Notch regulates BMP responsiveness and lateral branching in vessel networks via SMAD6

Functional blood vessel growth depends on generation of distinct but coordinated responses from endothelial cells. Bone morphogenetic proteins (BMP), part of the TGFβ superfamily, bind receptors to induce phosphorylation and nuclear translocation of SMAD transcription factors (R-SMAD1/5/8) and regulate vessel growth. However, SMAD1/5/8 signalling results in both pro- and anti-angiogenic outputs, highlighting a poor understanding of the complexities of BMP signalling in the vasculature. Here we show that BMP6 and BMP2 ligands are pro-angiogenic in vitro and in vivo, and that lateral vessel branching requires threshold levels of R-SMAD phosphorylation. Endothelial cell responsiveness to these pro-angiogenic BMP ligands is regulated by Notch status and Notch sets responsiveness by regulating a cell-intrinsic BMP inhibitor, SMAD6, which affects BMP responses upstream of target gene expression. Thus, we reveal a paradigm for Notch-dependent regulation of angiogenesis: Notch regulates SMAD6 expression to affect BMP responsiveness of endothelial cells and new vessel branch formation.

B lood vessel networks expand via endothelial cell (EC) sprouting, migration, anastomosis and lumenization to form new conduits, a process called sprouting angiogenesis 1 . Heterogeneous yet coordinated responses to pro-angiogenic signals are essential for proper angiogenesis and for subsequent maintenance of a functional vasculature [2][3][4][5][6] . Numerous signalling pathways integrate to provide this heterogeneity in ways that are not completely understood 5,7,8 .
Bone morphogenetic protein (BMP) signalling is essential to the proper form and function of blood vessels [8][9][10][11][12][13][14] . BMPs belong to the tranforming growth factor-b superfamily of secreted growth factors and they signal through cell-surface heterotetramers of type II and type I serine/threonine kinase receptors 15,16 . Phosphorylation of type I receptors by type II receptors induces phosphorylation of the receptor-associated SMADs 1, 5 and 8 (R-SMADs 1/5/8). R-SMAD phosphorylation induces association with SMAD4 and this complex translocates to the nucleus, to regulate expression of target genes. SMAD6 is an inhibitory SMAD that competitively binds type I receptors or SMAD4 to inhibit SMAD1/5/8 phosphorylation or nuclear translocation, respectively, and downregulate BMP signalling 17,18 .
Recent studies highlight novel roles for BMPs during angiogenesis [8][9][10][11][12][13][14] . Global genetic deletion of SMAD5 in mice led to multiple angiogenesis defects during embryogenesis 19 . EC-selective deletion of SMAD1 and SMAD5 severely impaired angiogenesis, resulting in defective yolk sac and cranial vasculature, and SMAD1/5 double knockout EC displayed reduced proliferation and migration, implicating BMP signalling in cellular processes essential to angiogenesis 8 . In zebrafish, Bmp2b is a venous-specific pro-angiogenic signal that promotes Vegfaa-independent sprouting from the posterior cardinal vein 9 . The type I BMP receptor ALK1 is either anti-angiogenic or pro-angiogenic when inhibited by Fc-conjugation or a highly specific blocking antibody, respectively 10,13 . These data suggest that the angiogenic activity of BMP ligands is context dependent.
EC differentially respond to BMP signalling 9 , but how BMP responsiveness is set remains largely unknown. Therefore, we sought to identify a mechanism by which EC intrinsically regulate the magnitude of their responses to BMP ligands. Previous studies indicated that BMP6 and BMP2 can be pro-angiogenic 8,9,20 . Here we show that these ligands are pro-angiogenic, and that lateral vessel branching requires threshold levels of R-SMAD phosphorylation. EC responsiveness to pro-angiogenic BMP ligands is regulated by Notch status. Notch sets responsiveness by regulating the cell-intrinsic BMP inhibitor, SMAD6 and SMAD6 affects BMP responses upstream of target gene expression, revealing a new paradigm for Notch-dependent regulation of angiogenesis.

Results
BMP2 and BMP6 promote lateral vessel branching. We first determined the effects of BMP6 and BMP2 on vessel branching in a three-dimensional (3D) sprouting angiogenesis assay 21 using human umbilical vein endothelial cells (HUVEC). Addition of exogenous BMP6 or BMP2 significantly increased both branching frequency and branch angle compared with controls ( Fig. 1a-d and Supplementary Fig. 1a-d). Exogenous BMP2 had a similar effect on branching frequency in mouse embryonic stem (ES) cell-derived vessels (Supplementary Fig. 1e-g). Consistent with a more highly branched phenotype, both ligands increased the percentage of nuclei in the tip position of growing HUVEC sprouts ( Fig. 1e-g and Supplementary Fig. 1h-j). Moreover, the BMP-induced branching phenotype was remarkably well-organized, in contrast to the effects of elevated vascular endothelial growth factor (VEGF)-A signalling, which often lead to gross dysmorphogenesis 22,23 . Thus, BMP6 and BMP2 are pro-angiogenic and increase vascular density by inducing new branch formation. Furthermore, the increased branch angles lead to lateralization of the branching pattern and suggest that pro-angiogenic BMP signalling tunes vessel branching between a 'bush-like' and a 'bamboo-like' vessel pattern, in a liganddependent manner.
We hypothesized that BMP-induced branching depended on EC sprouting in response to activation and nuclear localization of phosphorylated SMADs 1 and 5 (pSMAD1/5). We examined levels of nuclear pSMAD1/5 in EC during sprouting angiogenesis and found a striking inverse correlation between the distance of nuclei from the sprout tip and intensity of nuclear pSMAD1/5 staining (Fig. 1h,j,m). Thus, EC with the highest levels of BMP signalling were more likely to be in the tip cell position. As expected, short-term BMP6 treatment significantly increased overall nuclear pSMAD1/5 levels compared with controls ( Fig. 1i-k and Supplementary Fig. 1k). However, although acute BMP6 exposure significantly increased the average nuclear pSMAD1/5 levels in stalk cells, it had only a modest and nonsignificant effect in tip cells (Fig. 1l), suggesting that EC at sprout tips have maximal BMP pathway activation. In fact, stalk EC in BMP6-treated sprouts had average nuclear pSMAD1/5 levels comparable to levels in tip EC of controls (Fig. 1l), suggesting that BMP signalling increases vessel branching by inducing additional tip cells from the pool of stalk cells. This idea is also consistent with the finding that stalk cells, even after BMP stimulation, retain extensive heterogeneity in nuclear pSMAD1/5 levels and their pSMAD1/5 levels do not as strongly correlate with position in the sprout as controls (Fig. 1l,m).
To verify the necessity of SMAD1/5 for lateral branch formation, we reduced the levels of each protein via small interference RNAs (siRNAs) and analysed effects on sprouting. Reduction of either SMAD1 or SMAD5 (Supplementary Figs 1l,m and 5b) reduced BMP6-induced lateral branching. Although SMAD1/5 signalling has also been implicated in stalk cell maintenance 5 , this effect is thought to be downstream of BMP9 engagement with complexes that include the type I receptor ALK1 and induce vascular quiescence 13,24,25 . To determine whether ALK1 signalling contributes to the effects we observed, we reduced ALK1 levels via siRNA and found no effect on BMP6-dependent increases in lateral branching (Supplementary Figs 1n and 5b), suggesting that BMP6-induced branching occurs independently of BMP9/ALK1 signalling. Thus, pSMAD1/5 is required to mediate the pro-angiogenic effects of BMP6 in EC independent of effects from ALK1 signalling. Taken together, these data suggest that BMP-dependent lateral vessel branching depends on EC that are 'tunable' and have an innate set point of BMP responsiveness. For each EC, this innate responsiveness sets a ligand threshold required to produce sufficient pro-angiogenic pSMAD1/5 signalling and provoke an EC response. In this scenario, short-term BMP6 treatment increases the number of EC that achieve the threshold and are capable of responding, leading to additional lateral branching.
Notch sets EC BMP responsiveness. If BMP responsiveness is innate to EC, then factors that set this responsiveness and the extent of R-SMAD phosphorylation are predicted to profoundly affect BMP-dependent vascular architecture. Notch signals via its intracellular domain (NICD) and this signalling represses the tip cell phenotype and supports a non-branching stalk cell phenotype [26][27][28][29] . Although recent work describes Notch-BMP pathway crosstalk at the level of target gene expression 8,13,30,31 , it is not clear whether Notch signalling alters the responsiveness of EC to BMP via effects on upstream BMP pathway components.
In the developing zebrafish, venous EC are BMP sensitive and form ectopic sprouts after heat-shock induction of Bmp2b, whereas intersegmental vessels (ISVs) are relatively unresponsive at 24-48 hpf (ref. 9). To investigate regulators of BMP responsiveness in this model, we used Tg(Tp1bglob:eGFP) (Notch reporter) zebrafish embryos and asked how Notch activity and BMP sensitivity align in the vasculature. Notch reporter activity was strong in the dorsal aorta (DA) and ISVs, consistent with other reports 32,33 , and these vessels do not respond to ectopic BMP ligand 9 ; however, the reporter signal was undetectable in the BMP-responsive caudal vein plexus (CVP) (Fig. 2a,b). To determine whether Notch influences BMP responsiveness, we induced Notch signalling via heat-shock induction of NICD. As we have described, ectopic induction of  Bmp2b led to excessive sprouting from the CVP (Fig. 2c,e) 9 . However, concomitant induction of Bmp2b and NICD significantly reduced the frequency of CVP sprouts, suggesting that ectopic Notch signalling dampens the sensitivity of EC to BMPs (Fig. 2d,e). Conversely, to determine whether arterial EC could be sensitized to Bmp2b overexpression, we blocked Notch signalling by treatment with N- [   g-secretase inhibitor that prevents cleavage and release of NICD. Bmp2b induction induced a low level of ectopic vessels from arterial EC ( Fig. 2f-h,l) and DAPT treatment alone induced some ectopic arterial angiogenesis, consistent with previous reports (Fig. 2i,l) 5,27 . However, Notch inhibition combined with Bmp2b induction resulted in a significantly higher frequency of ectopic arteries compared with either manipulation alone ( Fig. 2j-l).
These results indicate that Notch is an intrinsic regulator of the magnitude of the BMP response in EC in vivo.
To quantitatively determine the impact of Notch manipulations on BMP pathway activation, we determined nuclear pSMAD1/5 levels on exposure of HUVEC to different amounts of ligand. A twofold serial dose-response curve to BMP6 yielded a prototypical sigmoidal semi-log curve for BMP-mediated EC activation, as measured by nuclear pSMAD1/5 levels ( Supplementary Fig. 2a,b). We next tested the effect of Notch activation by plating HUVEC onto Fc-conjugated Dll4 ligand (Dll4-Fc) before short-term treatment with BMP6 and found that the EC 50 for BMP-mediated EC activation increased significantly compared with controls ( Fig. 2m). We confirmed, using inducible NICD expression in HUVEC, that elevated Notch signalling increased the EC 50 ( Supplementary Fig. 2c). This relationship also held at the single-cell level, as EC expressing NICD had reduced levels of pSMAD1/5 ( Fig. 2n, Fig. 2f-j). The results of in vivo and in vitro Notch manipulations support our hypothesis that Notch regulates the innate BMP responsiveness of EC, and that the increased BMP responsiveness of EC with low Notch signalling promotes lateral branching.
SMAD6 integrates Notch and pro-angiogenic BMP responsiveness. Notch regulates VEGF signalling by modulating levels of VEGF receptor RNAs 34,35 . Therefore, we reasoned that Notch would regulate BMP responsiveness via expression of BMP receptors. Surprisingly, we detected no significant changes in expression levels of several type I and type II BMP receptors after Notch stimulation of HUVEC via Dll4-Fc plating or Notch blockade via DAPT ( Supplementary Fig. 3a,b). BMP signalling is also intrinsically regulated by an intracellular inhibitory protein, SMAD6 (refs 17,18,36,37) and SMAD6 messenger RNA levels increased with Notch stimulation and decreased with Notch blockade in HUVEC (Fig. 3a). This relationship was maintained at the single-cell level, as EC expressing NICD had elevated levels of SMAD6 protein (Fig. 3b,c). In other cell types, SMAD6 inhibits BMP signalling by preventing R-SMAD phosphorylation and nuclear localization 17,18 , but its activity in EC and effects on angiogenesis are unknown. Therefore, we generated HUVEC expressing doxycycline-inducible SMAD6 fused to tdTomato. The tagged SMAD6 protein reacted with a SMAD6 antibody by immunofluorescence ( Supplementary Fig. 3c-f) and suppressed nuclear pSMAD1/5 levels in HUVEC in a cell autonomous and dose-dependent manner (Fig. 3d,e). Conversely, reduction of SMAD6 protein levels via siRNA knockdown (Supplementary Figs 3g and 5d) increased BMP6-induced nuclear pSMAD1/5 (Fig. 3f). These findings show that an intrinsic BMP pathway inhibitor, SMAD6, modulates BMP signalling in EC. As elevated pSMAD levels were associated with increased lateral branching and SMAD6 suppressed BMP signalling, we hypothesized that loss of SMAD6 would increase BMP responsiveness of EC and promote branching. Consistent with this hypothesis, reduced SMAD6 levels via siRNA significantly increased branching of sprouting HUVEC with added exogenous BMP6 (Fig. 3g-k). These results show that SMAD6 regulates angiogenesis, probably by intrinsically modulating the magnitude of EC responses to BMP inputs.
To determine the functional relevance of SMAD6 in vivo, we first examined expression by fluorescence-activated cell sorting (FACS) in EC from zebrafish embryos that expressed the Notch signalling reporter (Fig. 2a). Zebrafish have two SMAD6 orthologues, smad6a and smad6b. We assessed expression of smad6b in EC in vivo, because its sequence is more homologous to that of human SMAD6. EC sorted for high levels of Notch reporter signalling had significantly elevated levels of RNAs that are expected to be elevated in arterial Notch-positive EC such as notch1b and ephrinb2. They also had significantly elevated levels of smad6b RNA relative to EC sorted from the same embryos with little or no Notch reporter signalling (Fig. 4a). We next manipulated embryonic SMAD6 expression. As global manipulations of smad6b are predicted to profoundly perturb dorsal-ventral axis formation in zebrafish embryos 38 , we used the Tol2 system to generate F0 mosaic embryos expressing green fluorescent protein (GFP)-tagged smad6b, or GFP alone, under control of the vascular-specific fli1 promoter. We predicted that Smad6b expression would preferentially affect BMP-responsive EC with low Notch levels, such as those in the cardinal vein and the CVP. EC expressing Smad6b-eGFP were significantly reduced in the cardinal vein and CVP, and enriched in the DA and ISVs, compared with EC expressing only GFP ( Fig. 4b-d), consistent with our finding that Smad6b expression is elevated in EC of embryos with high Notch signalling reporter expression, presumably including EC from the DA and ISVs. These findings suggest that forced Smad6b expression prevents EC from colonizing the vein and CVP where Notch signalling is low, but is irrelevant in arterial EC where Notch signalling is relatively high and support a role for SMAD6 in angiogenesis in vivo.
We next asked whether reduced levels of smad6b affect blood vessel branching and predicted that if SMAD6 is important for Notch-mediated effects in vivo, its reduction would more profoundly affect Notch-dependent ISV sprouting that is normally refractory to BMP 9 . To overcome the effects of global smad6b manipulations on dorsal-ventral axis formation, we developed a tissue-specific mRNA knockdown system for zebrafish based on a CRISPRi strategy [39][40][41][42] . However, CRISPRi is poorly efficient in eukaryotic cells 40 ; thus, dCas9 was fused to the Drosophila engrailed repressor domain 43 to generate dCas9-Engrailed Repressor domain (EnR), a chimeric transcriptional repressor protein that is targeted by short guide (sg) RNAs. We validated the efficacy of dCas9-EnR in global knockdown using sgRNAs targeting bmp6 or smad6b and observed the expected dorsalization (bmp6 loss-of-function) or ventralization (smad6b loss-of-function) phenotypes ( Supplementary Fig. 4a-d). We also validated sgRNA effects at the mRNA level by quantitative reverse transcriptase-PCR (qRT-PCR) for the targeted genes ( Supplementary Fig. 4e). Next, transgenic embryos expressing a fli1 promoter-driven dCas9-EnR were generated using the Tol2 system. F1 embryos expressing vascular restricted dCas9-EnR were indistinguishable from wild-type (WT) siblings and dCas9-EnR mRNA expression was confirmed by RT-PCR (Supplementary Figs 4f-h anf 5e). To determine whether Smad6b regulates BMP responsiveness in vivo, we injected sgRNAs targeting smad6b into Tg(fli1:dCas9-EnR);Tg(hsp701: bmp2b);Tg(kdrl:GFP) embryos, then heat shocked to induce ectopic Bmp2b expression at 26 hpf, well beyond the early developmental window leading to axis defects. WT heat-shocked embryos without sgRNAs had normal ISV formation ( Supplementary Fig. 4i,j,m). Likewise, Tg(fli:dCas9-EnR);Tg(kdrl:GFP) embryos injected with scrambled sgRNAs or smad6b sgRNAs and heat-shocked had normal ISV formation (Fig. 4e,f,i). Heat-shock induction of Bmp2b expression did not perturb axis formation, but induced some ectopic ISV formation in controls and Tg(fli:dCas9-EnR); Tg(hsp701:bmp2b);Tg(kdrl:GFP) embryos injected with scrambled sgRNAs (Fig. 4g,i and Supplementary  Fig. 4k,m). However, the frequency of ectopic ISV formation significantly increased in transgenic embryos overexpressing Bmp2b and injected with smad6b sgRNAs (Fig. 4h,i; Supplementary Fig. 4l,m). Remarkably, the phenotype of BMP-overexpressing, vascular smad6b knockdown embryos resembles that of BMPoverexpressing embryos with reduced Notch signalling. Combined, these in vivo results support a model whereby SMAD6 represses BMP responsiveness and vessel branching.
Based on the similarity of phenotypes in vivo and the Notch responsiveness of SMAD6 RNA and protein in EC, we further explored the mechanism by which Notch regulates SMAD6 gene  expression. Analysis of the SMAD6 5 0 -promoter region revealed a consensus sequence for Recombination Signal Binding Protein for Immunoglobulin Kappa J Region (RBPJ) binding to DNA, which is required for Notch-dependent transcription in cells. A chromatin immunoprecipitation (ChIP) assay using two separate antibodies to RBPJ revealed that RBPJ binds at the identified site in the SMAD6 promoter ( Fig. 5a and Supplementary Fig. 5a), suggesting that Notch directly regulates transcription of SMAD6 RNA.
Finally, we hypothesized that SMAD6 is functionally downstream of Notch in regulating the BMP responsiveness of EC. We tested this hypothesis by activating Notch in HUVEC with reduced levels of SMAD6, and we assessed BMP pathway signalling via nuclear localization of pSMAD1/5. Short-term BMP6 treatment of HUVEC led to elevated nuclear pSMAD levels that were significantly suppressed by Notch stimulation (Fig. 5b,c,f,g,j). In contrast, HUVEC with reduced SMAD6 levels and BMP6 stimulation did not respond to Notch stimulation with suppression of SMAD1/5 activation, but had nuclear pSMAD1/5 levels equivalent to that of HUVEC without Notch stimulation (Fig. 5d,e,h-j). These data indicate that SMAD6 is functionally downstream of Notch and is required to mediate the effects of Notch on BMP responsiveness in EC.

Discussion
Collectively, our data support a model whereby pro-angiogenic BMPs increase vascular complexity through regulation of lateral branching. The magnitude of intrinsic BMP responsiveness in EC is modulated by Notch-mediated regulation of the BMP inhibitor SMAD6 to 'tune' BMP responsiveness (Fig. 6), and combined effects on branch frequency and angle lead to a more arborized or 'bush-like' vascular network as BMP ligand levels increase. In this model, Notch signalling sets levels of SMAD6 in a given EC and SMAD6 levels regulate the amount of pSMAD1/5 that translocates to the nucleus with a given input of BMP ligand. As ligand levels increase, more EC reach a threshold that allows them to assume a tip cell phenotype and sprout to form a new branch. This includes a subset of EC in a classic 'stalk' position, and our finding of differential BMP responsiveness in the stalk cell compartment indicates dissimilarity among stalk cells. Notch is a major regulator of the tip cell versus stalk cell phenotype in EC of angiogenic sprouts 26,44 , and we now propose that Notch also differentiates stalk cells by modulating expression of SMAD6. We predict that Notch-dependent changes in SMAD6 expression among dynamically competing sprouting EC shifts BMP responsiveness relative to a threshold necessary for lateral branching. Subsequently, as arteries reach homeostasis and Notch signalling becomes more uniform, these changes may lead to more uniform SMAD6 levels and overall dampened BMP responsiveness in arteries versus veins.  Allthough convergence of Notch and BMP pathways at downstream target genes has been described 8,13,30 , our data strongly suggest that intrinsic responsiveness to BMP signals is independently preset by Notch to regulate lateral branching. Recent work suggests that Notch also maintains the stalk cell phenotype via downregulation of Neuropilin-1 (ref. 45). Low Neuropilin-1 in stalk cells relieves inhibition of ALK1 and ALK5, and promotes tranforming growth factor-b-mediated pSMAD2/3 signalling to maintain the stalk cell phenotype. Our work shows that SMAD6, a Notch-regulated cell-intrinsic BMP pathway inhibitor, modulates EC responses to pro-angiogenic BMP ligands that stimulate pSMAD1/5 signalling and a tip cell phenotype. Thus, Notch-mediated SMAD6 regulation 'tunes' branching responses to BMP ligands among stalk cells and coordinately regulates the magnitude of BMP pathway activation and sprouting, to determine the patterning of growing vessel networks.

Methods
Cell maintenance and processing. No cell lines used in this study are found in the International cell line authentication committee (ICLAC) commonly misidentified cell line database.
Mouse ES cells were maintained and differentiated as previously described 46 . Briefly, cells were maintained undifferentiated by culture in DMEM media supplemented with conditioned media from 5,637 human bladder cancer cells (ATCC HTB9) and passaged at 2-4 days. For differentiation, ES cells aged for 2-4 days beyond passage were trypsinized, resuspended and inoculated into droplets on a tissue culture plate lid. The lid was inverted, hanging drops cultured for 2 days to generate embryoid bodies (EBs), then inverted again and flushed with media. After culture in a new plate for 1 day, EBs were seeded at 20-30 EBs per well of a 24-well plate and differentiated in DMEM, Hi-glucose, 20% FCS, 150 mM monothioglycerol and 50 mg ml À 1 gentamicin. Media was changed every 2 days until fixation. Human BMP2 (R&D Systems 355-BEC) was added at d6 and d7 at 200 ng ml À 1 . Day 8 cultures were rinsed with 1 Â PBS and fixed for 5 min in ice-cold methanol-acetone (50:50) 47 . Following 6 Â 5 min washes in 1 Â PBS, cultures were stained with rat anti-mouse PECAM-1 (BD Biosciences 553370) at 1:1,000 and then goat anti-rat IgG Alexa Fluor 488 (Life Technologies A-11006) at 1:2,000 overnight at 4°C in Staining Solution (Jackson ImmunoResearch 005-000-121), 5% goat serum, 1% BSA and 0.3% TritonX-100 in DPBS (Fisher MT-21-031-CV).
HUVEC (Lonza C2519A) were maintained as per the manufacturer's recommendations and were used in experiments from passages 2 to 5. All experiments were independently replicated using two distinct lots of HUVEC. HUVEC were certified mycoplasma-free by the UNC Tissue Culture Facility. Normal Human Lung Fibroblasts (NHLF; Lonza CC-2512) were maintained as per the manufacturer's recommendations. NHLF were certified mycoplasma-free by the UNC Tissue Culture Facility.
Sprouting angiogenesis assay. The 3D sprouting angiogenesis assay was performed as previously described 21 . Briefly, 1-2 Â 10 6 HUVEC were mixed with 100 ml of a 60,000 ml À 1 suspension of cytodex beads (GE Healthcare 17-0485-01) in 5 ml HUVEC growth media (EBM-2 þ EGM-2 BulletKit, Lonza CC-3156 and CC-3162, respectively), maintained in suspension for 4 h by mixing every 15 min, then placed in 60 mm dishes at 37°C overnight. Beads were recovered from the plate by gentle rinsing and washed 3 Â in HUVEC growth media, then resuspended in 10 ml of 2 mg ml À 1 fibrinogen/0.15 U ml À 1 aprotinin. Five microlitres of a 50 U ml À 1 thrombin solution was added to each well of a 24-well plate, followed by gently mixing in 500 ml of HUVEC-coated beads in fibrinogen/ aprotinin per well. After 5 min atRT, the plate was transferred to 37°C for 30 min for polymerization, then 1 ml of 20,000 cells ml À 1 NHLF was added to each well. Cells were fixed for 10 min in 4% PFA/DPBS, followed by permeabilization in 0.5% Triton-X 100/DPBS. Cells were blocked in staining solution for 1 h, followed by overnight incubation at 4°C with Rabbit anti-pSMAD1/5 (R&D) at 1:1,000 in staining solution (see Supplementary Table 1 for antibodies and concentrations). Goat anti-Rabbit Alexa Fluor 488 secondary antibody was added at 1:1,000 in staining solution for 2 h at room temperature (RT). For FLAG experiments, HUVEC were incubated with 1:1,000 a-pSMAD 1/5 antibody (R&D) and 1:5,000 a-FLAG antibody (Sigma) overnight, then washed and incubated with 1:500 Alexa Fluor 594 to detect pSMAD 1/5 and with 1:5,000 Alexa Fluor 488 to detect FLAG. DRAQ7 (1:1,000) was used to visualize the nucleus. Coverslips were mounted onto slides in ProLong Diamond Antifade Mountant (Life Technologies P36961). For SMAD6 staining, HUVEC were incubated in 1:100 a-SMAD6 (Abcam) and 1:5,000 a-FLAG antibody (Sigma) overnight, and then with 1:500 Alexa Fluor 594 to detect SMAD6 and with 1:5,000 Alexa Fluor 488 to detect the FLAG antibody. Alexa Fluor 647-conjugated Phalloidin was used to visualize cell boundaries.
PCR product was purified using NucleoSpin kit (Clontech 740609) and double-digested for 10 min at 37°with BglII and KpnI (Life Technologies FD0084 and FD0524, respectively). ptdTomato-N1 empty vector (Clontech 632532) was double-digested in parallel using the same enzymes and de-phosphorylated for 5 min at 37°using FastAP Thermosensitive Alkaline Phosphatase (Life Technologies EF0654). Products were ligated at a 3:1 (Insert:Vector) molar ratio for 10 min at RT using T4 Rapid Ligation Kit (Life Technologies K1422), followed by transformation, selection, plasmid purification and sequencing, to verify the fidelity of the insert.
The full-length hSMAD6-tdTomato chimera was PCR-amplified and cloned into gateway-compatible pME-MCS 49 from the newly generated template using a similar strategy.
Zebrafish transgenesis was performed using the Tol2 system 49 . To generate the pTol2 fli1:smad6b-GFP;cmlc2:GFP targeting vector, full-length smad6b was amplified from a 24 hpf embryo complementary DNA library using PrimeSTAR Max and cloned into pCS2 FLAG EV (a gift from Peter Klein, Addgene plasmid 16331) as described above, using the following primers: Forward: 5 0 -ATTCACGGATCCGCCACCATGTTCAGGACGAAACGC TCA-3 0 , containing a 6 nt overhang, BamHI site and Kozak consensus sequence, Reverse: 5 0 -ATTCACATCGATATCTGTGGTTGTTGAGGAGG-3 0 , containing a 6 nt overhang, ClaI (Life Technologies FD0143) site and 2 nt deletion from the stop codon to maintain frame with C-terminal tags.
Smad6b-GFP was then ligated into pME-MCS using a BamHI and NotI (Life Technologies FD0594) double digest as described above. A four-way LR reaction using p5E-fli1ep (a gift from Nathan Lawson, Addgene plasmid 31160), pME-smad6b-GFP, p3E-polyA 49  dCas9-EnR was subcloned into pME-MCS using BamHI/XbaI double digests and four-way LR reaction performed with the same 5E, 3E and DEST vectors described above.
siRNA transfections. siRNAs used in this study are listed in Supplementary  Table 2. HUVEC were plated at 20,000 cm À 2 onto 6 cm dishes. The following day, B80% confluent HUVEC were transfected with 10 nM siRNA using Lipofectamine RNAiMax (Life Technologies 13778150) according to the manufacturer's protocol. The following day, HUVEC were trypsinized, coated onto Cytodex beads for 3D experiments or plated onto 12 mm coverslips for two-dimensional (2D) experiments and re-transfected with siRNA to ensure maximal knockdown persisted throughout the duration of the experiments. Residual cells were analysed by western blotting, to verify knockdown efficacy.
Western blotting. HUVEC whole-cell lysates were prepared in RIPA buffer supplemented with protease/phosphatase inhibitor cocktail (Cell Signaling 5872S). Ten micrograms of whole-cell lysates were separated on 10% TGX Stain-Free FastCast SDS-PAGE gels (Bio-Rad 161-0183), followed by 2 min ultraviolet activation of TGX stain for total protein quantification. Proteins were transferred to polyvinylidene difluoride membranes (Bio-Rad 162-0177), blots were imaged under ultraviolet for total protein, blocked for 2 h in 5% non-fat milk (NFM) in PBS þ 0.1% Tween20 (PBST), then incubated overnight at 4°C with primary antibody in 1% NFM in PBST (see Supplementary Table 1 for primary and secondary antibodies and concentrations). Horseradish peroxidase-conjugated secondary antibodies (Life Technologies goat anti-rabbit G21234 or rabbit anti-mouse 816720) were added for 2 h at RT in 1% NFM in PBST. Clarity ECL (Bio-Rad 170-5060) was used for detection. Films were digitized using an AlphaImager (ProteinSimple) gel scanning station and band densities calculated in FIJI 52 , following established guidelines 53 .
Quantitative RT-PCR. Primers used in this study are listed in Supplementary  Table 3. mRNA was collected from experimental samples using Trizol reagent (Life Technologies 15596026). cDNA was generated from 1 mg mRNA using iScript reverse transcription kit (Bio-Rad 170-8891) and diluted 1:3 in water. qRT-PCR was performed using iTaq Universal SYBR Green SuperMix (Bio-Rad 172-5121) on an ABI 7900HT Fast Real-Time PCR System (Life Technologies 4329001). Threefold serial dilutions of pooled cDNA were used to generate standard curves for each amplicon and data were analysed via the Pfaffl method 54 .
RNA was isolated from FACS-enriched zebrafish cells using RNA-easy Micro Plus kit (Qiagen) and reverse transcribed with Invitrogen's Superscript III First-Strand Synthesis Supermix (Invitrogen) according to the manufacturer's instructions. For qRT-PCR, NCBI's Primer-BLAST was used to design exon-spanning, gene-specific SybrGreen primers (see Supplementary Table 3). All primers were validated by high-resolution melt analysis, size confirmation and no-template controls. SybrGreen real-time PCR was performed in triplicate on the Viia7 real-time PCR system (Invitrogen). For quantification, the DDCT method was used where raw CT values were normalized to elongation factor alpha (eef1a1l1) and paired sorting control, then calculated fold change as 2 ( À DDCT) . Statistical significance was determined by one-sample T-test comparing with a reference value of onefold change.
For global CRISPRi, WT embryos from AB parents were injected at the one-cell stage with 450 pg dCas9-EnR mRNA alone or in conjunction with 200 pg gene-targeting sgRNA. Embryos were analysed at 24 hpf for gross dorsal-ventral patterning defects consistent with early BMP manipulations 38 . For CRISPRi, 450 embryos from Tg(hsp:bmp2b);Tg(kdrl:GFP) Â Tg(fli1ep:dCas9-EnR;cmlc2:GFP) crosses were injected using a PicoSpritzer III positive pressure micro-injection apparatus at the one-cell-stage with 200 pg (in 1 nl) of scrambled (scram) or smad6b-targeting sgRNAs. Embryos were raised at 28°C in 1 Â E3 and heatshocked to overexpress Bmp2b at 26 hpf as described above. Embryos were fixed and imaged on an FV1200 confocal microscope at 44 hpf. sgRNA sequences used are listed in Supplementary Table 3.
For all zebrafish experiments, embryos were collected and sorted into treatments blindly with respect to genotype, to ensure appropriate randomization. For all experiments, any embryos that failed to gastrulate or exhibited severe developmental patterning defects were excluded from analysis. All animals are maintained in an institutional animal care and use committee (IACUC)-approved satellite facility according to public health service (PHS) policy on Humane Care and Use of Laboratory Animals.
FACS sorting. Staged fish embryos were obtained by breeding hemizygous Tg(Tp1:EGFP);Tg(myl7:dsRed) adult zebrafish with homozygous Tg(fli1:tdTomato; cmlc2:GFP) fish by group spawning methods. Non-transgenic embryos (for use in setting FACS gates) were obtained by inbreeding WT (Tubingen) fish. Embryos from individual spawning groups were pooled into single clutch biological replicates. At 26 hpf, embryos were manually sorted for Tp1:EGFP expression. At 28-32 hpf, embryos were prepared for FACS sorting as previously described 19 .
Briefly, embryos were manually dechorinated, yolk removed, then dissociated into single-cell suspension using a combination of TrypLE (Gibco) and FACSMax (Gentlantis). Cells were re-suspended in ice-cold L-15 (Gibco) supplemented with 5% heat-inactivated FBS (Hyclone), passed through 40 and 35 mm filters, then counterstained with Live/Dead fixable near-IR stain (Molecular Probes). A SH800 cell sorter (Sony) was used to sort live, tdTomato þ cells by Tp1:EGFP expression level and harvested directly into RTL lysis buffer, then stored at À 20 for later RNA purification.
Imaging and quantification. All images were acquired on an Olympus FluoView FV1200 confocal microscope, equipped with Green, Red and Far-Red lasers and detectors. Olympus OIB file formats were imported into FIJI using Bio-Formats Importer 20 for analysis and quantification, as described below.
Branch frequency was measured by skeletonizing ES-cell-derived vessels or HUVEC-derived 3D sprouting angiogenesis assay vessels and calculating branches per mm using Image J (NIH). For branching parameters, the sprout length was measured from the base of the spout to the most distal end. The maximum branch angle was calculated at sprout junctions, excluding branches that fused with other sprouts distal to the junction. The percentage of tip cells was quantified in all multinucleated sprouts. Tip cells were defined as the cell nucleus located most distal in a sprout. In zebrafish, the percentage of somites containing angiogenic sprouts was calculated. The first 12 segments from the end of yolk extension were analysed as described 9 . pSMAD1/5 fluorescence intensities were determined on a single-cell basis in 2D using FIJI as follows: HUVEC were stained for pSMAD1/5 (also FLAG if relevant) and incubated with DRAQ7 as described in detail above. The brightest slice in the DRAQ7 (nuclear) channel from confocal z-stack images was threshold adjusted into a binary image (black nuclei, white everywhere else). Next, using the 'Analysis' -'Set Measurements' menu, analysis was re-directed from the binary nuclear image to the pSMAD1/5 channel and 'Analyze' -'Analyze Particles' function was used with area cutoffs of 50-500 mm 2 to return mean grey values per nucleus. pSMAD1/5 fluorescence intensities were determined on a single-cell basis in 3D using FIJI as follows: pSMAD1/5 and DRAQ7 (nuclear) channels were independently compressed along the z axis by summing slices. DRAQ7 channel was threshold adjusted as described above and binary image used to re-direct measurements to the pSMAD1/5 channel as described above. The geodesic distance from the tip nucleus was measured along the length of the vessel from the most distal portion of the tip nucleus. For tip-cell distance:pSMAD1/5 fluorescence correlations, both distance and fluorescence were determined relative to the tip cell (0 mm distance and 100% fluorescence).
SMAD6 fluorescence intensities were determined on a single-cell basis in 2D as follows: HUVEC were stained for SMAD6 and FLAG, and incubated with Phalloidin to visualize cell boundaries as described in detail above. SMAD6 fluorescence intensity was measured by finding the brightest slice in the phalloidin channel, outlining individual cells and calculating integrated density within the SMAD6 channel with subtraction of background.
Statistics. All statistical analyses were performed using Prism v6.05 (www.graphpad.com), with an a of 0.05. Data were tested for normal distribution using the D'Agostino and Pearson omnibus normality test within Prism. For two-sample data sets with equal variances (control -v-a single experimental condition) unpaired, two-tailed Student's t-test was used as reported in figure legends. For data sets with greater than two conditions and equal variances, one-way analysis of variance with Tukey's post-hoc test was used as reported in figure legends. For data sets with greater than two conditions and unequal variances, Kruskal-Wallis with Dunn's post-hoc test was used as reported in figure legends. No a priori sample-size power analyses were performed.
Data availability. Data supporting the findings of this work are available within the article and its Supplementary Information files, and from the corresponding author on reasonable request. ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms13247