SoxF factors induce Notch1 expression via direct transcriptional regulation during early arterial development

Genetic specification of arterial fate has long been attributed to regulation downstream of the Vegfa and Notch pathways. Vegfa signalling is essential for arterial specification in both zebrafish and mammalian models, at least partially by stimulating the expression of components of the Notch pathway, while activation of Notch signalling can rescue defects in Vegfa-deficient zebrafish embryos (Lanahan et al., 2010; Lawson et al., 2002; Liu et al., 2003; Visconti et al., 2002). The Notch receptors Notch1 and Notch4 and the delta-like ligands Dll1, Dll4, Jag1 and Jag2 are expressed in endothelial cells, where they play crucial roles in both arteriogenesis and angiogenesis (Lawson et al., 2002; Liu et al., 2003; Phng and Gerhardt, 2009; Roca and Adams, 2007). Ligand binding to the Notch receptor releases the Notch intracellular domain (NICD), which translocates to the nucleus and forms a transcriptional activation complex with the otherwise repressive DNA-bound Rbpj [CSL, Su(H)] (Bray, 2006). Combined ablation of Notch1 and Notch4, which are both principally expressed in arterial endothelial cells during early vascular remodelling (Chong et al., 2011; Jahnsen et al., 2015), results in severe vascular remodelling defects (Krebs et al., 2000), as does ablation of the Notch downstream effector Rbpj or of Dll4, a Notch ligand specific within the vasculature to arteries (Duarte et al., 2004; Gale et al., 2004; Krebs et al., 2004). However, loss of Notch signalling does not fully recapitulate the arterial defects downstream of Vegfa ablation (Carmeliet et al., 1996; Krebs et al., 2000; Lawson et al., 2002), and the arterially restricted gene expression patterns of components of the Notch pathway do not fully overlap with activated Vegfa (Lawson, 2003), suggesting that additional factors are involved in the regulation of Notch-mediated arterial fate.

The SoxF group of transcription factors (Sox7, Sox17 and Sox18) are expressed in endothelial cells from early in development (Francois et al., 2010). While each SoxF member displays a subtly different endothelial expression pattern, all three factors are expressed early in arterial development (Corada et al., 2013; François et al., 2008; Zhou et al., 2015) and share considerable functional redundancy, complicating interpretation of the consequences of gene disruption (Hosking et al., 2009; Zhou et al., 2015). Combined loss of sox7 and sox18 in zebrafish, both by morpholino-based knockdown and genetic mutation, resulted in serious arteriovenous malformations similar to those seen after Notch ablation, suggesting that SoxF factors might genetically interact with the Notch pathway in endothelial cells (Cermenati et al., 2008; Hermkens et al., 2015; Herpers et al., 2008; Lawson et al., 2001). Further evidence was provided in mice, where endothelial-specific ablation of Sox17 caused arterial differentiation and remodelling defects (Corada et al., 2013), and conditional deletion of SoxF factors in the adult resulted in loss of major vessel identity in the retina (Zhou et al., 2015). Inhibition of Vegfa signalling can significantly impact SoxF expression and activity in both mouse and zebrafish models (Kim et al., 2016; Pendeville et al., 2008; Duong et al., 2014; Pennisi et al., 2000b), suggesting that members of the SoxF family lie downstream of Vegfa. Further, SoxF acts in a positive feed-forward loop to maintain Flk1 (Kdr; kdrl in zebrafish) expression (Kim et al., 2016). By contrast, the ablation of Notch signalling in the vasculature does not significantly impact the expression of SoxF (Abdelilah et al., 1996; Corada et al., 2013). Different experimental models have positioned SoxF genes both upstream (Corada et al., 2013) and downstream (Lee et al., 2014) of Notch signalling in endothelial cells, suggesting a complex relationship between these two pathways.

Analysis of the only known enhancers for the Notch pathway, namely the arterial-specific Dll4-12 (Sacilotto et al., 2013) and Dll4in3 enhancers (Sacilotto et al., 2013; Wythe et al., 2013), has demonstrated a crucial role for SoxF factors in the regulation of expression of the Notch ligand Dll4 in arteries, in combination with Rbpj/Notch binding and in the presence of Ets factors including Erg (Sacilotto et al., 2013; Wythe et al., 2013). Although this work clearly positioned the SoxF and Notch pathways as crucial regulators of arterial specification, and is supported by analysis placing Sox17 upstream of Notch signalling in mouse models (Corada et al., 2013), ablation of arterial marker expression only occurs after the removal of both SoxF factors and Notch signalling in combination (Sacilotto et al., 2013). Consequently, the precise transcriptional hierarchy of SoxF and Notch has yet to be fully established. Although studies of the Notch receptor genes Notch1 and Notch4 also identified SoxF binding motifs within putative promoter sequences (Corada et al., 2013; Lizama et al., 2015), a requirement for these SOX motifs in arterial-specific gene expression of Notch receptors has not been established. In this study, we have identified and characterised arterial-specific enhancers directing NOTCH1/notch1b gene expression in vivo, and used them to demonstrate a direct requirement for SoxF factors in the transcriptional regulation of Notch receptors during early arterial differentiation, positioning SoxF factors upstream of Notch in the acquisition of arterial cell identity.

Previous studies into the transcriptional regulation of the Notch receptors have not extended to the identification or analysis of gene enhancers (cis-regulatory elements) (Corada et al., 2013; Lizama et al., 2015; Wu et al., 2005). We therefore conducted a detailed in silico analysis of the NOTCH1 locus with the aim of identifying novel, arterial-specific enhancers. This analysis focused on human NOTCH1 in order to take advantage of the wealth of publicly available information describing chromatin modifications in human endothelial cell lines. Using this information, we were able to pinpoint four regions of DNA rich in endothelial cell-specific H3K4me1 and H3K27ac histone modifications and DNaseI digital genomic footprints, all marks closely associated with enhancer activity (Heintzman and Ren, 2009; Sabo et al., 2004) (Fig. 1A, Fig. S1).

The putative enhancer regions were named NOTCH1+33, NOTCH1+16, NOTCH1+3/5 and NOTCH1−68 to reflect their distance from the transcription start site (TSS) in kb. Each enhancer region was cloned upstream of the silent hsp68 minimal promoter and the lacZ reporter gene (Fig. 1B) and tested for its ability to drive reporter gene expression specifically in arterial endothelial cells of transient transgenic mice at embryonic day (E) 12-13. Although each of the four putative enhancer regions was able to drive detectable levels of lacZ in transgenic mice, this expression was primarily neural, an expression pattern commonly seen when using the hsp68 minimal promoter (e.g. Becker et al., 2016; Sacilotto et al., 2013) (Table 1, Fig. S2). Only the NOTCH1+33 and NOTCH1+16 enhancers were able to direct expression in endothelial cells (Table 1, Fig. S2). In the case of NOTCH1+33, vascular expression was detected in only one of the five transgenic mice analysed. This expression was not restricted to the arterial endothelium but was pan-endothelial (Table 1, Fig. S2). This agrees with previous reports indicating that the mouse orthologue of this region, termed Notch1_enh1, also directs occasional vascular enhancer activity but did not show arterial-specific expression (Zhou et al., 2017). Conversely, the 274 bp NOTCH1+16 enhancer was able to robustly direct expression specifically to arterial endothelial cells within the vasculature at E12 in multiple independent transgenic embryos (Table 1, Fig. 2 and Fig. S2). This indicates that the NOTCH1+16 enhancer represents a novel, arterially restricted enhancer within the NOTCH1 locus. Analysis of a stable mouse line expressing the NOTCH1+16:lacZ transgene clearly demonstrated that this enhancer is strongly active from the very early stages of vascular development, mimicking the expression of endogenous Notch by becoming restricted to the arteries by late E9.5 and then maintaining an arterial endothelial cell-restricted expression pattern throughout embryonic development (Chong et al., 2011; Jahnsen et al., 2015) (Fig. 2).

To identify the transcription factors that potentially regulate the NOTCH1+16 enhancer, we performed a ClustalW analysis of orthologous mammalian sequences (Fig. 3A). This analysis clearly identified nine conserved core consensus ETS binding motifs [GGAW (Hollenhorst et al., 2011)] and two consensus SOX binding motifs [WWCAAW (Mertin et al., 1999)] (Fig. 3A) within the 274 bp NOTCH1+16 enhancer. Because not all in silico consensus binding motifs are able to functionally bind the cognate protein, these motifs were then tested by electrophoretic mobility shift assay (EMSA). Both SOX motifs (termed hmSOX-a and hmSOX-b) were able to bind recombinant SOX7 and SOX18 proteins in EMSA (Fig. 3B), and three ETS motifs (termed hmETS-a, hmETS-b and hmETS-c) were able to bind the endothelial Ets protein ETV2 (Fig. 3C).

ETS motifs are common to all endothelial-expressed gene enhancers (De Val and Black, 2009). Although the Ets factor Erg has been implicated in arterial specification (Wythe et al., 2013), previous studies have shown that ETS motifs were unable to direct expression of the arterial-specific Dll4 and Flk1 enhancers without additional transcription factor binding motifs (Becker et al., 2016; Sacilotto et al., 2013; Wythe et al., 2013), suggesting that Ets factors alone are unlikely to regulate the NOTCH1+16 enhancer.

To test whether the SOX motifs play a role in NOTCH1+16 enhancer activity, we mutated the core nucleotides of these motifs (see Materials and Methods for sequences) and tested the ability of the resultant NOTCH1+16mutSOX-a/b enhancer to drive reporter gene expression. Strikingly, enhancer mutation resulted in a dramatic reduction in reporter gene expression in endothelial cells in transgenic mice, although transgene expression was detected outside of the vascular system (Fig. 4, Table 1). This result differs notably from that reported for the Dll4 enhancers, where mutations in SOX motifs, or loss of SoxF factors, resulted in no detectable decrease in Dll4 expression unless accompanied by ablation of Notch signalling (Sacilotto et al., 2013).

The SoxF-dependent NOTCH1+16 enhancer robustly directed arterial-restricted expression in transgenic mouse models. However, this enhancer did not exhibit sequence conservation beyond mammals (Fig. S1B), leaving it unclear how relevant these observations are to Notch signalling during arteriovenous specification in zebrafish, an extremely well-studied model system (Gore et al., 2012). We therefore investigated whether SoxF factors were able to transcriptionally regulate the zebrafish orthologue of NOTCH1, notch1b, by examining the binding patterns of the zebrafish SoxF transcription factors around the notch1b locus. Zebrafish SoxF are expressed in early endothelial cells and implicated in arteriovenous differentiation (Cermenati et al., 2008; Hermkens et al., 2015; Herpers et al., 2008; Pendeville et al., 2008). To probe for SoxF (Sox7, Sox17 and Sox18) genome-wide binding locations we used an endothelial-specific SOX18Ragged overexpression line. The SOX18Ragged dominant-negative protein has been shown to interfere with the endogenous function of all three SoxF transcription factors (James et al., 2003; Pennisi et al., 2000b). Using 26-28 hours post fertilisation (hpf) embryos from the tg(fli1a:Gal4FF;10×UAS:Sox18Ragged-mCherry) zebrafish line, in which a tagged SOX18Ragged is expressed specifically in endothelial cells (Fig. S3A), ChIP-seq analysis identified a SoxF binding event 15 kb upstream of the notch1b first exon (Fig. 5A). This binding peak was enriched in the enhancer-associated histone modifications H3K4me1 and H3K27ac at 24 hpf (Bogdanović et al., 2012; Kent et al., 2002) (Fig. 5A), suggesting that it might represent a novel enhancer of notch1b. A 1219 bp zebrafish DNA fragment corresponding to the SOX18-bound region, termed the notch1b-15 enhancer, was cloned upstream of a silent gata2a promoter and GFP reporter gene within the zebrafish enhancer detection (ZED) vector (Bessa et al., 2009) (Fig. 5B) and used to generate the stable tg(notch1b-15:GFP) fish line (Fig. 5C). The GFP transcript was detected in the vascular cord around the midline from 19 hpf, and persisted in the vascular rod as it formed the dorsal aorta at 22 hpf (Fig. 5C). GFP expression continued to be restricted to the dorsal aorta and the segmental arteries in larvae from 24 hpf until 48 hpf. This arterial-restricted pattern of expression within the vasculature was similar to that of endogenous notch1b (Fig. S3B), indicating that the SOX18-bound notch1b-15 element is a bona fide notch1b enhancer and suggesting that SoxF factors directly transactivate Notch receptor transcription in arterial endothelial cells in zebrafish.

ClustalW analysis comparing the orthologous enhancer sequences from fugu, stickleback and medaka revealed a remarkably similar pattern of conserved transcription factor motifs when compared with the NOTCH1+16 enhancer (Fig. 6A), with multiple ETS and two SOX binding motifs, termed zfSOX-a and zfSOX-b, confirmed by EMSA analysis (Fig. 6B,C). To establish whether the zfSOX-a and zfSOX-b binding motifs were required for notch1b-15 arterial enhancer function, we generated transient transgenic fish lines harbouring mutated SOX binding motifs (notch1b-15mutSOX-a/b:GFP) and compared the activity of the transgene with wild-type (WT) notch1b-15:GFP control transient transgenic animals (Fig. S4A,B). Simultaneous disruption of both zfSOX-a and zfSOX-b sites led to a reduction of arterial-specific GFP expression in endothelial cells. Although a minority of mutant fish still expressed GFP after SOX binding site mutation, the loss of expression was still much greater than that seen after SOX motif mutation in a previously published Dll4 enhancer in transgenic zebrafish, where vascular expression rates were unaffected by mutations of SoxF binding motifs (Sacilotto et al., 2013), supporting a key role for SoxF factors in notch1b activation. To further confirm our observation, we established stable transgenic notch1b-15mutSOX-a/b:GFP fish lines and compared them with the established WT notch1b-15:GFP lines. Analysis of the stably transgenic embryos (Fig. 7A) confirmed a GFP expression pattern in the dorsal aorta and segmental arteries for the WT transgene. By contrast, the tg(notch1b-15mutSOX-a/b:GFP) lines showed ectopic GFP expression in neurons and a significant decrease of GFP expression in the arterial endothelium (Fig. 7A). Quantitative analysis of GFP intensity in both dorsal aorta and segmental arteries showed lower expression in most tg(notch1b-15mutSOX-a/b:GFP) than in WT tg(notch1b-15:GFP) embryos (Fig. 7B,C, Fig. S4C). Taken together, these data clearly demonstrate that the zfSOX-a/b binding sites are required to guide notch1b-15-specific enhancer activity in vivo during arterial development.

We next investigated the consequences of morpholino (MO)-based knockdown of SoxF on tg(notch1b-15:GFP) fish. sox7/sox18 double morphants exhibit a severe vascular phenotype (Fig. S5), including fusions and shunts between the dorsal aorta and cardinal vein (Cermenati et al., 2008; Herpers et al., 2008; Pendeville et al., 2008), phenotypes that are shared with sox7;sox18 double-mutant fish (Hermkens et al., 2015). MO-induced transcript depletion of sox7 or sox18 resulted in downregulation of notch1b-15:GFP expression, while sox7/sox18 double-morphant tg(notch1b-15:GFP) fish demonstrated a near-complete loss of reporter gene expression (Fig. 7D,E).

Taken together, these results indicate that SoxF proteins directly modulate the activity of arterial-specific enhancers for both the mammalian NOTCH1 and zebrafish Notch1b receptors, positioning SoxF transcription factors directly upstream of Notch signalling during early arterial differentiation in both mammals and zebrafish.

To assess whether the endogenous notch1b-15 regulatory element is functionally relevant during arteriovenous differentiation in vivo, we deleted the endogenous notch1b-15 enhancer in zebrafish. The resultant notch1b-15^uq1mf allele was generated using two guide RNAs to drive rapid genome editing using the CRISPR/Cas9 system (Fig. 8A), resulting in excision of a 203 bp fragment overlapping the notch1b-15 enhancer. Analysis of endogenous notch1b expression in the F₂ generation demonstrated lower notch1b expression levels in both dorsal aorta and arterial intersomitic vessel (aISV) of the notch1b-15^uq1mf/uq1mf homozygous embryos as compared with their sibling controls, whereas no change was observed in the neural tube (Fig. 8B). Next, we assessed the notch1b transcript levels in purified flt1-positive arterial endothelial cell populations from the F₃ generation of homozygous fish (F₂ notch1b-15^uq1mf homozygous in-cross) compared with WT control larvae (F₂ notch1b-15^+/+ × flt1:YFP;lyve1:dsRed) at 24-28 hpf (Fig. 8C). As expected, notch1b transcripts were significantly downregulated in the homozygous animals, strongly supporting a role for the notch1b-15 enhancer in the transcription of endogenous notch1b in arterial endothelial cells.

To assess the phenotypic outcome of notch1b-15 loss of function, we took advantage of the tg(notch1b-15^uq1mf/+;flt1:YFP;lyve1:dsRed) line to analyse the developing vasculature after notch1b-15 enhancer deletion in both the F₂ and F₃ generations. Surprisingly, given the reduced levels of notch1b (Fig. 8B), no overt vascular phenotype was detected in F₂ notch1b-15^uq1mf/uq1mf homozygous zebrafish (F₁ heterozygous notch1b-15^uq1mf/+ in-cross) (Fig. S6A), suggesting partial enhancer redundancy. Such redundancy, which is potentially explained by the pervasiveness of redundant, or ʻshadow', enhancers around developmental genes (Cannavò et al., 2016), has previously been well documented in key endothelial genes, with examples including the Dll4, Flk1 and Tal1 loci (Cannavò et al., 2016). By contrast, we detected a subpopulation (20-30%) of larvae from the F₃ generation (F₂ homozygous notch1b-15^uq1mf/uq1mf in-cross) that displayed a phenocopy of the notch1b loss-of-function phenotype (Fig. S6B). This increase in the phenotypic severity in the F₃ generation suggests that, in the context of the notch1b-15^uq1mf/+ cross, maternal mRNA deposition is likely to help compensate for the disruption of notch1b transcription caused by deletion of the notch1b enhancer, a compensation that is reduced in a purer notch1b-15^uq1mf/uq1mf genetic background (Harvey et al., 2013).

To bypass potential rescue effects of shadow enhancers or maternally deposited transcripts, we also investigated arteriovenous differentiation and sprouting angiogenesis in F₂ notch1b-15^uq1mf/uq1mf zebrafish after low-level depletion of notch1b mRNA. A splice notch1b MO was injected into eggs from a notch1b-15^uq1mf/+ in-cross in the tg(flt1:YFP;lyve1:dsRed) background. The notch1b MO was used at suboptimal concentration (5 ng/embryo), which is known to result in minimal phenotypes (Sacilotto et al., 2013). The developing vasculature of each resulting embryo was analysed blindly at 3 dpf, and genotypes were assigned to embryos after image acquisition (Fig. 8D, red). Whereas most WT siblings had normal ISV development, we observed an increased number of hypersprouting ISVs across the notch1b-15^uq1mf/+ and notch1b-15^uq1mf/uq1mf population (Fig. 8E, asterisks; Fig. 8F, top) in a gene dosage-dependent manner, similar to the phenotype described previously in high MO concentration notch1b morphants and Notch signalling-deficient embryos (Geudens et al., 2010; Siekmann and Lawson, 2007). Further, mutant embryos also demonstrated loss of arterial connections between the dorsal aorta and dorsal longitudinal anastomotic vessel, similar to those described in Notch signalling-deficient zebrafish (Geudens et al., 2010; Quillien et al., 2014), while the notch1b-depleted notch1b-15^+/+ and notch1b-15^uq1mf/+ morphant embryos demonstrated an equal proportion of arterial and venous ISVs as previously reported (Bussmann et al., 2010) (Fig. 8E,F bottom).

To further confirm that interfering with notch1b-15 enhancer activity is additive to notch1b transcript depletion, we chemically treated the notch1b-15^uq1mf/+ cross with DAPT, a well characterised Notch signalling inhibitor, over the course of endothelial differentiation (15- to 16-somite stage through to 3 dpf) (Fig. 8D, blue). All embryos treated with a suboptimal concentration of DAPT (5 µM) showed a straight body axis, indicating that somitogenesis (and therefore Notch activity) was not significantly compromised (Fig. S7A). Interestingly, despite this lack of morphological defects, F₂ fish homozygous for the notch1b-15^uq1mf allele displayed a lower arterial-to-total ISV ratio (Fig. 8G, Fig. S7A). By contrast, fish homozygous for the notch1b-15^uq1mf allele treated with DMSO vehicle alone had a comparable aISV ratio to both notch1b-15^+/+ and notch1b-15^uq1mf/+ siblings (Fig. S7B), similar to the untreated control. This suggests that the observed loss of aISV is specific to an additive effect of DAPT treatment and notch1b-15 enhancer activity disruption. Overall, these data suggest a functional role of the notch1-15 enhancer in the endothelial-specific initiation of notch1b transcription to promote the acquisition of arterial cell identity.

Our results have clearly implicated SoxF factors as direct upstream regulators of arterial Notch enhancers, and therefore suggest a considerably greater role for SoxF in the regulation of the Notch receptors than of the Notch ligands. However, since the notch1b-15 enhancer is partially redundant with other notch1b shadow enhancers, we wished to establish whether SoxF regulation is required for endogenous notch1b expression itself, not just enhancer activity. Further, our results so far do not entirely rule out the possibility of SoxF/Rbpj combinatorial regulation of notch1b, as was previously shown for Dll4 enhancers (Sacilotto et al., 2013). Although neither the human NOTCH1+16 nor the zebrafish notch1b-15 enhancer contains conserved consensus Rbpj/Notch binding motifs, transcription factors can bind non-consensus motifs, and not all transcription factors necessarily bind conserved motifs (Wong et al., 2015). The nature of the SoxF/Notch combinatorial regulation of Dll4, where the SOX or RBPJ binding motifs play functionally interchangeable roles, indicates potential direct interactions between these two proteins, such that only a single SOX binding motif might be necessary for SoxF/Rbpj synergy. We therefore investigated the consequences of SoxF depletion on endogenous Notch1/notch1b expression in vivo.

Although Sox17 is robustly expressed in arterial endothelial cells (Corada et al., 2013; Hosking et al., 2009), compound Sox7;Sox18 deletion in mice resulted in a reduction of Notch1 mRNA levels in the trunk dorsal aorta and primitive heart cavities of E8.5 embryos (Fig. 9A, Fig. S8A). These results concur with observations in the mouse retina, where the vascular phenotype after Sox7;Sox17;Sox18 endothelial-specific triple deletion closely resembled defects caused by loss of Notch signalling (Zhou et al., 2015). Strikingly, both MO-induced gene knockdown and compound mutation of sox7 and sox18 in zebrafish embryos also led to a near-complete loss of notch1b transcript expression specifically in endothelial cells, as shown by in situ hybridisation analysis (Fig. 9B,C, Fig. S8B). These results further establish an essential role for SoxF transcription factors in the induction of Notch1/notch1b gene expression, and position SoxF proteins at the top of the transcriptional hierarchy regulating arterial specification.

Recent work has implicated SoxF, Ets and Rbpj, the Notch transcriptional effector, in the regulation of the Notch ligand Dll4 and many other key arterial genes (Corada et al., 2013; Lizama et al., 2015; Sacilotto et al., 2013; Wythe et al., 2013), but has not established the hierarchical arrangement of these diverse factors in arterial specification and differentiation. In this study, we demonstrate that arterial expression of the Notch receptor Notch1/notch1b, a key player in arterial specification, is directly downstream of SoxF regulation in both fish and mouse. Unlike other key arterial specification markers, including Dll4, Efnb2a and Dlc (Sacilotto et al., 2013), ablation of Notch1/Notch1b expression after depletion of SoxF factors occurred without concurrent inhibition of Notch signalling. Therefore, this work positions SoxF factors directly above Notch signalling in the transcriptional hierarchy initiating arterial development, and suggests that SoxF factors might initiate a feed-forward loop directing arterial identity. In this model, SoxF factors would first activate Notch signalling via the transcriptional activation of Notch receptors in combination with weak activation of Notch ligands (Sacilotto et al., 2013). This early SoxF-mediated activity would then be boosted by the initiation of Notch signalling, resulting in the sustained activation of other downstream genes, eventually activating the full cohort of genes necessary to acquire and maintain arterial endothelial cell identity.

Understanding the function of SoxF factors through mutational analysis has presented significant challenges. Strain-specific variations in mice after depletion of individual SoxF genes and varying levels of compensation from other SoxF factors have resulted in some contradictory reports (Corada et al., 2013; Lee et al., 2014), as is also the case for zebrafish morphant analysis (Cermenati et al., 2008; Herpers et al., 2008; Pendeville et al., 2008). Nonetheless, the results described here agree with an increasingly convincing body of work suggesting that SoxF factors influence Notch signalling yet are unaffected by Notch ablation. For example, overexpression of Sox17 upregulates components of the Notch pathway (Corada et al., 2013; Lizama et al., 2015), loss of functional SoxF factors results in defects similar to those observed after Notch inhibition in mice and fish (Corada et al., 2013; Sakamoto et al., 2007; Zhou et al., 2015), while Notch ablation results in little alteration to the endothelial expression of SoxF factors in mice and fish (Abdelilah et al., 1996; Corada et al., 2013). Data reported here combine with these reports to strongly support a role for SoxF factors as part of the initial transcriptional machinery that instructs arterial specification events.

However, some questions remain. In particular, it is notable that sox7;sox18 double-mutant fish, although exhibiting severe arteriovenous defects very similar to those seen in Notch-deficient fish (Lawson et al., 2001), do not fully recapitulate the effects of Vegfa depletion on arterial specification (Lawson et al., 2002). While this difference may in part be attributed to weak expression of zebrafish sox17, which is expressed in some arterial endothelial cells (Hermkens et al., 2015), it is also expected that the Vegfa pathway has a wider effect on arterial endothelial cells more generally, away from SoxF-mediated activation of Notch signalling. SoxF factors are also influenced by signalling pathways beyond Vegfa. The diverse nature of Vegfa roles in the vasculature, including the regulation of both sprouting angiogenesis and arteriogenesis, processes that inevitably involve different cohorts of downstream targets, make it necessary that multiple regulatory pathways interact with Vegfa during vascular development. While recent work has shown that Vegfa signalling increases the nuclear translocation of SoxF (Duong et al., 2014), and inhibition of Vegfa results in the loss of vascular sox7 in fish, sox18 is still expressed in the absence of intact Vegfa signalling (Pendeville et al., 2008). Additionally, loss of the Vegf co-receptor Nrp1 has little effect on SoxF expression in mouse retinal vasculature (Zhou et al., 2015), pointing to other upstream influences on SoxF function in endothelial cells. Other upstream effectors of SoxF function are likely to include canonical Wnt signalling and Vegfd, both of which have been shown to influence SoxF nuclear localisation (Corada et al., 2013; Duong et al., 2014; Zhou et al., 2015).

Recent evidence has also implicated a role for a coordinated Vegf-Mapk-Ets pathway in the induction of Notch signalling components and early arterial differentiation (Wythe et al., 2013). Notably, in addition to SoxF motifs, both the Notch1 and Dll4 enhancers share a number of highly conserved consensus motifs for the Ets family of transcription factors. While ETS motifs are common to all vascular enhancer elements, including many that are not preferentially expressed in the arterial vasculature (De Val and Black, 2009), the Ets factor Erg has been specifically implicated in arterial-specific regulation of Dll4 (Wythe et al., 2013). It is therefore likely that Vegfa-mediated activation of Ets factors may contribute to the transcriptional activity of Notch downstream effectors, and thus may influence arterial establishment independently of SoxF factors. However, it is notable that Vegfa-mediated activation of Ets transcription factors alone does not appear to be sufficient for arterial gene expression. Dll4 enhancers lacking SOX and RBPJ motifs but retaining all ETS motifs were unable to drive any transgene expression (Sacilotto et al., 2013), nor were Notch enhancers lacking SOX motifs (Figs 2-4). Similar results were found in other delineated arterial enhancers, including the Ece1 upstream enhancer (Robinson et al., 2014) and the Flk1 intron 10 enhancer, where loss of Rbpj-mediated repression resulted in expansion of enhancer activity into venous cells without alterations to ETS motif binding (Becker et al., 2016). Combined with recent observations demonstrating that Erg also plays a crucial role in venous specification through activation of Aplnr (Lathen et al., 2014), it is therefore likely that the role of Erg, and of other Ets factors, downstream of Vegfa in arterial-restricted gene expression occurs in co-operation with additional essential, arterial-specifying transcription factors. The data presented in this work, combined with the analysis of further arterial enhancers, including those of Dll4 and Ece1 (Robinson et al., 2014; Sacilotto et al., 2013; Wythe et al., 2013), increasingly suggest that SoxF may fulfil this role.

The 10×UAS:Sox18Ragged-mCherry plasmid was generated using the full-length mouse Sox18Ragged cDNA sequence, tagged with 10×UAS and mCherry and cloned into pDestTol2CG2 (Kwan et al., 2007). notch1b-15:GFP WT was generated by cloning a 1219 bp PCR fragment from zebrafish genomic DNA together with gata2a promoter and GFP reporter gene into the zebrafish enhancer detection (ZED) vector (Bessa et al., 2009). notch1b-15mutSOX-a/b was generated by site-directed PCR mutagenesis of the WT construct. The NOTCH1−68, NOTCH1+3/5 and NOTCH1+33 enhancers were generated by PCR from human genomic DNA, NOTCH1+16 WT and NOTCH1+16mutSOX-a/b enhancers were generated as custom-made, double-stranded linear DNA fragments (GeneArt Strings, Life Technologies). All mammalian fragments were cloned into the hsp68-lacZ Gateway vector (provided by N. Ahituv) (De Val et al., 2004). Primers and sequences for DNA fragments are listed in the supplementary Materials and Methods.

Animal procedures were approved by local ethical review and licensed by the UK Home Office or conformed to institutional guidelines of the University of Queensland Animal Ethics Committee. Transgenic mice were generated by oocyte microinjection and analysed as detailed in the supplementary Materials and Methods (De Val et al., 2004). Compound Sox7^−/−;Sox18^−/− (C57BL/6) mouse embryos were generated on the C57BL/6 background through crossing heterozygous Sox7:tm1 to Sox18:tm1 generating Sox7^+/−;Sox18^+/− mice, which were subsequently in-crossed (Pennisi et al., 2000a).

Transgenic zebrafish embryos were generated using the Tol2 system in conjunction with the ZED vector (Bessa et al., 2009). The sox7^hu5626; sox18^hu10320 double-homozygous mutant zebrafish have been described previously (Hermkens et al., 2015). The tg(fli1a:Gal4FF,10×UAS:Sox18Ragged-mCherry) zebrafish line was generated by crossing 10×UAS:Sox18Ragged-mCherry with fli1a:Gal4FF, 4×UAS Utrophin GFP. MO-mediated knockdown was performed as previously described (Herpers et al., 2008). CRISPR genome editing for notch1b-15 was performed as described by Gagnon et al. (2014) using the primers listed in the supplementary Materials and Methods to generate notch1b-15^uq1mf (203 bp deletion) allele.

The F₂ notch1b-15^uq1mf/uq1mf was generated by in-crossing notch1b-15^uq1mf/+, while F₃ notch1b-15^uq1mf/uq1mf was generated from the F₂ notch1b-15^uq1mf/uq1mf in-cross, both in the tg(flt1:YFP;lyve1:dsRed) background.

Positive tg(fli1a:Gal4FF;10×UAS:Sox18Ragged-mCherry) fish larvae were collected at 26-28 hpf and processed as described in supplementary Materials and Methods (Mohammed et al., 2013). DNA amplification was performed using the TruSeq ChIP-seq Kit (Illumina, IP-202-1012) following immunoprecipitation. The library was quantified using the KAPA library quantification kit for Illumina sequencing platforms (KAPA Biosystems, KK4824) and 50 bp single-end reads were sequenced on a HiSeq 2500 (Illumina) following the manufacturer's protocol. FASTQ files were mapped to GRCz10/danRer10 genome assembly using bowtie (Langmead, 2010), and peaks were called using MACS version 2.1.0. using input as a reference. To avoid false-positive peaks calling due to the mCherry epitope, ChIP-seq with the mCherry epitope only was performed in parallel to SOX18Ragged-mCherry ChIP-seq and peaks called in these experimental conditions were subtracted from the peaks called in the SOX18Ragged-mCherry conditions. For details, see the supplementary Materials and Methods.

Sequences were analysed for consensus sequence motifs by eye and using TRANSFAC (BIOBASE; http://genexplain.com/transfac/) (Matys et al., 2006). EMSAs were performed as previously described (De Val et al., 2004), as outlined in the supplementary Materials and Methods.

MO-mediated knockdown was performed as previously described (Duong et al., 2014; Cermenati et al., 2008). ATG MOs against sox7 and sox18 were injected into the tg(notch1b-15:GFP) stable line at the 1-2 cell stage at 5 ng/embryo (Herpers et al., 2008). To assess the effect of sox7/18 knockdown on endogenous notch1b transcripts, sox7 and sox18 MOs were injected into WT zebrafish larvae at 1 ng/embryo in parallel with a standard control MO (std-MO) injected at 2 ng/embryo (Cermenati et al., 2008). To characterise notch1b-15^uq1mf, notch1b MO was injected into the notch1b-15^uq1mf/+ cross at 5 ng/embryo. For MO sequences, see the supplementary Materials and Methods.

N-[(3,5-difluorophenyl)acetyl]-L-alanyl-2-phenyl]glycine-1,1-dimethylethyl ester (DAPT; Sigma-Aldrich) was used at 5 μM dissolved in 1% DMSO. Fish were treated from the 15- to 16-somite stage to 3 dpf and the medium containing DAPT was refreshed daily.

Whole-mount in situ hybridisation in zebrafish larvae was performed as described (Coxam et al., 2015; Duong et al., 2014). Section and whole-mount in situ hybridisation in mouse was performed as described (Metzis et al., 2013; Fowles et al., 2003). The Notch1 probe was generated by PCR from mouse embryo cDNA pool at E14.5, and reverse transcribed with T7 polymerase. Whole-mount immunohistochemistry for anti-GFP was performed as described (Koltowska et al., 2015). For details, see the supplementary Materials and Methods.

To characterise the vasculature in Fig. 8E-G and Figs S6, S7, intersomitic vessels (20-22 ISVs) expressing tg(flt1:YFP) were­­­ analysed across 10-11 somites through a z-stack using ImageJ (NIH) after image acquisition by confocal microscopy. Intersomitic vessels connecting the dorsal longitudinal anastomotic vessel to the dorsal aorta expressing YFP were assigned as arterial ISVs. Vessels were also scored for ectopic sprouting. The proportion of aISVs or hypersprouts among the total number of ISVs was analysed by two-tailed Mann–Whitney U-test. To quantify the GFP intensity of tg(notch1b-15:GFP) and tg(notch1b-15mutSOX-a/b:GFP) (Fig. 7A), two to three ISVs across five to six somites in the trunk region were analysed using ImageJ. A region of interest (ROI) covering a single ISV was selected and mean pixel intensity for each ISV was quantified from each individual stack across three z-sections. This value was further corrected by subtracting the background value. Average ISV GFP intensity (quantified from two to three ISVs) for each fish larva was subsequently corrected for its genomic GFP copy number. A similar method was used to quantify GFP intensity in the endothelial lining along the dorsal aorta.

Flt1:YFP-positive endothelial cells were isolated from WT and F₃ notch1b-15^uq1mf/uq1mf at 24-28 hpf. RNA was extracted, amplified and cDNA was synthesized as previously described (Coxam et al., 2014; Picelli et al., 2014). Primer sequences and details of the quantitative PCR analysis are provided in the supplementary Materials and Methods.

We thank M. Shipman for help with imaging. The ZED vector was kindly provided by F. Tessadori (Bakkers lab, Hubrecht Institute). The notch1b plasmid used to generate the probe was kindly provided by N. D. Lawson (University of Massachusetts Medical School).