Neuronal sFlt1 and Vegfaa determine venous sprouting and spinal cord vascularization

Formation of organ-specific vasculatures requires cross-talk between developing tissue and specialized endothelial cells. Here we show how developing zebrafish spinal cord neurons coordinate vessel growth through balancing of neuron-derived Vegfaa, with neuronal sFlt1 restricting Vegfaa-Kdrl mediated angiogenesis at the neurovascular interface. Neuron-specific loss of flt1 or increased neuronal vegfaa expression promotes angiogenesis and peri-neural tube vascular network formation. Combining loss of neuronal flt1 with gain of vegfaa promotes sprout invasion into the neural tube. On loss of neuronal flt1, ectopic sprouts emanate from veins involving special angiogenic cell behaviours including nuclear positioning and a molecular signature distinct from primary arterial or secondary venous sprouting. Manipulation of arteriovenous identity or Notch signalling established that ectopic sprouting in flt1 mutants requires venous endothelium. Conceptually, our data suggest that spinal cord vascularization proceeds from veins involving two-tiered regulation of neuronal sFlt1 and Vegfaa via a novel sprouting mode.

T he vascular network closely associates with the neuronal network throughout embryonic development, in adulthood and during tissue regeneration [1][2][3] . Close association of vessels and nerves allows reciprocal cross-talk involving diffusible molecules, which is important for physiological functions in both domains 4,5 . Arteries secrete factors that attract sympathetic axons, and adrenergic innervation of arteries allows the autonomic nervous system to control arterial tone and tissue perfusion 5 . The nervous system, on the other hand, requires a specialized network of blood vessels for its development and survival. Metabolically active nerves rely on blood vessels to provide oxygen necessary for sustaining neuronal activity 6 , and disturbances herein result in neuronal dysfunction 1,7 .
How nerves attract blood vessels is debated, but several studies addressing vascularization of the mouse and chicken embryonic nervous system suggest that the angiogenic cytokine VEGF-A is involved [8][9][10] . In the mouse peripheral nervous system axons of sensory nerves innervating the embryonic skin trigger arteriogenesis involving VEGF-A-Neuropilin-1 (NRP1) dependent signalling 11,12 . While these studies provide evidence for the physical proximity and cooperative patterning of the developing nerves and vasculature, relatively little is known about mechanisms controlling VEGF-A dosage at the neurovascular interface. This is of great importance considering that blood vessels are very sensitive to changes in VEGF-A protein dosage and even moderate deviations from its exquisitely controlled physiological levels result in dramatic perturbations of vascular development 13,14 . VEGF-A levels must therefore be well titrated, and several strategies have evolved to achieve this.
Mouse retinal neurons for example can reduce extracellular VEGF-A protein via selective endocytosis of VEGF-A-VEGF receptor-2 (KDR/FLK) complexes. Inactivation of this uptake causes non-productive angiogenesis 15 . In the vascular system, spatio-temporal control of VEGF-A protein dosage is thought to be achieved by soluble VEGF receptor-1 (sFLT1), an alternatively spliced, secreted isoform of the cell-surface receptor membranebound FLT1 (mFLT1) 16,17 . Soluble FLT1 binds VEGF-A with substantially higher affinity than KDR, thereby reducing VEGF-A bioavailability and attenuating KDR signalling 17 . While originally discovered as a vascular-specific receptor, evidence is emerging showing neuronal FLT1 expression 18 . To what extent endogenous neuronal Flt1 has a physiological role in titrating neuronal VEGF levels controlling angiogenesis at the neurovascular interface independent of vascular Flt1 remains to be determined.
Angiogenesis involves complex and dynamic changes in endothelial cell behaviour 19 . In the zebrafish embryo these events can be studied in detail at the single cell level in vivo through the use of vascular-specific reporter lines 20,21 . The stereotyped patterning of arteries and veins in the trunk of the zebrafish embryo prior to 48 hpf is mediated by cues derived from developing somites and the hypochord, controlling angiogenic sprout differentiation and guidance 22,23 . Sprouting of intersegmental arterioles (aISV) requires Vegfaa-Kdrl signalling, as loss of either kdrl or vegfaa completely abolishes ISV sprouting from the dorsal aorta (DA) 24 . Primary sprouting also involves a component regulated by Notch, as loss of Notch increases the endothelial propensity to occupy the tip cell position in this vessel, whereas gain of Notch restricts aISV development 25 . Secondary vein sprouting requires Vegfc-Flt4 signalling, as loss of either ligand or receptor blocks venous growth 26,27 . Developing somites are regarded as the main source for Vegfaa, while the hypochord provides Vegfc during early development 22,23 .
In this study we show that developing spinal cord neurons located in the trunk of the zebrafish embryo produce Vegfaa and sFlt1 affecting the angiogenic behaviour of intersegmental vessels at the neurovascular interface. We find that during early development neuronal sFlt1 restricts angiogenesis around the spinal cord. We demonstrate that on genetic ablation of neuronal sFlt1 this brake is relieved resulting in the formation of a vascular network supplying the spinal cord in a Vegfaa-Kdrl dependent manner. Using inducible neuron-specific vegfaa gain-of-function approaches and analysis of several mutants with vegfaa gain-of-function scenarios, we furthermore show that the neuronal Vegfaa dosage determines the extent of the neovasculature supplying the spinal cord, as well as sprout invasion into the spinal cord. Interestingly, loss of flt1 or augmenting neuronal vegfaa promotes sprouting from intersegmental veins involving distinctive angiogenic cell behaviours including nuclear positioning and a molecular signature not observed in primary arterial or secondary venous sprouting. Cell transplantation experiments confirm the role of neuronal flt1 in venous sprouting and furthermore show that vascular flt1 is dispensable herein. Taken together, our data suggest that spinal cord vascularization proceeds from veins and is coordinated by two-tiered regulation of neuronal sFlt1 and Vegfaa determining the onset and the extent of the vascular network that supplies the spinal cord via a novel sprouting mode.

Results
Spinal cord neurons express sflt1, mflt1 and vegf ligands. Analysis of TgBAC(flt1:YFP) hu4624 ;Tg(kdrl:hsa.HRAS-mCherry) s916 transgenic embryos showed flt1 expression in the aorta, arterial intersegmental vessels (aISVs), dorsal part of venous intersegmental vessels (vISVs) and spinal cord neurons located in the neural tube ( Fig. 1a,b,d-g) 18 . Spinal cord neurons were in close proximity to blood vessels ( Fig. 1c-e) and 3D-rendering of confocal z-stacks obtained from Tg(kdrl:EGFP) s843 ;Tg(Xla.Tubb:DsRed) zf148 double transgenic embryos showed the dorsal aspect of ISVs 'indenting' the neural tube indicative of close contact ( Fig. 1c; Supplementary Movie 1). Optical sections confirmed close contact between the outer neuronal layers of the neural tube and the dorsal part of ISVs, as well as the dorsal longitudinal anastomotic vessel (DLAV) (Fig. 1d-g). Such anatomical juxtapositioning of trunk vessels and neurons may provide a template for molecular cross-talk (Fig. 1d,e; pink box).
TaqMan analysis using FAC-sorted neuronal cells from two different neuronal reporter lines (Supplementary Fig. 1a-k) showed expression of mflt1, sflt1, kdrl, kdr, flt4 and the ligands vegfaa, vegfab, and plgf ( Supplementary Fig. 1b,e) 24 . Flt1 was expressed in a comparable range as neuronal guidance molecules ( Supplementary Fig. 1c,f). Real-time qPCR analysis for vegfaa and vegfab in the trunk of developing zebrafish embryos confirmed expression of both isoforms ( Supplementary Fig. 1l,m).
Zebrafish homozygous for the flt1 -1 nt allele (flt1 ka601 ) displayed severe hyper-branching of the trunk vasculature at 3-4 dpf (Fig. 2a,b). Supernumerous amounts of branches developed in the dorsal aspect of the trunk at the level of the neural tube (Fig. 2b,c). Comparable observations were made in embryos homozygous for the flt1 À 5 nt allele (flt1 ka602 ) and the flt1 þ 5 nt allele (flt1 ka603 ) (Fig. 2d,e,f). Analysis of four mflt1 mutant alleles (flt1 ka605-608 , Supplementary Fig. 2e) did not reveal any obvious vascular malformations or alterations in vascular branching morphogenesis (Fig. 2g). These observations are compatible with absence of angiogenic defects in mouse Flt1 TK À / À embryos lacking mFlt1 signalling 28,29 . The vascular phenotype observed in the flt1 ka601 mutants thus most likely involved soluble Flt1.
No significant changes in heart frequency were noted ( Supplementary Fig. 4a). The vascular phenotype of flt1 ka601 mutants emerged around day 2.5 ( Fig. 2q,r) with sprouts emanating exclusively from the dorsal aspect of the venous ISVs at the level of the neural tube (Supplementary Movie 2); ectopic arterial ISV sprouting was not observed (Fig. 2r). In flt1 ka601/ þ heterozygotes ( Fig. 2l-o) ectopic sprouting was rarely observed (Fig. 2n,o; Supplementary Movie 3). In wild-type (WT) embryos such endothelial cell behaviours were not observed ( Fig. 2h-k, Supplementary Movie 4).
We furthermore examined whether flt1 targeting morpholino could recapitulate the flt1 ka601 mutant phenotype ( Supplementary  Fig. 4b-h). We evaluated two dosages of a published flt1 ATG targeting morpholino (MO) and found that 1 ng flt1 MO induced hyper-branching in WT at levels comparable to flt1 ka601 (Supplementary Fig. 4g,h) 18,30 . Injection of 1 ng MO into flt1 ka601 mutant background did not induce additional sprouting defects ( Supplementary Fig. 4d) additional branches at 2 dpf that were not observed in the flt1 ka601 mutant at this stage ( Supplementary Fig. 4e). Since we did not observe maternal contribution of flt1 these observations suggest that 3 ng MO introduced non-specific effects 31 .
Sprouts in flt1 ka601 display distinctive cell behaviours.
Compatible with ectopic sprouting we identified hyperactive endothelial cells extending filopodia in the dorsal aspect of vISVs of flt1 ka601 mutants ( Fig. 3a; Supplementary Movie 2). About 55% of hyperactive endothelial cells investigated generated a patent sprout (Fig. 3a); in the remaining 45%, filopodia and sprouts retracted (Fig. 3b). From the population of patent ectopic venous sprouts 95% formed an anastomosis with an aISV, whereas only 5% made a connection with a vISV (Fig. 3c).
The preference for arterial anastomosis may be physiologically     relevant as it creates a pressure gradient promoting blood flow perfusion. Sprout filopodia length ranged from 1 to 20 mm, and filopodia projected at an angle between 90 and 120°with respect to the vISV compatible with arterial anastomosis formation ( Fig. 3d-f). Current models posit that Flt1 produced in angiogenic sprouts mainly prevents back-branching of nascent sprouts 32 . We find that in the absence of Flt1 sprouts retain their directionality and migrate away from the parent vessel. Within ISVs endothelial nuclei migrated at velocities of up to 1 mm min À 1 (Fig. 3g-j,l). Careful analysis of nuclear positioning within endothelial cells revealed an association between nuclear position and sprout initiation ( Fig. 3k-m, Supplementary Movie 2). Nuclei migrated actively into the direction of future sprout initiation points (SIP), and in more than 80% of the studied sprout initiations nuclear positioning was directly linked with sprout initiation (linkage was defined as nucleus-SIP distance of o5 mm at spout initiation) (Fig. 3l,m). This nuclear migration behaviour is in contrast to rearward nuclear positioning in migrating angiogenic endothelial cells in vitro 33 and is not described in vivo for primary artery or secondary venous sprouting events in zebrafish. Analysis of endothelial cell numbers at 4 dpf showed increased endothelial cell numbers in aISVs, vISVs, DLAV, DA and PCV of flt1 ka601 compared with WT ( Fig. 3n-q); DLAV size was not statistically different ( Fig. 3r). At earlier stages (17 hpf) we found no differences in endothelial cell numbers between flt1 ka601 and WT (Supplementary Fig. 4i-l).
flt1 ka601 display upregulation of angiogenic sprout markers. We next performed RNA sequencing of flt1 ka601 mutants and analysed genes implicated in sprouting angiogenesis (Supplementary Fig. 5a-c). Expression of the classical tip-stalk cell markers including notch1a, notch1b, dll4, nrarpa, nrapb, hey1, hey2, her6 and flt4 were not altered 34 ( Supplementary Fig. 5b). This result may not be surprising since ectopic venous sprouts emanated from venous ISVs, and Dll4-Notch signalling is absent in this domain 35 . Instead we found upregulated expression of other genes implied in sprouting cell behaviour. RNA-seq and qPCR of flt1 mutants showed significantly increased levels of apelin receptor-a (aplnra), angiopoietin-2a (angpt2a), and endothelial cell specific molecule-1 (esm1) ( Supplementary  Fig. 5b,c), genes previously shown to be enriched in angiogenic vessels 36,37 . In addition, we observed a significant upregulation of plgf, which encodes the Flt1-specific pro-angiogenic ligand Plgf, and lyve1, a gene expressed in veins and implied in lymphangiogenesis, in line with the venous expansion phenotype in flt1 ka601 mutants ( Supplementary Fig. 5c) 38 .

Origin of endothelial cells in ectopic venous sprouts.
It is established that artery-derived ECs, on arteriovenous (AV) remodelling, contribute to the dorsal aspect of vISVs (ref. 39). Besides these remodelled artery-derived cells, another source may be PCV-derived venous endothelial cells as they can migrate over long distances 40  artery-derived cells in green). Scenario (I), which we refer to as 'mixed' (both artery and vein-derived EC), accounted for 43.2% of cases (Fig. 4h,i). Of the mixed population 67.9% of endothelial cells were of venous origin and 32.1% of arterial origin (Fig. 4h, right panel). In scenario (II), the dorsal part of vISV only contained PCV-derived venous endothelial cells; artery-derived endothelial cells were absent. Scenario (II) accounted for 48.6% of cases ( Fig. 4h,i). In scenario (III) we find that the dorsal part of vISV only contained artery-derived ECs; in this scenario the dorsal aspect of vISVs was not colonized by migrating PCV-derived venous endothelial cells (Fig. 4h,i). This scenario was observed in 8.2% of cases. The flt1 enh promoter marks ISV-ECs of arterial origin 39 . Loss of flt1 in Tg(flt1 enh :Tdtomato; flt4:mCitrine) showed ectopic venous sprouts containing flt1 enh -expressing ECs (Fig. 4k,l,n-p). In the same embryo, we furthermore noted ectopic venous sprouts devoid of flt1 enh expressing ECs (Fig. 4j,m,p), suggesting that these sprouts were only made of vein-derived ECs (Fig. 4m).
To confirm a contribution of PCV-derived venous endothelium we performed cell tracking experiments in Tg(kdrl:nlskikGR) hsc7 on loss of flt1 and indeed we found that PCV-derived venous ECs were capable of contributing to ectopic sprouting (Supplementary Fig. 6a). Interestingly, besides sprouts exclusively containing artery-derived, or venous-derived endothelium ( Fig. 4m,o), we observed composite sprouts with artery and venous-derived endothelial cells juxtapositioned (Fig. 4k,n).
Mechanistically, loss of vhl and flt1 augments Vegfaa function at different levels, through increased vegfaa transcription and higher Vegfaa protein bioavailability, respectively. We reasoned that combining both mutants should increase Vegfaa   ARTICLE and activate downstream Kdrl signalling even further. Indeed, flt1 ka601 ;vhl hu2114 double mutants showed more severe hyper-branching of the trunk vasculature when compared with single mutants (Fig. 5e,h). Accordingly, flt1 ka601 ;vhl hu2114 double mutants developed more ectopic venous sprouts when compared with either single flt1 ka601 or single vhl hu2114 mutants (Fig. 5e). The flt1 ka601 ;vhl hu2114 double mutants also developed a small number of ectopic arterial sprouts after 2.5 dpf (Fig. 5e).
Ectopic sprouting in flt1 ka601 mutants requires veins. To prove that in flt1 ka601 mutant sprouts indeed emanated from veins, we interfered with early arterial-venous remodelling by blocking flt4 (ref. 26) (Fig. 5f,g; Supplementary Fig. 6l). Loss of flt4 in flt1 ka601 mutants interfered with arterial-venous remodelling; as a consequence almost all trunk ISVs remained arterial 26 ( Supplementary Fig. 6l). In line with the requirement for veins, the flt1 ka601 hyper-branching phenotype was rescued (Fig. 5f,g). Furthermore, flt4 loss-of-function in the flt1 ka601 ;vhl hu2114 double mutants (denoted as double in Fig. 5h,i) also significantly reduced branching complexity (Fig. 5h-j; method quantification of branch points in Supplementary Fig. 1n). As ectopic sprouting requires venous endothelium, we next reasoned that promoting vISV formation in flt1 ka601 mutants should augment branching. Vessel identity and Notch signalling are linked. In zebrafish, it is established that loss of the Notch ligand Dll4 promotes venous cell fate and dll4 loss-of-function embryos display a trunk vasculature consisting almost exclusively of vISVs 45 . Accordingly, loss of dll4 in flt1 ka601 mutants significantly augmented ectopic branching when compared with control flt1 ka601 mutants (Fig. 5k,l).
Notch, pericytes and ectopic venous sprouting in flt1 ka601 . One explanation for the low arterial responsiveness in vegfaa gain-of-function scenarios may involve high arterial Notch activity since Notch acts as a repressor of sprouting in arteries, downstream of Vegfaa signalling 25,35,46 . To inhibit endothelial Notch signalling in arterial ISVs of flt1 ka601 mutants, we expressed a dominant negative form of the Notch co-activator MAML (DN-MAML-EGFP) in an endoxifen inducible manner (Fig. 6a,b) 47 . We used the flt1 enh promoter construct which is mainly active in aISVs (ref. 39) to drive gal4-ERT2;UAS:DN-MAML-eGFP (notch iDEC ) in flt1 ka601 mutants. Transgene expression was initiated at 52 hpf by adding endoxifen. Endothelial-specific DN-MAML gain-offunction in flt1 ka601 mutants induced ectopic aISV sprouting at the level of the neural tube (Fig. 6a,b,f). Even more pronounced ectopic arterial sprouting was observed with the g-secretase inhibitor LY-411575 that blocks Notch activation; adding LY-411575 at 2 dpf activated ectopic arterial sprouting in flt1 ka601 mutants (Fig. 6c-f). Venous sprout numbers were not significantly changed upon DN-MAML (16.1±3.45 versus 17.1 ± 2.88) or LY-411575 treatment (15.9 ± 2.89 versus 14.2±1.69). Addition of LY-411575 to WT at 2 dpf had no effect. To explain differential AV responsiveness, we also considered differences in pericyte cell coverage (Fig. 6g-j). Overall, pericytes were scarce with 88% of all ISVs investigated not being covered by pericytes. In the remaining 12% of cases, pericytes were found in both aISVs (9.94%) and vISVs (1.91%) along the ISV ventral-dorsal axis. In the most dorsal aspect of aISV and vISV, the region where ectopic sprouting occurs in flt1 ka601 , pericytes were comparable between aISV and vISV (2.48% and 1.91% respectively, Fig. 6k).
Vegf and Flt1 determine extent of spinal cord vascularization. Neurons expressed vegfaa ( Supplementary Fig. 1), and neuronal cells of both 3 dpf WT and vhl loss-of-function embryos had significantly higher vegfaa levels than non-neuronal cells (Fig. 7a,b; FACS settings in Supplementary Fig. 7a-d). Furthermore, neuronal vegfaa expression was significantly increased in vhl loss-of-function when compared with WT (Fig. 7a,b). Thus, at this stage of development neurons are the major source of vegfaa, and not other tissues like developing muscle 48 . We next examined whether neurons can direct sprouts into the neural tube (Fig. 7c-i). We compared the flt1 ka601 ;vhl hu2114 double mutant (Fig. 7c,d) with flt1 ka601 mutant and WT and found striking changes in optical sections of the neurovascular interface (Fig. 7e-h). In flt1 ka601 , sprouts occasionally projected into the neural tube (Fig. 7g), whereas in flt1 ka601 ;vhl hu2114 double mutants many branches invaded the neural tube (Fig. 7h,i).
In the mutants with vegfaa gain-of-function, the spinal cord becomes vascularized relatively early, between 3 and 4 dpf. In WT, the spinal cord is vascularized much later in development starting in the period between 12 and 14 dpf (Fig. 7j-l). In those older WT embryos, sprouts preferentially emanated from venous ISVs, displayed nuclear positioning as described for the flt1 mutant ( Supplementary Fig. 7e,f) and the onset of vascularization of the WT spinal cord coincides with decreased sflt1 expression during this stage of development ( Supplementary  Fig. 7g).
Neuronal sFlt1 and Vegfaa regulate sprouting from veins. We next generated tissue-specific and inducible flt1 and vegfaa gainof-function models. Loss of neuronal sFlt1 in flt1 ka601 mutants may augment neuron-derived Vegfaa availability and promote ISV sprouting. Hence, restoring neuronal sFlt1 in flt1 ka601 mutants should provide a rescue, whereas neuronal-specific flt1 loss-of-function should induce hypersprouting. To test the first scenario we expressed -3.2elavl3:gal4-ERT2;UAS:GFP-p2A-sflt1 (sflt1 iNC ) in flt1 ka601 mutants (Fig. 8a-d, branch quantification method in Supplementary Fig. 1n). This construct allows precise time-controlled expression of sflt1 specifically in neurons. We found that transgene activation in neurons at 52 hpf, just before the emergence of the ectopic sprouts in flt1 ka601 mutants, rescued the vascular hyper-branching phenotype (Fig. 8b-d).
We next explored whether neuron-specific loss of flt1 is sufficient to induce ISV hyper-branching (Fig. 8e-i).
To substantiate the contribution of neuronal sflt1 we next employed multiplexed custom designed miRNAs directed against sflt1 3 0 UTR arranged with a common miR-155 backbone 50 ( Supplementary Fig. 8b). The constructs were expressed under control of vascular (flt1 enh ) and neuronal (Xla.Tubb) specific promoters. Targeting neuronal sflt1 resulted in ectopic sprouting ( Supplementary Fig. 8c), but targeting vascular sflt1 failed to induce sprouts (Supplementary Fig. 8d).
Next we performed cell transplantation experiments, which demonstrated that neuronal flt1 and not vascular flt1 is the physiologically relevant mediator of sprouting at the level of the neural tube (Fig. 8j-l). Transplantation of flt1 mutant neurons into WT hosts induced ectopic sprouting (Fig. 8k).
In contrast, transplantation of flt1 mutant endothelial cells into WT hosts failed to induce sprouting (Fig. 8l).
To prove that neuron-derived Vegfaa promotes hyper-branching, we generated neuronal tissue-specific and inducible vegfaa165 gain-of-function zebrafish ( Supplementary Fig. 8e,f; quantification in Fig. 5e). Transgenic expression was initiated by adding endoxifen after completion of AV remodelling at 52 hpf. In this scenario hyper-branched neovascular networks formed at the level of the neural tube, similar to flt1 ka601 mutants ( Supplementary Fig. 8e,f). Neuronal vegfaa121 was also capable of inducing sprouting (Supplementary Fig. 8g). In contrast, neuron-specific and inducible vegfc gain-of-function, induced at 54 hpf, did not induce ectopic sprouts (Supplementary Fig. 8h). Timing of transgene expression was relevant as inducible neuronspecific vegfaa165 overexpression prior to completion of AV remodelling resulted in thickened abnormal vascular structures ( Supplementary Fig. 8i,j). In the same line, neuronspecific constitutive overexpression of sflt1 completely annihilated ISV formation (Supplementary Fig. 8k).
To confirm that the flt1 ka601 phenotype involved gain of Vegfaa, we titrated vegfaa levels using a low dose vegfaa targeting morpholino 51 . Reducing vegfaa in flt1 ka601 mutants rescued the hyper-branching phenotype (Fig. 8m-o). Vegfaa signals via Kdrl and application of ki8751, an established Kdrl tyrosine kinase inhibitor in zebrafish 52 to flt1 ka601 mutants at 2.5 dpf annihilated the formation of the ectopic neovascular networks (Supplementary Fig. 8l,m,o). In contrast, the Flt4-specific tyrosine kinase inhibitor MAZ51 (ref. 52) did not rescue hyper-branching in flt1 ka601 ( Supplementary  Fig. 8l,n,o). Vegfaa-driven primary artery sprouting can occur in the absence of blood flow perfusion. To test if Vegfaadriven ectopic venous sprouting in flt1 ka601 mutants is affected by blood flow, we modulated cardiac activity and flow with BDM or tricaine. We found that loss of flow completely rescued ectopic hyper-sprouting in flt1 ka601 mutants ( Supplementary  Fig. 9a-e). Inhibition of PI3 kinase with wortmannin also significantly reduced ectopic sprouting ( Supplementary  Fig. 9f).

Discussion
Intimate cross-talk between vessels and the nervous system is important for tissue homeostasis. During embryonic development, neuronal stem cells differentiate into mature neurons, a process that associates with a change in cellular metabolism 53 . Concomitantly with developmental neurogenesis, changes occur in the vascular network feeding the spinal cord. We show in the zebrafish embryo that neurons in the developing spinal cord express the pro-angiogenic ligand Vegfaa and antiangiogenic soluble Vegf receptor-1, sFlt1, which acts as a Vegfaa scavenger (Fig. 9). Spinal cord neurons are in close contact to the developing trunk vasculature, and we show that these vessels are responsive to changes in neuronal sFlt1 and Vegfaa. Using a combination of global and tissue-specific loss of flt1 mutants, and further substantiated by vegfaa loss-and gain-of-function experiments as well as cell transplantations, we demonstrate that neuronal sFlt1 restricts neuronal Vegfaa and vessel branching morphogenesis at the neurovascular interface. Differential regulation of vegfaa and sflt1 allows orchestration of the onset and extent of spinal cord vascularization (Fig. 9). We propose that neurons may use sFlt1-Vegfaa to adjust vascularization according to their developmental needs.
Flt1 ka601 mutants develop ectopic sprouts emanating from venous ISVs around embryonic day 2.5. Neuron-specific targeting of flt1 or sflt1 using CRISPR/Cas9-and miRNA-based approaches respectively, result in flt1 ka601 comparable phenotypes. Transplantation of flt1 mutant neurons into WT hosts induces ectopic sprouting which is not observed after transplantation of flt1 mutant endothelial cells, suggesting that neuronal flt1 is the physiologically relevant mediator in our mutant. Neuronal-specific gain of sflt1, reducing vegfaa levels, or inhibition of Kdrl signalling provide a rescue suggesting that ectopic venous sprouting is mediated by the Vegfaa-Kdrl signalling axis. Accordingly, ectopic venous sprouting from the dorsal aspect of vISVs is conserved in five independent vegfaa gain-of-function scenarios.
Previous studies have shown that during AV remodelling, aISV-derived endothelial cells remain integrated in the dorsal aspect of vISVs (ref. 39). We confirm that remodelled artery-derived endothelial cells indeed contribute to this domain although they are not the sole or most important endothelial source. Using in vivo cell tracking we find posterior cardinal vein-derived endothelial cells migrating against the direction of blood flow to populate venous ISVs including the dorsal aspect where sprouts are formed. Here, venous-derived endothelium can co-exist with the artery-derived endothelium. With respect to the endothelial cells populating the dorsal part of vISVs, our data now reveal three different scenarios. The dorsal aspect can contain a mix of both artery and venous-derived endothelium (43%), only vein-derived endothelium (48%) or only artery-derived endothelium (8%). On loss of flt1, both artery-and vein-derived endothelium give rise to ectopic vISV sprouts. This prompts toward the concept that integration into the local venous ISV environment constitutes a permissive factor for sprouting, regardless of the endothelial origin.
Flt1 ka601 mutants display ectopic sprouting in vISVs but not in aISVs, indicating that AV vessel identity or compartmentspecific cues may be involved in the novel sprouting type described here. Notch is tightly linked to both AV vessel specification and sprouting, as Notch programs arterial identity and Notch signalling represses sprouting of arteries 25,45 . Lack of Notch associates with venous identity and Notch signalling is absent in venous ISVs (ref. 35). Interference with AV remodelling or Notch indeed affects sprouting numbers in flt1 ka601 . Inhibiting the remodelling of arterial ISVs into venous ISVs in flt1 ka601 , and thus creating a trunk vasculature that almost exclusively consists of aISVs, rescues ectopic hypersprouting. Conversely, promoting venous cell fate and creating a trunk that consists of vISVs by knock-down of dll4 augments vascular branching in flt1 ka601 . Inhibiting Notch by endothelial expression of DN-MAML or pharmacological treatment with LY-411575 stimulate ectopic arterial sprouting in flt1 ka601 , without affecting venous sprout numbers. Thus active Notch signalling in arteries most likely accounts for the observed AV sprouting differences. We propose that the artery-derived endothelial cells that become incorporated in vISVs on AV remodelling are relieved from Notch and adopt a venous fate, enabling them to respond to changes in local Vegfaa.
Our data show that vISVs contribute to the vascularization of the spinal cord with Vegfaa-Kdrl signalling mediating vISV branching events. Interestingly, in other domains of the zebrafish trunk venous vasculature, endothelial cells are responsive to alternative signals. The Vegfc-Flt4 signalling pathway drives secondary venous sprouting from the PCV contributing to the formation of the lymphatic vasculature. BMP-Smad signalling has been shown to promote venous sprouting in the caudal vein plexus 54 . Furthermore, PCV-derived endothelial cells contribute to the formation of the gut vasculature 55 . Regeneration of the tail fin vasculature upon injury also starts from the venous side 56 . The concept emerging from these observations is that initiation of organ vascularization is initiated or proceeds from the venous vasculature. The mouse embryo coronary vasculature is vein-derived, and coronary arteries form by developmental reprogramming of venous endothelium 57 . Context-dependent heterogeneity in signalling mechanisms establishing venous branching, may allow versatile control of tissue vascularization in a spatio-temporal manner. Using sFlt1 as a rheostat to control Vegfaa bioavailability, constitutes a means to regulate Vegfaa independent of vegfaa promoter activity, vegfaa mRNA or protein stability. We propose that this enables neurons to dynamically fine-tune the extent and onset of peri-neural vascular network formation and sprouting into the spinal cord. While the peri-neural network may serve to sustain growth of the developing nervous system, vessel sprouting into the spinal cord and relief of hypoxia has been associated with changes in neuronal stem cell metabolism, triggering differentiation events 53 . Therefore, untimely or excessive vascularization of the spinal cord is potentially harmful as it may promote premature stem cell differentiation and disrupt the carefully orchestrated neuronal specification process. We propose a two-tiered checkpoint mechanism involving sFlt1 and Vegfaa, requiring two decisions to guide vascularization, namely Vegfaa up-and sFlt1 downregulation, to protect neurons from harmful angiogenesis and oxygen variations during early stages on the one hand, and on the other hand to enable more mature neurons to attract sufficient vessels into the spinal cord after stem cell differentiation has been completed.

Methods
Ethics statement. Zebrafish husbandry and experimental procedures were performed in accordance with the German animal protection standards and were approved by the government of Baden-Württemberg, Regierungspräsidium Karlsruhe, Germany (Akz.: .81/G-93/15).
mRNA injection and generation of transgenic and mutant lines. For the generation of mutants 1 nl of a mixture containing 600 ng ml À 1 capped and polyadenylated Cas9-nls mRNA and 50 ng ml À 1 sgRNA was injected into one-cell stage embryos 63 . Cas9 mRNA was produced by in vitro transcription from the MLM3613 plasmid using the mMessage mMachine T7 Ultra Kit (Ambion). The MLM3613 plasmid was a gift from Keith Joung (Addgene plasmid #42251). For the generation of transgenic lines 1 nl of a mixture of 12.5 ng ml À 1 transposase mRNA and 25 ng ml À 1 plasmid DNA was injected into one-cell stage embryos.
Cell transplantations. Cell transplantations were performed using 3.5 hpf donor and host blastula-stage embryos. Approximately 50-100 cells were taken from the donor's animal pole and transferred close to the host's lateral marginal zone (for ECs) or slightly above for neuronal cells. Donors and hosts carried distinct neuronal and endothelial-specific reporters to identify the source of ECs and neurons within chimeras.
Generation of mflt1-specific mutants. To generate mflt1 mutants we used a CRISPR/Cas approach and designed an sgRNA targeting E11b, the first specific mflt1 exon 18 . In this scenario splicing of intron 10 and exon 11a relevant for generating sflt1 mRNA remains unaffected. Oligos Flt1E11_O1_A_15 and Flt1E11_O2_A_15 were annealed and cloned into DR274 as described for flt1 mutants. Founders were identified by PCR and subsequent Sanger sequencing, using primers Flt1E11A2386576F and Flt1E11A2386151R. We identified four frame shift mutants harbouring a PTC in exon 11b. Flt1 ka605 (exon 11b þ 28 nt), flt1 ka606 (exon 11b þ 20 nt), flt1 ka607 (exon 11b À 1 nt) and flt1 ka608 (exon 11b À 1 nt and one mutation) mflt1 mutants were outcrossed to Tg(kdrl:EGFP) s843 and Tg(Xla.Tubb:DsRed) zf148 . All four mflt1 mutants were phenotypically comparable and in this manuscript only the mflt1 mutant flt1 ka605 is shown. All sgRNA sequences and oligos used for annealed oligo cloning into DR274 are listed in Supplementary Table 4,5.
Generation of p5E entry clones. The NBT_tauGFP plasmid was a kind gift by Enrique Amaya. The 3.8 kb regulatory element derived from neural specific beta tubulin was removed from the NBT_tauGFP using SalI and HindIII and subcloned into SalI and HindIII digested and dephosphorylated p5E_MCS (ref. 64). The 1 kb flt1 enhancer/promoter fragment from the pMiniTol2_flt1_ECR5a_pro_181_YFP (ref. 39) construct was subcloned into p5E_MCS using KpnI and HindIII. The resulting plasmids were named p5E_ Xla.Tubb-3.8 and p5E_flt1 enh .
Generation of a universal p2A-GFP middle entry clone. To easily detect transgenic cells the pME_eGFP (#455) from the Tol2kit (ref. 64) was modified by site-directed mutagenesis PCR. The p2A sequence was added before the stop codon of GFP using pME_eGFP specific primer with 5 0 end extension coding for the p2A peptide and a SmaI restriction site just downstream of p2A for convenient subcloning (pME_eGFP_p2A_fw and pME_eGFP_p2A_rev primer). pME entry clones used for gain-of-function experiments. pME_eGFP-p2A_SmaI was digested with SmaI and XhoI. The inserts vegfaa165, vegfC and sflt1 were amplified from zebrafish cDNA using primers vegfaa_p2A_fw/rev, vegfc_-p2A_fw/rev and sflt1_p2A_fw/rev. The PCR products were digested with XhoI and gel purified. Vector and inserts were ligated following the manufactures instructions (NEB T4 DNA Ligase). The resulting plasmids were named pME_eGFP-p2A_vegfaa165, pME_eGFP-p2A_vegfc and pME_eGFP-p2A_sflt1.
gal4ERt2 middle entry clone generation. To spatially and temporally regulate transgene expression an inducible gal4-ERT2 fusion protein was constructed. The Gal4 DNA binding domain was fused at its C-terminus with a mutant oestrogen ligand-binding domain ERT2 that carries a VP16-derived non-deleterious transactivation domain TA4 (ta4, 39 aa) at its C-terminus 65 . Among all possible sequential orders of domains, this arrangement was inferred to have a low background with a high induction rate 66 . A middle entry clone pENTR/D-creERT2 was modified by replacing Cre recombinase domain (1,053 bp, flanked by NotI and XhoI sites at its 5 0 and 3 0 termini, respectively) with a PCR product encoding Gal4 DNA binding domain (1-146 aa) with Kozak sequence in the 5 0 vicinity of the start codon. To replace the stop codon at the 3 0 terminus of ERT2 domain with TA4 domain, a C-terminal half of a ERT2 domain (115-316 aa, flanked by in-frame NcoI at its 5 0 terminus) was replaced with a PCR product encoding the C-terminal half of the ERT2 domain without a stop codon (115-315 aa) with in-frame AgeI site at its 3 0 terminus. The AgeI site, and 3 0 downstream EcoRI site were utilized to insert two synthetic double-stranded oligonucleotides encoding the TA4 domain and the stop codon.
FACS. Approximately 500 embryos Tg(mnx1:GFP) ml2 , Tg(HuC:EGFP) as8 , Tg(Xla.Tubb:DsRed) zf148 or vhl MO injected Tg(Xla.Tubb:DsRed) zf148 embryos were dechorionated at 24 hpf using pronase (0.5 mg/ml). Cells were dissociated using FACSMax as recommended by the manufacturer. Tg(mnx1:GFP) ml2 , Tg(HuC:EGFP) as8 embryos were dissociated and sorted at 24 hpf, control and vhl MO injected Tg(Xla.Tubb:DsRed) zf148 embryos were dissociated and sorted at 3 dpf. Dissociated cells were FACS sorted using BD-FACS-Aria I and Aria II. The sorted cells (B0.5 Â 10 6 cells per experiment) were spun down at 310 g for 5 min and resuspended in lysis buffer contained in the RNeasy mini kit (Qiagen). RNA was extracted as described in the manual. Because of limited amounts of RNA the QuantiTect Whole Transcriptome Kit (Qiagen) was used to preamplify and reverse transcribe the RNA to make cDNA. cDNA was diluted 1:250 for real-time qPCR.
Gene expression analysis by real-time qPCR and TaqMan. Total RNA of zebrafish embryos was isolated with TRIzol, purified with RNeasy mini kit (Qiagen) and quantity and quality were measured using an Agilent 2,100 Bioanalyzer (Agilent Technologies) according to the manufacturer's instructions. We performed DNase on-column digestion using RNase-free DNase Set (Qiagen) according to the manufacturer, followed by cDNA synthesis using the Thermoscript First-Strand Synthesis System (Thermo Fisher Scientific). Primer probe sets (FAM and TAMRA labels) were obtained from Thermo Fisher Scientific. Amplification was carried out using an ABI Prism 7,000 thermocycler (Applied Biosystems). qPCR was conducted with SYBR Green PCR Master Mix (Thermo Scientific) in a StepOnePlus real-time qPCR system (Applied Biosystems). Primers for real-time qPCR were ordered from Eurofins Genomics. Gene expression data were normalized against zebrafish elongation factor 1-alpha. Primers and probes are listed in Supplementary Table 1-3. RNA-seq library preparation and sequencing. Zebrafish RNA was isolated and purified from 4 dpf zebrafish larvae using TRIzol and RNeasy mini kit (Qiagen) as recommended by the manufacturers. A cDNA library was generated using the TruSeq Ilumina RNA sample prepv2 kit according to the manufacturer's protocol. The cDNA library was sequenced on a HiSeq2000 according to the manufacturer's protocols (Illumina).
Identification of differentially expressed genes. Raw sequencing reads were mapped to the transcriptome and the zebrafish reference genome (GRCz10 danRer10) using Bowtie2.0 and TopHat 2.0 (ref. 67). On average 44,490,573 reads (81,6% of total reads) were assigned to genes with Cufflinks and HTSeq software package. Differentially expressed genes (control vs. mutant) were identified using DESeq and Cuffdiff 67,68 . Genes were defined as differentially expressed if Z2 fold significantly regulated (Po0.05) with two independent methods (DEseq and Cuffdiff).
Zebrafish histological sectioning. Dechorionated larvae were fixed in 4%PFA for 2 h and subsequently transferred to 20% DMSO/ 80% Methanol and incubated overnight at À 20°C. Larvae were then washed in 100 mM NaCl, 100mMTris-HCl, pH7.4 for 30 min at room temperature. Washed larvae were embedded in gelatin from cold water fish skin/sucrose (Sigma). Larvae were sectioned (20 mm) in a cryomicrotome.
Inhibitor treatments. All stock solutions were prepared in DMSO. Embryos were dechorionated at 24 hpf using Pronase (Roche, Basel, Switzerland). For Notch signalling inhibition embryos were incubated from 2 dpf with 10 mM of LY-411575 (Sigma, St Louis, MO, USA) and imaged at 3 dpf. For VEGFR2 and VEGFR3 inhibition embryos were treated with 25 mM MAZ51 (Merck Millipore, Billerica, Massachusetts, USA) from 2.5 dpf or from 3 dpf with 0.125 mM ki8751 (Sigma, St Louis, MO, USA) and imaged at 4 dpf. To inhibit PI3K/Akt signalling embryos were incubated with 1.25 mM wortmannin from 3 dpf and imaged at analysed at 4 dpf. Heartbeat was blocked using 15 mM 2,3-Butanedione 2-monoxime (BDM) dissolved in E3 media. Control embryos were mock treated with DMSO (Sigma, St Louis, MO, USA). Embryos were randomly assigned to experimental groups. Investigators were blinded to inhibitor treatment.
Photoconversion of kikGR and migration tracking. Dechorionated embryos were embedded in 0.7% low-melting agarose at 30 hpf and a small part of the posterior cardinal vein of Tg(kdrl:nlskikGR) hsc7 transgenics was converted for several seconds using UV-light with the smallest available field diaphragm of the Leica Sp8 confocal microscope. Subsequently embryos removed from the agarose and allowed to develop in E3 medium until imaging or were immediately used for time-lapse imaging.
Gal4ERT2 endoxifen activation. Endoxifen (Sigma) was solved in DMSO. Zebrafish embryos expressing Gal4ERT2 were incubated from 52 hpf onwards in 0.5 mM endoxifen in E3 medium in the dark. GFP positive cells could be observed approximately 1.5 h after induction.
Vascular network analysis. To assess sprout number and length, we developed a semi-automated analysis of the DLAV-ISV vessel network using ImageJ ( Supplementary Fig. 1n). Image-stacks of ISVs were acquired using the Leica SP8 confocal microscope. Stack projections of one side of the trunk were generated. Dorsal region of the ISVs was used for analysis. Using ImageJ a Gaussian blur filter was applied followed by a black/white threshold and subsequent skeletonization to generate a skeleton of the vasculature. Segment number, branch point number and total branch length were calculated using the 'analyse skeleton' plugin. The semi-automated pipeline was applied for analysis of 4 dpf vascular networks, while sprout numbers in 2-3 dpf zebrafish embryos were counted manually.
Imaging. Zebrafish larvae were embedded in 0.7% low-melting agarose with 0.112 mg ml À 1 Tricaine (E10521, Sigma) and 0.003% PTU (P7629, Sigma) in glass bottom dishes (MatTek, P35G-0.170-14-C). Images presented in this study were acquired using a Leica SP8 confocal microscope with Â 20 multi-immersion and Â 40 water immersion objectives and LAS X software. Images were processed using ImageJ. Vascular branching was quantified using a semi-automated ImageJ pipeline (Supplementary Fig. 1n). Animal numbers used are indicated in figure legends. For zebrafish mutants more than 100 embryos per genotype were analysed. In morpholino experiments morphologically malformed embryos were excluded from analysis.
Statistical analysis. Statistical analysis was performed using GraphPad Prism 6. Each dataset was tested for normal distribution (D'Agostino and Pearson test). Parametric method (unpaired Students t-test) was only applied if the data were normally distributed. For non-normal distributed data sets, a non-parametric test (Mann Whitney U test) was applied. When appropriate in case of multiple comparisons, ANOVA plus Bonferroni correction was applied. P values o0.05 were considered significant. Data are represented as mean±s.e.m., unless otherwise indicated. *Po0.05, ** Po0.01 and *** Po0.001.
Data availability. The authors declare that all data supporting the findings of this study are available within the article and its Supplementary Information files or from the corresponding author on reasonable request. The RNA-seq data generated in this study has been deposited into the Gene Expression Omnibus database with the accession code http://www.ncbi.nlm.nih.gov/geo/query/ acc.cgi?acc=GSE89350.