) failed to induce a cell size change upon exposure to light ( Fig. 5h-l). We found no evidence for a functional role of the Rac1B splice-isoform (Supplementary Table 1; Supplementary  Fig. 8c-g). Endothelial overexpression of a dominant active RhoG form (GFP-RhoG-Q61L) significantly increased endothelial cell area (Fig. 5m). Conversely, silencing of Rac1 or RhoG in endothelial cells overexpressing TrioN significantly reduced the TrioN-induced cell size increase (Fig. 5n- Fig. 8i-l). Activation of Rac1 involved Trio as shRNA-mediated silencing of Trio using two different shRNAs reduced VEGF-induced Rac1 activation (Supplementary Fig. 8i-l). Besides Trio, there are a number of other known regulators for Rac1, including Tiam1, and multiple regulators may coordinate in regulating the function of Rac1 in this context. In endothelial cells overexpressing TrioN, we observed that Tiam1 was expressed in junctional regions, as opposed to the more cytosolic localization observed in controltransfected cells (Fig. 5q-v). Endothelial overexpression of Tiam1 resulted in larger endothelial cells, similar to the effects observed upon overexpression of TrioN (Fig. 5w). Combining TrioN and Tiam1 overexpression had no additive effect on increasing EC size when compared to TrioN alone (Fig. 5w). These data suggest that one part of cell enlargement may require careful spatial positioning of GEFs, and activation of RhoGTPases in junctional regions.
Trio activates Rac1 in junctional regions. Using a Rac1 FRETbased biosensor we found that Trio activated Rac1 in particular at endothelial cell junctional regions (Fig. 6a, b). TrioN and Trio-GEF1 were equally potent in activating Rac1 at junction regions (  (Fig. 6e, f). After 17 h post-TrioN transfection, no additional enlargement was observed and the VE-Cadherin expression pattern was reminiscent of a honeycomb structure, suggestive of a force-equilibrium between the interacting cells 41 .
Also, the majority of active myosin, distinguished by staining for mono-phosphorylated myosin light chain S19, localized strongly at the junction region upon TrioN overexpression (Fig. 6i, j). In addition, we observed that fully di-phosphorylated myosin light chain (T18/S19) localizes even more specifically to junctional Factin bundles (Fig. 6k, l). Inhibition of myosin activity from 1 h after TrioN transfection significantly abrogated Trio-induced endothelial cell size enlargement (Supplementary Fig. 9k-m). Inhibiting myosin activity in already enlarged endothelial cells resulted in the formation of gaps in particular at tri-cellular junctions ( Supplementary Fig. 9n, o).
We next reasoned that Trio-induced changes in myosin-II and cytoskeletal remodeling affect the tension distribution along the F-actin cytoskeleton. To test this, we examined local tension at To test if VE-Cadherin-based cell-cell junctions are functionally relevant for the TrioN-induced enlargement of endothelial cells, we overexpressed TrioN while simultaneously silencing endothelial VE-Cadherin ( Fig. 7d-m). Loss of VE-Cadherin had no significant impact on TrioN-induced endothelial cell enlargement (Fig. 7m). To test if coordination between adjacent endothelial cells is critical for endothelial enlargement in vitro, we performed mosaic experiments by mixing TrioN overexpressing endothelial cells with control-transfected endothelial cells ( Supplementary Fig. 10m), and compared cell size with a scenario in which all (neighboring) cells were expressing TrioN (homogeneous expression scenario). In both the mosaic and in the homogeneous scenario we find increased EC size in TrioN overexpressing cells (Fig. 7n). The magnitude of cell size increase was not significantly different between the two scenarios ( Fig. 7n).
Endothelial cells anchor to the extracellular matrix (ECM) through focal adhesions. In endothelial lamellipodia extensions, anchoring to the ECM is necessary for the buildup of tractional forces that regulate forward extension and cell shape. We hypothesized that to maintain their shape, the enlarged endothelial cells have to be able to stably anchor to the ECM. In line with this, we found that TrioN massively increased the number of focal adhesions, in particular near cell-cell junctions, as judged by phosphorylated Paxillin staining (Fig. 7o-s). TrioN  (Fig. 8g). Instead, and in-line with our in vitro observations, TrioN increased endothelial surface area in aISVs (Fig. 8f). The aISV diameter increase was therefore solely attributed to larger endothelial cells. The larger aISVs upon TrioN overexpression were perfused and nonleaky as evidenced from dextran extravasation analyses (Supplementary Fig. 11a-d).
We next overexpressed TrioN under control of the flt1 enh promoter in plgf musc embryos ( Fig. 8h-o). Overexpression of TrioN in plgf musc embryos resulted in significantly larger aISVs and endothelial surface area, when compared to plgf musc alone ( Fig. 8m, n, o, u). On an average, combining plgf and TrioN resulted in 2.5-fold larger aISV diameters (Fig. 8o); whereas in plgf musc embryos the increase was 2-fold (Fig. 8o). The enlarged aISVs in plgf musc + TrioN gain of function embryos were actively perfused (Supplementary Movie 8). Overexpression of TrioNmut (TrioN with inactive GEF1 domain) in plgf musc transgenics failed to produce any additive increase in aISV diameter or endothelial surface area (Fig. 8o). Taken together, these data suggest that combining plgf with artery specific Trio transgenic gain-offunction results in arteries having multiple enlarged endothelial cells, which additively facilitate the arterial outward lumen remodeling.
We furthermore found that VE-Cadherin was not required for TrioN-induced endothelial cell size increase in vivo. Overexpression of TrioN in aISVs of ve-cadherin morphants (cdh5-MO) or ve-cadherin −/− mutants augmented endothelial area at a similar magnitude as observed in TrioN gain-of-function embryos with intact ve-cadherin expression (Fig. 8p-t; Supplementary  Fig. 11e).
diameters, and EC size, in WT and in plgf musc embryos (Fig. 8v, w, Supplementary Fig. 11g, h). In vitro, human aorta (HAEC) and human umbilical artery derived endothelial cells (HUAEC) showed a significantly increased size upon TrioN overxpression ( Supplementary Fig. 11f). As Vegf is the major driver of Trio activation in our setting, we hypothesized that our findings regarding enlargement may be conserved in other vegf gain of function scenarios. Von Hippel Lindau (Vhl) is a protein relevant for probing Hypoxia Inducible Factor-1α (HiF-1α), the main driver of vegf expression, for proteasomal degradation. Vhl −/− mutants show increased vegfaa expression, in particular in developing neuronal tissue 29 . Accordingly, vhl −/− mutants showed increased cerebral arterial diameters, without significant change in endothelial cell numbers ( Supplementary Fig. 11i-p). Vhl morphants also showed increased vessel diameter, concomittantly with increased EC surface area (Supplementary  ] NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-19008-0 ARTICLE Fig. 11q-u). Blocking Trio function using ITX3 inhibited the diameter increase in vhl morphants ( Supplementary Fig. 11s, u). This suggests that Trio-mediated cell and vessel dimension changes is conserved in this Vegf gain of function scenario.
In zebrafish embryos, loss of the TGFβ co-receptor endoglin augments aortic diameter in response to flow increases 10 . Morpholino-mediated loss of endoglin increased aorta diameter at 3dpf, not 2dpf, in-line with the endoglin mutant phenotype (Supplementary Fig. 12a-r, u). We found no evidence for loss of endoglin and gain of Trio acting synergistically to influence aISV or aortic diameters in 2dpf embryos ( Supplementary Fig. 12a-i; Supplementary Fig. 12j-r). In vitro, cell size of TrioN overexpressing cells was not significantly different from the size of TrioN expressing cells in which endoglin was silenced (Supplementary Fig. 12s, t). At 3dpf endoglin morphants showed an increased aorta diameter similar to the phenotype reported in endoglin mutants ( Supplementary Fig. 12u) 10 . Since in the endoglin mutant the change in aorta diameter was attributed to an increase in EC size and not EC number, we considered a potential contribution of Trio in mediating the diameter increase. Accordingly, loss of Trio in endoglin morphants, or inhibiting Trio function using ITX3 in endoglin morphants inhibited the aorta diameter increase and reduced EC surface area (Fig. 9a, b). This suggests that at 3dpf the TGFβ pathway may act as an inhibitor of Trio function and diameter control.

Discussion
Here we identified the GEF Trio as a central signaling hub in driving endothelial cell enlargement and outward arterial lumen diameter remodeling, downstream of Vegf receptor-2 signaling. Endothelial cell enlargement requires the GEF1 domain of Trio, and activation of the Trio-GEF1 targets Rac1 and RhoG, inducing F-actin remodeling events, focal adhesions, and actomyosin tension specifically at the cell periphery. Actomyosin-activity induced tension at extracellular matrix anchors in lamellipodia extensions is subsequently conveyed into a homothetic expansion of the endothelial cell. Actin remodeling in the endothelial cell periphery requires activation of Rho GTPases specifically in junctional regions. In line with this, we show that Trio activates Rac1 in the cell periphery. Active Trio is expressed at junctional regions, and Trio also directs other GEFs including Tiam1 to the junctions. Tiam1 has thus far not been linked to Vegf receptor signaling. Tiam1 is a known activator of Rac1, and we show that Tiam1 promotes endothelial cell size, similar to Trio. Combining Trio and Tiam1 overexpression has no additive effect on augmenting endothelial cell size, thus leaving open the option that Trio and Tiam1 may act in a linear way to induce Rac1 activity at junctional regions. Trio and Tiam1 are not the only GEFs that can activate Rac1 and it is conceivable that activation of additional Rac1 pools, at different intracellular locations and coupled to other downstream effectors, also contributes to the complex process of cell enlargement.
Besides activation of Rac1, the GEF1 domain of Trio is also recognized for its ability to exchange GTP on the related small GTPase RhoG. Similar to Rac1, we show that RhoG gain-offunction results in larger endothelial cells, while loss of RhoG reduces Trio-induced cell size. How RhoG contributes to endothelial cell size increase may however differ from Rac1, as RhoG is dispensable for lamellipodia formation 37 . We noticed that upon Trio activation, the distribution of the Integrins β1 and α5 shifted from a well-known focal adhesion pattern towards a more junctional pattern. Also Paxillin, a well-recognized marker for focal adhesions shifted to the junction regions. Integrins may be involved in the regulation of Trio-mediated increase in cell size by controlling ECM contacts near junctions as a means to keep the enlarged cell in its shape and preventing it from collapsing. RhoG may contribute herein as it has been shown to modulate focal adhesion turnover 46 and microtubule dynamics, at least in cancer cells and epidermal keratinocytes [46][47][48] . Vegf has also been shown to activate RhoG independently of Trio 49 . Interestingly, here RhoG subsequently activated Rac1 through the atypical GEF DOCK4, whereas SGEF, instead of Trio, was the responsible upstream activator of RhoG. Thus, downstream from VEGF, SGEF, and Trio may work in parallel, to activate RhoG and Rac1, respectively. Both pathways may also function as a reciprocal backup, as it was reported that SGEF knock out animals do not show any vascular malformations 50 .
Increasing Vegf-Trio signalling strength in developing arteries and augmenting both arterial endothelial cell size and their number allows for a 2-2.5-fold structural increase of arterial lumen diameter in vivo (Fig. 10a, b). Using Vegf to obtain arterial diameter growth in aISVs in vivo requires precise fine-tuning of local Vegf signaling strength in arterial endothelium. This can be achieved by targeting the arterial Flt1-Vegf binding equilibrium with the Flt1 specific ligands Plgf and Vegfb. We demonstrate that developing arteries express Flt1 protein, and analyses of flt1 mutants and flt1 morphants shows that arterial Flt1 traps sufficient endogenous Vegf to modulate arterial diameter growth. Overexpression of Plgf or Vegfb, and displacing the arterial Flt1trapped Vegf, results in an artery specific Vegf gain of function scenario. This triggers an increase in arterial endothelial cell size and number, culminating in a two-fold increase in arterial diameter. Plgf-induced arterial diameter growth requires Vegfaa-Kdrl signaling, but is independent of mFlt1 signaling, as the  mCherry-TrioN (TrioN), or mCherry-TrioGEF1 (TrioGEF1). Scale bar, 25 µm. g Endothelial cell surface area in ECs expressing mCherry, mCherry-TrioN, or mCherry-TrioGEF1. Mean ± s.e.m, unpaired two-sided students t-test, n = 59, 69, 63 cells per indicated group. ***p < 0.001. h-k Confocal images of ECs transfected with mCherry, photo-activatable Rac1 (mCherry-Rac1-PA-WT), mutated photo-activatable Rac1 (mCherry-Rac1-PA-C450A), or constitutively active Rac1 construct (mCherry-Rac1-Q61L). Red arrowheads indicate cell-cell junctions. Scale bar, 25 µm. l Quantification of images in h-k. Mean ± s.e.m, two-sided Mann-Whitney U test, n = 67, 56, 65, 74 cells per group. *p = 0.0477, ***p < 0.001. m Surface area of ECs transfected with control plasmid or constitutive active RhoG Q61L. Mean ± s.e.m, unpaired two-sided students t-test, n = 89, 83 cells per group, from three independent experiments. ***p < 0.001. n Western blots for Rac1 and RhoG; control of knockdown efficiency for shRNA shRac1 and shRhoG as used in O, P and actin for protein loading control, as indicated. o Change in EC size upon TrioN overexpression after silencing of Rac1. Mean ± s.e.m, unpaired two-sided students t-test, n = 73, 97 cells from three independent experiments. ***p < 0.001. p Change in EC size upon TrioN overexpression after silencing of RhoG. Mean ± s.e.m, unpaired two-sided students t-test, n = 101, 146 cells per group from three independent experiments. ***p < 0.001. q-s Endothelial cells transfected with control plasmid (ctrl) and stained for Tiam1 (q), F-actin with phalloidin (r) and VE-Cadherin as junction marker (s). Tiam1 localizes cytosolic and to a lesser extent to junction regions. Scale bar, 20 μm. t-v Endothelial cells transfected with TrioN and stained for Tiam1 (t), F-actin with phalloidin (u) and VE-Cadherin as junction marker (v). Tiam1 localizes at junction regions (arrowheads). Scale bar, 20 μm. w Endothelial cell size of ECs transfected with Tiam1, TrioN, or transfected with both Tiam1 and TrioN. Cell size was measured of n = 78, 82, 103 and 80 cells derived from three separate experiments. Mean ± s.e.m, unpaired two-sided students t-test. ns not significant, ***p < 0.001.
growth effect of Plgf is maintained in mflt1 mutants. Arterial endothelial cells of aISVs enlarge considerably in plgf musc transgenics. Consistent with Trio driving endothelial enlargement in this Vegf gain of function scenario, we find that inhibiting Trio reduces endothelial cell size, and aISV diameter in plgf musc . Our model suggests that arterial diameter growth in plgf musc involves a combination of both increased cell number and enlarged size, enabling the shaping of a structurally larger lumen (Fig. 10a, b).
Trio gain-and loss-of-function experiments in WT and plgf musc embryos show that Trio augments arterial lumen dimensions through modulation of endothelial cell shape. The in vitro data suggest that Trio regulates cell shape in a cell-  autonomous manner. Trio furthermore augments junctional integrity and stabilizes endothelial-endothelial cell-cell junctions. Such increased junctional stability would be beneficial to keep expanding vessels sealed and prevent plasma leakage during the outward remodeling process 45,51 . The TGFβ co-receptor Endoglin controls arterial diameter in response to shear stress through modulation of endothelial cell shape 6,10 . With respect to the regulation of 2dpf aISV and aorta diameter, we find no interaction between Vegf-Trio signaling and Endoglin. However, observations in 3dpf aorta suggest that Trio may be required for mediating aorta diameter increase upon loss of endoglin. How loss of endoglin couples to Trio remains to be determined but human endothelial cells devoid of Endoglin show altered VEGFR-2 kinetics, and exhibit differential activation of VEGFR-2 downstream pathways, including AKT 6 . We hypothesize that Vegf can activate Trio during early stages of arterial network remodeling, whereas the TGF beta pathway may act to restrict Trio function once flow becomes more dominant during the maturation of the arterial network. Although we originally investigated activation of Vegf-Trio as a means to augment arterial diameter in the context of improving perfusion for ischemic cardiovascular disease, the notion that inhibiting Trio and reducing EC and arterial size, may have therapeutic relevance as well for preventing shunt formation and vascular complications in, for example hereditary hemorrhagic telangiectasia (HHT) 10,11 . In early zebrafish embryos, mural cells are absent and vessel size is determined solely by the endothelium. In adult resistance sized arteries the presence of several layers of smooth muscle cells, and sympathetic nerve induced vascular tone may restrict the impact of Trio on diameter remodeling. It therefore seems unlikely that activating Trio and promoting EC size, can directly affect arterial diameter independent of vascular smooth muscle function in mature resistance sized arteries. Flt1 and its ligands PlGF and VEGF-B can regulate vascular adaptation in ischemic cardiovascular conditions in mouse, rat and pig models [52][53][54][55][56] . In rats, constitutive cardiomyocyte specific expression of full length Vegf-b results in enlargement of all coronary arteries 53 . In the coronaries, Flt1 and Kdr only colocalize in the most distal areas and capillaries, whereas coronary arteries in the proximal part express only Flt1. Based on the juxtapositioning of Flt1 and Kdr, it is therefore conceivable that arterial enlargement commences in precapillary arteries and that this is propagated retrogradely to the proximal areas 53 . We previously showed that such retrograde communication mechanism between micro-and macrocirculation, exists in an ischemic setting and promotes arterial remodeling 57 . In a more physiological context, increases in metabolism or tissue hypoxia can trigger the local release of Vegf. Vegf-Trioinduced endothelial enlargement and diameter increase reduce the resistance to flow. At a given pressure gradient this results in a redistribution of flow toward the enlarged vessel segment in the Vegf-producing or hypoxic region. We hypothesize that such vessel adaptations allow matching of changes in metabolism with flow delivery. We propose that this mechanism may operate in particular in vessel segments that lack the ability to actively regulate vascular tone by smooth muscle cells, which is typically the case in the most distal parts of the microcirculation, the areas where Flt1 and Kdr are co-expressed.
Generation of mutant and transgenic lines. Zebrafish embryos were injected at the one-cell stage with a glass microneedle and a microinjector (World Precision Instruments). For transgenesis tol2 mRNA, transcribed from pCS2FA with the mMESSAGE mMACHINE SP6 Transcription Kit, a kind gift from Koichi Kawakami, was injected at a concentration of 25 ng µl −1 together with Tol2 destination vectors 69 . For mutagenesis Cas9 mRNA was in vitro transcribed from the MLM3613 plasmid (Addgene plasmid #42251) using the mMESSAGE mMA-CHINE T7 ULTRA Transcription Kit (Thermo Fisher Scientific); sgRNA target sequences were cloned into DR274 (Addgene plasmid # 42250) and transcribed with the MAXIscript T7 transcription kit (Thermo Fisher Scientific). Subsequently, 1 nl of a mixture of 600 ng ml −1 Cas9 mRNA and 50 ng ml −1 sgRNA were coinjected into one-cell stage embryos 29 .
Generation of Vegfaa gain of function lines. In order to generate a constitutive overexpression construct for vegfaa165 in skeletal muscle tissue, a Gateway reaction using p5E_503unc, pME_GFP-p2A-vegfaa165, p3E_polyA, and pDestTol2CG2 was performed (Supplementary Fig. 1a). For constitutive overexpression 25 ng µl −1 of 503unc:eGFP-p2A_vegfaa were co-injected with tol2 transposase mRNA. Embryos were identified by means of the transgenesis marker cmlc2:EGFP as well as green fluorescence in the skeletal muscle tissue.
Generation of an early stop flt4 mutant allele. For the generation of the flt4 mu407 allele via CRISPR-Cas9 genome editing, sgRNA synthesis was performed according to the standard protocol 70 . The target sequence 5′-GGCTGTTATTGATGGC ACCA-3′ resides in exon 3 and expanded a BsaJI restriction site, which was used for screening and efficacy testing after PCR amplification using the primer pair 5′-GAAGGAGTGTCAAGGGGTGG-3′ and 5′-CGGTTGCACATTCCCCAAAG-3′. A mix containing 300 pg of zebrafish codon-optimized Cas9 mRNA and 20 pg of  Table 2) for homologous recombination. The oligonucleotide encodes two HA tags which are flanked by two 33 nt homology arms for the flt1 genomic region surrounding the DSB (Supplementary Fig. 2f). Founders were identified by the presence of an additional 60 bp larger PCR product, amplified with flt1_E3_gDNA_fw and flt1_E3_gDNA_rev, and in-frame integration of the double HA tag was verified by Sanger sequencing of the PCR product. The knockin line was viable, showed no vascular defects compatible with HAtagged Flt1 being functional.
Generation of the Tg(flt1 enh :sflt1_Δ7-HAHA) ka612 line. For overexpression of sFlt1 tagged with two HA tags, sflt1 was amplified with Phusion High-Fidelity DNA polymerase (NEB) from cDNA with the primers Flt1_XhoI_fw and sFlt1_HAHA_Xba and cloned into the MiniTol2 vector containing flt1 promoter/ enhancer elements (flt1 enh ) 63 using the restriction enzymes XhoI and XbaI. The resulting sflt1 overexpression construct flt1 enh :sflt1_wt-HAHA ( Supplementary  Fig. 2g, upper panel) was injected at a concentration of 12.5 ng µl −1 together with 25 ng µl −1 tol2 mRNA. Due to abrogation of ISV growth with this sflt1 overexpression construct ( Supplementary Fig. 2i-k), the Vegfaa binding site of sFlt1 was deleted (Supplementary Fig. 2h). Therefore, interface residues of the human sFLT1 receptor domain 2 binding to human VEGF-A were identified with PyMOL (www.pymol.org). The corresponding positions in the zebrafish sFlt1 homologue were identified as E134 to Y140 (Supplementary Fig. 2h), encoded by the base pairs 403-423. These were deleted from flt1 enh :sflt1_wt-HAHA by site-directed mutagenesis (SDM) with the primers SDM_Flt1delta7_fw and SDM_Flt1delta7_rev. For SDM the PCR product was phosphorylated with T4 polynucleotide kinase, followed by blunt end ligation with T4 DNA ligase (both NEB). Remaining template plasmid was digested with DpnI. 25 ng µl −1 of the resulting flt1 enh :sflt1_Δ7-HAHA Tol2 expression construct ( Supplementary Fig. 2g, lower panel) were injected with tol2 transposase mRNA, fish were raised and founders of the Tg(flt enh :sflt1_Δ7-HAHA) ka612 line were identified by PCR using the primers sflt1_fw and HA_rev.
Trio gain of function. TrioN contains the N-terminal part of Trio (amino acids 1-1685) including the Sec14 domain, spectrin repeats, the Rac1/RhoG GEF1 domain and SH3 domain, but excluding the RhoA-specific GEF2 domain and kinase domain 72 . For TrioN overexpression under the flt1 enh promoter in zebrafish embryos, the p5E_flt1enh, pENTR1a_GFP-TrioN and p3E_polyA sequences were recombined into the pDestTol2CG2 in a Gateway reaction (LR clonase II plus; Thermo Fisher Scientific). The resulting expression construct was injected into the cytoplasm of one-cell stage embryos at a concentration of 25 ng µl −1 together with tol2 mRNA. For generation of the stable transgenic line Tg(flt1 enh :GFP-TRION) ka615 (referred to as flt1 enh :TrioN), injected fish were raised and founders were identified by means of the transgenesis marker cmlc2:EGFP. Overexpression of TrioN under the kdrl and fli1a promoter in zebrafish embryos was performed likewise via Gateway recombination reactions using the respective p5E_kdrl and p5E_fli1a plasmids. The mutant GFP-TrioNmut construct (TrioNmut) was generated by site-directed mutagenesis of nucleotides 4399-4405 (NM007118) AAT-GAC to GCTGCC. This results in the amino acid mutations N1406A and D1407A, located in the C-terminus of the DH1 domain. The mutated GEF domain is still able to bind Rac1, but unable to induce guanine nucleotide exchange. Overexpression of TrioNmut under the flt1 enh promoter was performed similarly to TrioN. For Trio gain of function in vitro, primary HUVECs, which were acquired from Lonza (Verviers, Belgium) and regularly checked for mycoplasma contamination, were seeded on culture flasks. HUVECs were grown in EGM-2 medium (Promocell, Heidelberg, Germany). Cells were cultured on fibronectin-coated glass coverslips and transfected with human GFP-TrioN, mCherry-TrioN, mCherry-TrioNmut, mCherry-TrioGEF1 or the isoform TrioB. The mCherry-TrioN/TrioNmut/TrioGEF1 constructs were generated by replacing GFP in GFP-TrioN/TrioNmut/TrioGEF1 with mCherry using AgeI and Kpn2I restriction enzymes. After treatment, cells were washed with warm PBS, containing 1 mM CaCl 2 and 0.5 mM MgCl 2 , and fixed in 4% (v/v) formaldehyde for 10 min. After fixation, cells were permeabilized in PBS supplemented with 0.1% (v/v) Triton X-100 for 10 min followed by a blocking step in PBS supplemented with 2% (w/v) BSA and stained as indicated 45 .

RNA interference. Inhibitory shRNA constructs (Supplementary
RNA interference using siRNA constructs (Supplementary Table 4) was performed as published 37 .
Chemical and inhibitor treatments. Embryos were dechorionated prior to chemical and inhibitor treatments using 1 mg ml −1 Pronase (Roche, Basel, Switzerland). The chemicals/inhibitors were added at 32 hpf, followed by incubation at WT + ITX3 Fig. 9 Trio and endoglin interact to determine 3dpf aortic EC size. a Dorsal aorta diameter at 3dpf in endoglin morphants (magenta bar), endoglin morphants injected with 0.8 ng Trio targeting morpholino (blue bar), endoglin morphants treated with Trio inhibitor ITX3 (dark green bar) and WT treated with ITX3 (light green bar). Note that loss of Trio or inhibiting Trio reduced diameter growth in endoglin morphants. Mean ± s.e.m, unpaired two-sided students t-test, n = 8, 9, 8, 10 independent embryos per group. **p = 0.0011, ***p = 0.001. b Aorta EC surface area at 3dpf in endoglin morphants (magenta bar), endoglin morphants injected with 0.8 ng Trio targeting morpholino (blue bar), endoglin morphants treated with Trio inhibitor ITX3 (dark green bar) and WT treated with ITX3 (light green bar). Note that loss of Trio or inhibiting Trio reduced EC surface area in endoglin morphants. Mean ± s.e.m, unpaired twosided students t-test, n = 18, 20, 22, 16 cells derived from six independent embryos per group; ***p < 0.001; **p = 0.0055; *p = 0.0100. 28.5°C in the dark until the embryos were analysed at 50 hpf. For chemicals/ inhibitors dissolved in DMSO, control embryos were mock treated with DMSO (Sigma). Inhibitor and chemical concentrations are listed in Supplementary Table 5 and 6. Embryos were randomly assigned to experimental groups. Investigators were blinded to inhibitor treatment. In vitro chemical treatments were performed at indicated time points. Inhibitor concentrations are further specified in Supplementary Table 5.
Immunohistochemistry of zebrafish embryos. Fixed embryos were permeabilized for 1 h in PBT (1% (v/v) Triton X-100 in 1x PBS). In order to reduce unspecific background, samples were subsequently incubated for 2 h in blocking buffer (1% (v/v) Triton X-100, 3% (w/v) BSA in 1× PBS). The primary antibody (Supplementary Table 7) was diluted in 0.5% (v/v) Triton X-100, 1% (w/v) BSA in 1× PBS, and samples were incubated overnight at 4°C. Embryos were washed five times for 10 min with PBT before the secondary antibody (Supplementary Table 7) was added for 2 h at room temperature. Before imaging, samples were washed five times for 20 min in PBT.
Cell lines and immunofluorescent stainings in vitro. Antibodies used for immunofluorescent staining of human endothelial cells in culture are indicated in Supplementary  Table 7).
Western blotting. SDS-PAGE samples were analyzed on 7.5, 10, or 15% (w/v) polyacrylamide gels, depending on the size of the proteins of interest, and transferred onto nitrocellulose membranes (Whatman, Piscataway, NJ). Subsequently of blocking in 5% (w/v) low-fat milk in Tris-buffered saline Tween 20 (TBST), the blots were incubated with respective primary antibodies (Supplementary Table 7) for 1 h at room temperature, washed in TBST and incubated with respective HRPcoupled secondary antibodies (Supplementary Table 7) for 1 h at room temperature. Blots were developed via enhanced chemiluminescence (ECL) (Thermo-Scientific, Etten-Leur, The Netherlands). For Trio protein expression, 3-8% (w/v) Tris-acetate precast gels (ThermoFisher) were used according to the manufacturer's instructions, and samples were transferred onto nitrocellulose membranes by blotting for 18 h at 20 mA. Full Western blots in Supplementary Fig. 13.
Laser ablation. For laser ablation experiments HUVECs were transduced with LifeAct-mScarlet expressing lentiviral particles. The 442 nm laser was used to locally cut specific F-actin bundles. F-actin bundles were ablated for 10 seconds using full laser power and retracted beam expander to concentrate the laser beam intensity. Initial recoil of the cut F-actin bundles was measured immediately after ablation.
Confocal microscopy. Zebrafish larvae were embedded in 0.7% (w/v) low-melting agarose (NuSieve GTG Agarose, Lonza) in 35 mm glass bottom microscopy dishes (MatTek). The agarose was covered with E3 medium supplemented with 0.112 mg ml −1 Tricaine and 0.003% (w/v) PTU (Sigma). Confocal t-(for time-lapse imaging) and z-stacks were acquired using a Leica SP8 confocal microscope with ×20 multi-immersion and ×40 water immersion objectives, resonance scanner, HyD detectors and LAS X software. Images are displayed as maximum intensity projections of the z-stacks. For imaging of red blood cell perfusion, six images per second were acquired at a single sagittal plane in the middle of the aISVs with the SP8 confocal microscope. The transmitted light channel was used for the red blood cells and the mCherry channel for the Tg(kdrl:hsa.HRAS-mcherry) s916 signal marking the endothelial cell membranes. Images were processed with ImageJ/ Fiji. aISV numbers are indicated in figure legends. For cell culture experiments, t-and z-stack image acquisition was performed on a confocal laser scanning microscope (Leica SP8) using a ×40 NA 1.3 or ×63 NA 1.4 oil immersion objective and LAS X software. Maximum intensity projections were also generated and analyzed using ImageJ/ Fiji software.
Transmitted light microscopy. Zebrafish larvae were anaesthetized in 0.112 mg ml −1 Tricaine and transferred to E3 medium in a glass bottom microscopy dish. Red blood cell perfusion was visualized by imaging a single sagittal plane in the middle of two ISVs using a Zeiss Axioskop 5 microscope with ×40 objectives and ×4 optical gain for 40 ms exposure time.
Microangiography. Embryos were anaesthetized with 0.112 mg ml −1 Tricaine in E3 medium and mounted in 0.7% (w/v) low-melting agarose (NuSieve GTG Agarose, Lonza) in glass bottom microscopy dishes (MatTek). For the microangiography Texas Red-dextran with a molecular weight of 70 kDa (Thermo Fisher Scientific, Supplementary Table 6) was solubilized in E3 medium to a concentration of 2 mg ml −1 . Using a glass microneedle and a microinjector (World Precision Instruments), the dextran was injected into the sinus venosus. Subsequently, images were taken with an SP8 confocal microscope (Leica).
Measurements of aISV diameter, DA diameter, endothelial cell sizes, and numbers. For vessel parameter measurements, confocal z-stacks of ISVs plus dorsal aorta were acquired of up to five aISVs per animal in the middle of the yolk sac elongation and parameters were measured with Fiji using maximum projection images of the z-stacks. For aISV diameter analysis, lateral lines were drawn at seven evenly distributed positions along the aISV perpendicularly to the vessel wall between the dorsal aorta and the DLAV, approximately every 18 µm. The average of these seven lines was calculated as the average aISV diameter. The remodeling index (RI) was calculated as follows: RI = (aISV diameter of the respective treatment group in µm)/(average aISV diameter of the WT control group). For zebrafish mutants and transgenic lines more than 50 embryos derived from more than three different breedings were analysed per genotype. In morpholino experiments morphologically malformed embryos were excluded from analysis. To determine the dorsal aorta diameter, seven to nine vertical lines were drawn at ISV sprouting positions and between two ISVs perpendicularly to the aortic vessel wall at the level of the yolk sac elongation. Zeiss Zen 2012 microscope software was used to control the system. Offline ratio analyses between Cer3 and Venus images were processed using the MBF ImageJ collection. Image stacks were background corrected, subsequently aligned, and a smooth filter was applied to both image stacks to improve image quality by noise reduction. An image threshold was applied exclusively to the Venus image stack, converting background pixels to "not a number (NaN)," eliminating artefacts in ratio image derived from the background noise. Finally, the Venus/Cer3 ratio was calculated with high activation shown in red/white and low activity in blue/black. For FRET quantification, data were analysed using a MatLab script (MATLAB, The MathWorks, Inc., Natick, Massachusetts, United States). Prior to any of the ratiometric FRET analyses, donor CFP and acceptor YFP channels were background corrected by subtracting the modal pixel value, and both channels were aligned to each other. Prior to watershed segmentation, cells were manually selected by adding seed points and touching cells were separated by manually drawing boundaries between them. A local threshold was applied to the watershed region, exclusively including pixels that were higher than 15% of the maximum intensity in that region. For all channels the mean fluorescence intensity was calculated for each segmented region. Ratiometric FRET analysis was applied to each segmented region in the cell. The endothelial junction marker VE-cadherin was used as positive control. Quantification of FRET signal overlapping with VEcadherin was done between the cells edge and 10 pixels inwards. These areas were defined as junction regions.
Resistance measurements. Monolayer integrity was determined by measuring the electrical resistance using ECIS as published 37 . Electrode-arrays (8W10E; IBIDI, Planegg, Germany) were treated with 10 mM l-cysteine (Sigma) for 10 min and subsequently coated with 10 μg/ml fibronectin (Sigma) in 0.9% NaCl for 1 h at 37°C. Cells were seeded at 40.000 cells per well (0.8 cm 2 ) and grown to confluence. Electrical resistance was continuously measured at 4000 Hz at 37°C under 5% CO 2 using ECIS model 9600 (Applied BioPhysics, New York, MA).
Statistical analysis and reproducibility. Statistical analysis was performed using GraphPad Prism 6. Each dataset was tested for normal distribution with the D'Agostino-Pearson test. Only if the data were normally distributed, a parametric method (unpaired two-sided Students t-test) was applied. A nonparametric test (two-sided Mann-Whitney test) was applied for non-normally distributed data sets. In case of multiple comparisons, one-way ANOVA plus Bonferroni correction or 2-way ANOVA were applied. P values < 0.05 were considered significant. Data are represented as mean ± s.e.m., unless otherwise indicated. *P < 0.05, **P < 0.01 and ***P < 0.001, # P < 0.001 to indicated ctrl group. No data were excluded from statistical analysis. No statistical method was used to predetermine sample size. For every treatment, treated and control embryos were derived from the same egg lay; embryo were selected on the following pre-established criteria: normal morphology, a beating heart and circulating red blood cells. All images shown in the figures are representative examples of the respective phenotypes and expression patterns.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.