Abstract
The forkhead transcription factors Foxc1 and Foxc2 are essential to establish intact vascular networks in mammals. How these genes interact with endothelial signalling pathways to exert their functions remains incompletely understood. We have generated novel zebrafish mutants in foxc1a and foxc1b, the zebrafish orthologues of mammalian Foxc1, to determine their function during angiogenesis. foxc1a mutants display abnormal formation of cranial veins including the primordial hindbrain channels (PHBC), reduced Vascular Endothelial Growth Factor (VEGF) receptor expression in these and loss of central arteries. foxc1b mutants are normal, whereas foxc1a; foxc1b double mutants exhibit ectopic angiogenesis from trunk segmental arteries. Dll4/Notch signalling is reduced in foxc1a; foxc1b double mutant arteries and ectopic angiogenesis can be suppressed by induction of Notch or inhibition of Vegfc signalling. We conclude that foxc1a and foxc1b play compensatory and context-dependent roles to co-ordinate angiogenesis by promoting venous sprouting via induction of VEGF receptor expression whilst antagonising arterial sprouting by inducing Dll4/Notch signalling. foxc1a/b mediated induction of both pro- and anti-angiogenic axes of VEGF-Dll4/Notch negative feedback imparts competition to balance arterial and venous angiogenesis within developing vascular beds.
Summary Statement foxc1a and foxc1b promote angiogenesis from veins and suppress angiogenesis from arteries by promoting competing pro-angiogenic Vascular Endothelial Growth Factor signalling, and anti-angiogenic Dll4/Notch signalling in zebrafish embryos.
Introduction
The cardiovascular system is the first functional organ system to form during vertebrate embryogenesis and is essential to facilitate growth and development. Within the cardiovascular system, blood vessels represent the primary conduit for nutrient transport and metabolic exchange and these are formed from endothelial progenitor cells, or angioblasts, which are first specified within the lateral plate mesoderm. During somitogenesis, angioblasts differentiate into arterial and venous cell fates and undergo medial migration to coalesce into a solid linear mass of cells or vascular cord. These vascular cords undergo subsequent lumenisation and are remodelled into a functional and complex vascular network by a process of selective cell sprouting termed angiogenesis. Angiogenesis requires complex co-ordination of cell signalling and cellular behaviours to generate a branched vascular morphology capable of maximising surface area for metabolic exchange while minimising transport distances. While angiogenesis is tightly regulated by interactions between pro-angiogenic Vascular Endothelial Growth Factor (VEGF) and anti-angiogenic Notch signalling, reviewed in (Blanco and Gerhardt, 2013), how these pathways are transcriptionally regulated is not fully understood.
VEGF is a morphogen which signals via distinct ligands to induce migratory behaviour within endothelial cells (ECs) and drive blood vessel sprouting. In zebrafish, angiogenesis from arteries is promoted primarily by interaction of Vegfa with its cognate receptor Vegfr4/Kdrl or Vegfr2/Kdr (Bahary et al., 2007; Covassin et al., 2006; Habeck et al., 2002; Nasevicius et al., 2000), whereas Vegfc promotes sprouting from veins via Vegfr3/Flt4 (Hogan et al., 2009b; Le Guen et al., 2014; Villefranc et al., 2013). Within an angiogenic sprout, leading ECs are termed tip cells and these are followed by trailing stalk cells. As an angiogenic sprout forms, a migrating EC extends filopodia to sense VEGF signals and upregulate Dll4 transcription (Gerhardt et al., 2003; Hellstrom et al., 2007), inducing Notch signalling in neighbouring cells and this acts to limit excessive angiogenic sprouting (Siekmann and Lawson, 2007). Notch signalling inhibits expression of VEGF receptors in neighbouring stalk cells (Lobov et al., 2007), thereby limiting the ability of these cells to respond to VEGF and controlling the number of tip cells per sprout. Flt4 is expressed in angiogenic tip cells (Gore et al., 2011; Shin et al., 2016) and hypersprouting of arteries induced by knockdown of dll4 can be rescued by inhibition of Vegfc-Flt4 signalling (Hogan et al., 2009b; Villefranc et al., 2013). Thus, the interplay between VEGF and Dll4/Notch signalling dynamically controls behaviour of angiogenic sprouts within developing vascular networks, reviewed in (Blanco and Gerhardt, 2013).
The forkhead family of transcription factors possess a highly conserved forkhead DNA binding domain (Kaufmann and Knochel, 1996; Lai et al., 1993) and more than 40 forkhead box (Fox) proteins have been identified in mammals (Ivanov et al., 2013). Members of the C class of Fox proteins, FOXC1 and FOXC2 have been implicated in vascular development (Kume, 2009) and mutations in FOXC1 have been described in cerebral small vessel disease which can lead to stroke (French et al., 2014). FOXC1 is also mutated in Axenfeld-Rieger syndrome which causes craniofacial defects including iris hypoplasia (Smith et al., 2000). Mutations in FOXC2 cause lymphoedema-distichiasis syndrome and result in primary lymphoedema (Mellor et al., 2011). Murine Foxc1 or Foxc2 mutants exhibit pre-or perinatal lethality with variable cardiovascular and skeletal defects (Iida et al., 1997; Smith et al., 2000; Topczewska et al., 2001a; Winnier et al., 1997; Winnier et al., 1999). Foxc1; Foxc2 double mutant mice exhibit similar developmental abnormalities to Foxc1 single mutants but with increased severity and abnormal arteriovenous specification, indicating redundant functions between these genes (Seo et al., 2006; Topczewska et al., 2001a). Murine Foxc1; Foxc2 double mutants die by E9.5 preventing detailed analysis of angiogenesis, however some Foxc1+/-; Foxc2-/- embryos survive until E12.5 and exhibit arteriovenous malformations (Seo et al., 2006).
In teleosts, Foxc2 has been lost, while Foxc1 has undergone duplication to generate the paralagous genes foxc1a and foxc1b (Topczewska et al., 2001b) and this study. Functional studies of foxc1a and foxc1b during vascular development have relied on gene knockdown and mutant analysis has been lacking. Knockdown of foxc1a and foxc1b by morpholino has been reported to induce cerebral haemorrhage and circulation defects (De Val et al., 2008; Skarie and Link, 2009) and this was suggested to arise from arteriovenous malformations and reduced vascular basement integrity (Skarie and Link, 2009). However, reported disruption of angiogenesis induced by knockdown of foxc1a and foxc1b are inconsistent between studies (De Val et al., 2008; Skarie and Link, 2009). Foxc transcription factors can directly activate the Dll4 and Hey2 promoter in cultured ECs, indicating these genes function to promote Notch signalling (Hayashi and Kume, 2008; Seo et al., 2006). In zebrafish, foxc1a synergises with Notch signalling during kidney development and Foxc1a can activate Notch targets in vitro (O’Brien et al., 2011), however whether foxc1a/b act upstream of Notch during zebrafish vascular development is not known. Foxc2 combinatorically activates EC genes in partnership with Etv2 in Xenopus embryos and foxc1a/b knockdown in zebrafish reduced angiogenesis within the developing trunk (De Val et al., 2008). In addition, knockdown of foxc1a/b in zebrafish has been reported to reduce Etv2 promoter activity, suggesting these genes also function upstream of Etv2 (Veldman and Lin, 2012). Several recent studies using foxc1a mutants have indicated a requirement for foxc1a during neural circuit development (Banerjee et al., 2015) in addition to anterior somite formation (Hsu et al., 2015; Li et al., 2015) which is consistent with previous knockdown studies (Topczewska et al., 2001b). However, mutant analysis of foxc1a during vascular development and investigation of potential genetic interaction between foxc1a and foxc1b during this process remain unaddressed.
We have generated novel zebrafish foxc1a and foxc1b mutants and characterised the function of these transcription factors during angiogenesis. foxc1a mutants display abnormal cranial angiogenesis including delayed formation of primordial hindbrain channels and almost total loss of central arteries within the developing hindbrain, whereas foxc1b mutants are morphologically normal. By contrast, foxc1a; foxc1b double mutants exhibit ectopic sprouting of segmental arteries within the developing trunk and reduced segmental vein formation, in addition to abnormal cranial vessel formation observed in foxc1a single mutants. We find that foxc1a promotes venous expression of VEGF receptors including vegfr3/flt4 and vegfr4/kdrl, thereby promoting venous angiogenesis, whilst foxc1a and foxc1b genetically interact to limit angiogenesis from arteries by suppressing Vegfc/Flt4 signalling via induction of Dll4/Notch signalling. Our data indicates foxc1a and foxc1b play compensatory and context-dependent roles to co-ordinate angiogenesis from arteries and veins via differential regulation of pro-and anti-angiogenic signalling.
Results
foxc1a mutants display multiple vascular abnormalities while foxc1b mutants are morphologically normal
Mammals have two Foxc genes, Foxc1 and Foxc2 (Fig. S1, S2), whereas teleost lineages have no identifiable foxc2 ortholog (Fig. S2) but possess two paralogous Foxc1 orthologues, foxc1a and foxc1b (Topczewska et al., 2001b). To investigate the specific functions of foxc1a and foxc1b during vascular development, we generated zebrafish mutants using genome editing (see methods). Wild type foxc1a and foxc1b comprise single exon genes, which encode proteins containing 476 and 433 amino acids respectively (Fig. 1A, B) (Topczewska et al., 2001b). The foxc1ash356 mutant allele contains a 4bp insertion, generating a protein which is predicted to retain the first 56 amino acids of wild type Foxc1a before shifting frame and truncating following 14 incorrect amino acids (Fig. 1A; Fig. S3 A, B). The foxc1bsh408 mutant allele contains a 13bp deletion, which is predicted to shift frame after the first 58 amino acids and prematurely truncate the protein following 10 incorrect amino acids (Fig. 1B; Fig. S3 C, D). Both foxc1ash356 and foxc1bsh408 alleles are predicted to truncate prior to the conserved Forkhead DNA binding domain (Fig. 1A, B Green box) and are therefore likely to represent severe loss of function or null mutations. These alleles are hereafter referred to as foxc1a and foxc1b.
foxc1a single mutants displayed absent or substantially reduced blood circulation throughout the developing trunk, variable pooling of blood within the caudal vein plexus (Fig. 1E black arrowheads) and severe pericardial oedema in comparison to wild type siblings (Fig. 1C, E red arrowheads). Some foxc1a mutants retained the cranial circulatory loop comprised of the heart, ventral aorta, lateral dorsal aortae, primitive internal carotid artery, basilar artery, posterior hindbrain channel and common cardinal vein (not shown). foxc1a; foxc1b double mutants were morphologically similar to foxc1a single mutants (Fig. 1F) and displayed similar circulation abnormalities. Most foxc1a and foxc1a; foxc1b mutants died before 6dpf whereas foxc1b mutants were morphologically normal (Fig. 1D), displayed normal circulation and were homozygous viable and fertile (not shown).
No observable vascular defects were present in foxc1b mutants (not shown), however, foxc1a mutants displayed straightening of intersegmental vessels (ISVs) by 3dpf (Fig. 1I white arrowhead, J yellow arrowheads) in comparison to the normal chevron shape illustrated by ISVs in WT sibs (Fig. 1G white arrowhead, H) as these follow vertical myotomal boundaries. Previous studies have demonstrated functions for foxc1a during somite formation and patterning (Hsu et al., 2015; Li et al., 2015; Topczewska et al., 2001b) and abnormal somite formation is known to influence the migratory path of ECs (Shaw et al., 2006). We therefore examined somite morphology in our foxc1 mutants and identified that while somite morphology was normal in foxc1b mutants consistent with normal ISV patterning in these (Fig. S4B arrowheads), the formation of somites 1-4 was abnormal in foxc1a mutants (Fig. S4A, C, E, F red arrowheads, G) consistent with previous reports (Hsu et al., 2015; Li et al., 2015). In addition, foxc1a mutants displayed a significant increase in somitic angle of posterior somites (Fig S4E, F, H), which likely accounts for straightened ISVs in these mutants. No difference in somite morphology were observed between foxc1a mutants and foxc1a; foxc1b double mutants (Fig. S4C, D). While ISV formation in foxc1a mutants was unremarkable (Fig. 1H, J arrowheads, M), foxc1a; foxc1b double mutants surprisingly displayed a significant increase in frequency of misbranched ISVs within the developing trunk at 5dpf (Fig. 1J, L arrowheads, M). By contrast, both foxc1a and foxc1a;foxc1b double mutants exhibited significantly fewer central arteries (CtAs) within the developing brain at 3dpf in comparison to WT (Fig. 1G, I, K, yellow arrowhead and asterisks, N). Strikingly, these data suggest foxc1a and foxc1b differentially regulate angiogenesis within the developing zebrafish trunk and brain.
foxc1a is required for central artery formation
Since blood circulation was absent throughout the trunk in the majority of foxc1a and foxc1a; foxc1b double mutants but cranial circulation was retained in some embryos, we first examined the hindbrain region where these two circulatory loops interconnect, using confocal time lapse microscopy (Fig. S5). foxc1a mutant embryos displayed delayed angioblast migration during the formation of primordial hindbrain channels (PHBC) (Fig. S5B, Red asterisks). Formation of PHBCs and the basilar artery (BA) were delayed by approximately 3 hours in foxc1a mutants (Fig. 2 A-B’, Supplementary Movies 1 and 2) and increasingly delayed in foxc1a; foxc1b double mutants by approximately 6 hours (Fig. S5C) Formation of common cardinal veins (CCV) (Fig. S6A-D) and lateral dorsal aortae (LDA) (Fig. S6E-G) were also abnormal in foxc1a mutants. In addition, angiogenic sprouting of CtA was greatly reduced in foxc1a mutants at 60hpf (Fig. 2A’, B’ C-F arrowheads, K, L, Supplementary Movie 2). foxc1a mutants also displayed significantly increased frequency of arterial venous connections (AVCs) between the PHBC and the BA or posterior communicating segments (PCS) (Fig. 2M). EC number within basal cranial vessels (PHBC, BA, PCS) were not significantly altered in foxc1a mutants (Fig. 2N), indicating that reduced CtA formation was not due to reduced EC number in basal cranial vessels. Overexpression of full length foxc1a mRNA partially rescued CtA (Fig. 2C-L, arrowheads) and AVC frequency (Fig. 2L, M) in foxc1a mutants, confirming that loss of these vessels was due to mutation of foxc1a.
foxc1a mutants display reduced expression of genes required for normal formation of PHBCs and CtAs
Delayed PHBC formation in foxc1a mutants (Fig. 2, S5) is very similar to that induced by vegfr3/flt4 and vegfc loss of function (Covassin et al., 2006; Hogan et al., 2009b; Shin et al., 2016; Villefranc et al., 2013). By contrast, while Vegfc signalling through Flt4 is dispensable for CtA formation (Hogan et al., 2009b; Le Guen et al., 2014), the VEGF receptor kdrl is required for CtA sprouting from the PHBC but is dispensable for PHBC formation (Bussmann et al., 2011; Habeck et al., 2002). Consistent with this, expression of flt4 was reduced within cranial vessels including the PHBC (Fig. 3A, B black arrowheads) and CCV (Fig. 3A, red arrowheads, B, red asterisks). Furthermore, persistent reductions in flt4 expression were observed in the PHBC in foxc1a mutants after it had formed (Fig. 3C, D black arrowheads). kdrl expression was also substantially reduced in cranial vessels including the PHBC (Fig. 3E, F black arrowheads) and CCV (Fig. 3E red arrowheads, F red asterisks) in foxc1a mutants, which would explain lack of CtAs. In addition, expression of sox7, a transcription factor required for formation of central arteries (Hermkens et al., 2015), was substantially reduced in cranial vessels including the PHBC (Fig. 3G black arrowhead, H asterisk) and the CCV (Fig. 3G’ red arrowheads, H’ red asterisks). Taken together, these data indicate foxc1a contributes to expression of genes which control formation and sprouting of the PHBC and suggest reduced expression of these in foxc1a mutants induce defective PHBC and CtA formation.
Previous studies have reported expression of foxc1a within cranial mesenchyme and hyaloid vasculature (Skarie and Link, 2009). Since foxc1a mutants lacked central arteries (Fig. 2) and expression of genes which co-ordinate their formation (Fig. 3), we examined whether foxc1a was expressed within PHBCs using fluorescent in situ hybridisation and immunocytochemistry (Fig. S7). foxc1a was expressed widely throughout cranial mesenchyme and the developing eye (Fig. S7A, A”) as previously reported (Skarie and Link, 2009). foxc1a also co-localised with the endothelial reporter fli1:EGFP in the mid cerebral vein (MCeV) (Fig S7. A-B”, white arrowheads), CCV (Fig S7. A-B”, white arrowheads), LDA (Fig S7. A-B”, red arrowheads) and PHBC (Fig S7. A-C”, blue arrowheads) at 28hpf and PHBC, CCV and LDA formation are abnormal in foxc1a mutants (Fig. S5, S6). Interestingly, while foxc1a was clearly expressed within ECs of cranial blood vessels and co-localised with an endothelial marker in the PHBC, CCV and LDA foxc1b expression was excluded from these (Fig. S7 D-F”, arrowheads) and displayed perivascular expression (Fig. S7F-F”, blue arrowheads).
foxc1a; foxc1b double mutants display ectopic arterial angiogenesis and reduced venous angiogenesis
Whereas ISVs were straightened in foxc1a mutants, their patterning within the developing trunk was relatively normal (Fig. 1I, J, arrowheads). By contrast, foxc1a; foxc1b double mutants displayed ectopic ISV branching (Fig. 1L, yellow arrowheads, M) suggesting foxc1a and foxc1b interact genetically and negatively regulate angiogenesis in this region. Ectopic ISV sprouting in foxc1a; foxc1b double mutants was observed as early as 28hpf (Fig. 4A-C) and since secondary angiogenesis originating from the posterior cardinal vein (PCV) does not begin until 32hpf (Yaniv et al., 2006) this suggests ectopic sprouts were arterial. Consistent with this, the majority of ectopic vessels within the trunk of foxc1a; foxc1b double mutants at 4dpf expressed the arterial marker Tg(0.8flt1:RFP), hereafter referred to as Tg(flt1:RFP) (Bussmann et al., 2010), (Fig. 4D-F’’’ red arrowheads), indicating these were ectopic segmental arteries (SeA). At 28hpf, foxc1a was expressed within the dorsal aorta (DA), SeAs, PCV and surrounding tissue (Fig. S8A-C), however foxc1b expression was excluded from trunk ECs at this stage (Fig. S8D-F).
Since arterial angiogenesis was increased in foxc1a; foxc1b double mutants while cranial vein formation was inhibited in both foxc1a single and foxc1a; foxc1b double mutants, we examined segmental vein (SeV) sprouting in these embryos (Fig. S9). Quantification of the relative distribution of SeAs and SeVs in foxc1a and foxc1a; foxc1b double mutants using a Tg(fli1a:EGFP;flt1:RFP) background, revealed a 50% reduction in SeV frequency in both mutants (Fig S9A). Since flt4 expression was reduced in cranial veins of foxc1a mutants (Fig. 3A-D) and is known to promote SeV sprouting from the PCV (Hogan et al., 2009b), we examined its expression within the developing trunk vasculature. From 48hpf, flt4 expression was also reduced in SeV sprouts of foxc1a and foxc1a; foxc1b double mutants at 48hpf (Fig. S9B-D, green arrowheads). Collectively, these data indicate venous angiogenesis is impaired in the developing trunk of foxc1a and foxc1a; foxc1b double mutants and suggest foxc1a and foxc1b positively regulate expression of flt4 throughout the developing vasculature.
VEGF signalling is essential for SeA formation in zebrafish (Bahary et al., 2007; Covassin et al., 2006; Habeck et al., 2002; Lawson et al., 2003; Nasevicius et al., 2000; Rossi et al., 2016; Shin et al., 2016; Weinstein and Lawson, 2002) and VEGF receptor expression was reduced in trunk (Fig. S9) and cranial (Fig. 3) vessels of foxc1a mutants. We therefore examined expression of additional components of VEGF signalling in foxc1a and foxc1a; foxc1b double mutants (Fig. 5). Vegfa is the major VEGF isoform which promotes SeA formation in zebrafish and vegfaa expression was normal in foxc1a and foxc1a; foxc1b double mutants (Fig. 5A-C). However, kdrl expression was moderately reduced in the DA (Fig. 5D-F, red arrowheads) and SeAs (Fig. 5D-F, black arrowheads) of foxc1a and foxc1a; foxc1b double mutants, in keeping with our observations of reduced kdrl expression within cranial vessels of foxc1a mutants (Fig. 3E, F). By contrast, flt4, which is preferentially expressed in venous ECs, was normal within the PCV (Fig. 5 G-I, blue arrowheads), but its expression was not downregulated within SeAs in foxc1a; foxc1b double mutants by 24hpf in comparison to foxc1a mutants (Fig. 5G-I, black arrowheads). Expression of the soluble VEGF decoy receptor, sflt1 was reduced in the DA (Fig. 5J-K, red arrowheads) and SeAs (Fig. 5J-K, black arrowheads) of foxc1a mutants and more substantially reduced in foxc1a; foxc1b double mutants (Fig. 5L arrowheads). Given that Kdrl promotes SeA formation (Covassin et al., 2006; Habeck et al., 2002) and its expression was not significantly altered between foxc1a and foxc1a; foxc1b double mutants (Fig. 5E, F), we reasoned that the moderate reduction in kdrl expression in both mutants compared to wild type sibs was unlikely to account for the ectopic SeA sprouting observed in foxc1a; foxc1b double mutants. By contrast, sflt1 expression was substantially reduced in foxc1a; foxc1b double mutants compared to foxc1a mutants (Fig. 5N, O arrowheads) and this receptor has been shown to antagonise sprouting angiogenesis (Krueger et al., 2011; Zygmunt et al., 2011). We therefore overexpressed sflt1 in foxc1a; foxc1b double mutants and quantified frequency of ectopic SeA sprouts (Fig. S10). Frequency of ectopic SeAs were comparable in foxc1a; foxc1b double mutants in the presence or absence of sflt1 overexpression (Fig. S10A-D, arrowheads) indicating reduced sflt1 expression did not account for ectopic sprout formation in double mutants (Fig. S10E). These data are consistent with recent studies which have demonstrated that sflt1 limits sprouting from veins, but not arteries (Matsuoka et al., 2016; Wild et al., 2017).
foxc1a and foxc1b negatively regulate arterial angiogenesis by promoting Notch dependent suppression of vegfc/flt4 signalling
Foxc1 directly activates the promoters of the Notch ligand Dll4 and Notch target Hey2 in vitro (Hayashi and Kume, 2008). Dll4/Notch signalling limits angiogenic sprouting from arteries in vivo (Geudens et al., 2010; Leslie et al., 2007; Siekmann and Lawson, 2007) and suppresses Vegfc/Flt4 signalling in SeAs (Hogan et al., 2009b). Thus a loss of Dll4/Notch signalling in foxc1a; foxc1b mutants might explain the ectopic sprouting. We therefore examined whether Dll4/Notch signalling was disrupted in these mutants. Activity of a dll4 enhancer (Sacilotto et al., 2013) was substantially reduced in developing arteries of foxc1a mutants and was further reduced in foxc1a; foxc1b double mutant SeAs (Fig. 6A-C, black arrowheads) and DA (Fig. 6A-C, red arrowheads). Furthermore, expression of dll4 (Fig. 6D-F) and the Notch target hey2/gridlock (Fig. 6G-I) were substantially reduced in foxc1a; foxc1b double mutant SeAs (Fig. 6D-I, black arrowheads) and DA (Fig. 6D-I, red arrowheads), indicating Dll4/Notch signalling was reduced in arteries. To determine if ectopic sprouting in foxc1a; foxc1b double mutants was Notch dependent we crossed Tg(hs:gal4); Tg(5xUAS-E1b:6xMYC-notch1a) (Scheer and Campos-Ortega, 1999), into the foxc1a/+; foxc1b/+ background and subjected progeny to heat shock between 18-20 somite stage to induce expression of the Notch intracellular domain. Heat shock induction of NICD substantially reduced ectopic SeA sprouting in foxc1a; foxc1b double mutants, thereby demonstrating increased arterial angiogenesis in foxc1a; foxc1b double mutants was due to reduced Notch signalling.
Dll4/Notch signalling suppresses Vegfc/Flt4 signalling in SeAs (Hogan et al., 2009b). Since ectopic SeA sprouting in foxc1a; foxc1b double mutants was Notch dependent (Fig. 6J-O) and flt4 expression was not retained in SeAs in these mutants (Fig. 5I, black arrowhead), we hypothesised that reduced Dll4/Notch signalling in foxc1a; foxc1b double mutant arteries may induce ectopic SeA angiogenesis by enhanced Vegfc/Flt4 signalling. We therefore injected vegfc morpholino (Hogan et al., 2009a) into foxc1a; foxc1b double mutant embryos and quantified frequency of ectopic SeAs (Fig. 6P-T). We observed reduced formation of parachordal lymphangioblasts in vegfc morphants (Fig. 6R, S, asterisks) as previously described (Hogan et al., 2009a; Hogan et al., 2009b) indicating Vegfc function was inhibited. The frequency of ectopic SeAs in foxc1a; foxc1b double mutants was significantly reduced by vegfc knockdown in comparison to control double mutants (Fig. 6Q, S, T, arrowheads). Collectively, our data indicates foxc1a and foxc1b antagonise angiogenesis from arteries by promoting Dll4/Notch-mediated repression of Vegfc/Flt4 signalling. Conversely, foxc1a also promotes venous angiogenesis by positively regulating VEGF receptor expression in veins (Fig. 7). Thus, foxc1a and foxc1b act in concert to balance angiogenesis from arteries and veins by promoting context dependent expression of both pro-and anti-angiogenic genes during development.
Discussion
Here we report a detailed genetic analysis of foxc1a and foxc1b during zebrafish vascular development. Zebrafish foxc1a mutants display vascular defects localised to the cranial vasculature, foxc1a; foxc1b double mutants display additional defects in trunk vasculature, while foxc1b mutant vasculature is normal. This indicates foxc1a can fully compensate for loss of foxc1b during blood vessel development in zebrafish, whereas foxc1b can only partially compensate for loss of foxc1a during formation of trunk vasculature. Furthermore, a single copy of foxc1a is compatible with normal embryonic development following loss of both foxc1b alleles.
Mammals have 2 Foxc genes, Foxc1 and Foxc2, however, Foxc2 has been lost in teleost lineages and Foxc1 has undergone duplication (Fig. S1, S2). Structural analysis suggests FOXC proteins have the same binding specificity (van Dongen et al., 2000) and are likely to regulate the same downstream targets when co-expressed (Hayashi and Kume, 2008). It therefore seems surprising that foxc1a and foxc1b mutants have such different phenotypes given that the forkhead DNA binding domain is 97% identical at the protein level in Foxc1a and Foxc1b (Topczewska et al., 2001b) and is probable these transcription factors have highly similar or identical targets. Our data are consistent with this, since we observed greater reductions in expression of the Foxc targets dll4 and hey2 (Hayashi and Kume, 2008) in foxc1a; foxc1b double mutants than in foxc1a mutants. The morphological difference between foxc1a and foxc1b mutants are therefore not likely to be explained by differential regulation of target genes, but by differential expression of foxc1a and foxc1b during development, for example, foxc1a is expressed in zebrafish ECs whereas foxc1b is excluded from these (Fig. S7, S8).
Mutation of FOXC1 has been proposed to induce human pathologies including cerebral small vessel disease (CSVD) (French et al., 2014). In mice, Foxc1 is expressed in both pericytes and ECs (Kume et al., 2001; Siegenthaler et al., 2013). Conditional knockout of Foxc1 in mouse pericytes induces cerebral haemorrhage and as such Foxc1 function in pericytes has been proposed to maintain integrity of the blood brain barrier in mammals (Siegenthaler et al., 2013). Constitutive Foxc1 knockout mice exhibit similar defects to pericyte-specific Foxc1 knockout mice but with much greater phenotypic severity (Kume et al., 1998; Siegenthaler et al., 2013). Interestingly, EC specific inactivation of Foxc1 in mice does not recapitulate cerebral vascular defects displayed in either pericyte-specific Foxc1 knockout or constitutive Foxc1 KO mice (Mishra et al., 2016), indicating Foxc1 functions non-cell autonomously to promote cerebral blood vessel formation in the mouse. In zebrafish, combined knockdown of foxc1a and foxc1b by morpholino has been reported to induce cerebral haemorrhage in embryos (French et al., 2014; Skarie and Link, 2009) and this was proposed to occur via reduced vascular basement membrane integrity (Skarie and Link, 2009) or inhibition of foxc1a/foxc1b-mediated induction of Pdgf signalling in pericytes (French et al., 2014). While we also observe reduced expression of pdgfrb in foxc1a mutants (not shown) and find both foxc1a and foxc1b are expressed perivascularly in the zebrafish brain prior to mural cell emergence (Fig. S7), we have not observed cerebral haemorrhage in either foxc1a or foxc1a; foxc1b double mutant embryos, even in those which retain anterior circulation. These differences could be due to incomplete gene knockdown in morphants, off target effects of morpholinos, or potential compensatory mechanisms which promote vascular integrity in our mutants. However, recent studies in zebrafish have demonstrated that while pdgfrb is essential for recruitment of mural cells to cranial vessels including CtAs, mural cells are recruited to cranial vessels such as CtAs only after they have formed and cranial angiogenesis is normal in pdgfrb mutants which lack mural cells (Ando et al., 2016). Therefore, since foxc1a mutants display reduced cranial angiogenesis, foxc1a-mediated induction of pdgfrb expression is unlikely to contribute to cranial angiogenesis in zebrafish.
Deletion of Foxc1 in mouse neural crest-derived cells reproduces cerebrovascular phenotypes of global mouse Foxc1 mutants (Mishra et al., 2016), indicating its function in this tissue is essential to co-ordinate cranial angiogenesis, however, neural crest cell specification is dispensable for cranial angiogenesis in zebrafish (Ando et al., 2016; Wang et al., 2014). This indicates substantial divergence between Foxc1-mediated regulation of vascular development between mouse and zebrafish and suggests that in contrast to mouse, endothelial expression of foxc1a is important during cranial angiogenesis in zebrafish. Consistent with this, foxc1a is expressed in cranial vessels (Fig. S7), which are abnormal in foxc1a mutants while foxc1b is excluded from ECs (Fig. S7) and foxc1b expression is not induced in ECs in the absence of foxc1a (not shown). Our analysis of foxc1a and foxc1b mutants suggests foxc1a is a master regulator of cranial angiogenesis which functions in ECs to promote angiogenesis from cranial veins by inducing VEGF receptor expression and other pro-angiogenic factors including sox7. By contrast, foxc1a is dispensable for angiogenesis from arteries, however, loss of both foxc1a and foxc1b induces ectopic angiogenesis from arteries within the developing trunk through reduced Dll4/Notch signalling (Fig. 6, 7). Since foxc1b is not expressed in ECs within the developing zebrafish trunk, but is expressed in neighbouring somitic tissues including sclerotome (Fig. S8) (Topczewska et al., 2001b), this suggests foxc1b induces Dll4/Notch signalling in ECs non-cell autonomously.
In mice, non-cell autonomous anti-angiogenic functions for Foxc1 have been described, for example, Foxc1 expression in neural crest suppresses corneal angiogenesis via a mechanism which antagonises EC response to VEGF signalling (Seo et al., 2012). Counterintuitively, increased angiogenesis induced by Foxc1 KO in neural crest correlated with increased corneal expression of sVegfr1/sFlt1, suggesting Foxc1 regulates competing pro-and anti-angiogenic mechanisms (Koo and Kume, 2013; Seo et al., 2012). Similarly in zebrafish, we find context-dependent functions of foxc1a and foxc1b in suppressing angiogenesis from arteries, while promoting angiogenesis from veins. In contrast to the mouse cornea where Foxc1 expression in neural crest cells suppresses sFlt1 (Seo et al., 2012), foxc1a and foxc1b promote expression of anti-angiogenic sflt1 within the DA in zebrafish (Fig. 5K, L). sflt1 has recently been demonstrated to antagonise sprouting of veins in zebrafish (Matsuoka et al., 2016; Wild et al., 2017), and induction of sflt1 may therefore represent an additional mechanism by which Foxc1 limits venous angiogenesis within the developing trunk, for example, to balance its pro-angiogenic effects on SeV sprouting from the PCV (Fig. S9). Interestingly, we observed reduced expression of kdrl in SeAs of foxc1a and foxc1a; foxc1b double mutants, which is consistent with studies which demonstrate Foxc proteins function co-operatively with ETS transcription factors to induce endothelial gene expression and directly bind enhancers within vegfr2/kdr (De Val et al., 2008). However, these studies also reported reduced sprouting of intersegmental vessels (ISVs) at 24hpf following combined knockdown of foxc1a/b by morpholino (De Val et al., 2008), whereas our mutant analysis showed that despite reductions in kdrl expression following loss of foxc1a and foxc1b, SeA sprouting was not reduced, but enhanced via a Dll4/Notch dependent mechanism (Fig. 6). However, since inhibition of ISV formation was observed at very high doses of morpholino, off target effects cannot be excluded as a potential cause for these differences (De Val et al., 2008). Collectively, our data suggests that foxc1a and foxc1b control angiogenesis throughout the zebrafish vasculature by positively regulating both the pro-and anti-angiogenic inputs of the VEGF-Dll4/Notch negative feedback loop. foxc1a and foxc1b promote VEGF signalling by inducing expression of VEGF receptors within both veins and arteries. Since Notch signalling is active in arteries and actively repressed in veins (Lawson et al., 2001; You et al., 2005) foxc1a and foxc1b mediated induction of Dll4/Notch signalling serves to counteract the pro-angiogenic effects of these transcription factors in arteries but not veins. In doing so, foxc1a/b provide an additional level of transcriptional control to balance arterial and venous angiogenesis within developing vascular beds (Figure 7).
Materials and Methods
Zebrafish strains
All zebrafish were maintained according to institutional and national ethical and animal welfare guidelines. The following zebrafish lines were employed: Tg(fli1a:EGFP)y1 (Lawson and Weinstein, 2002), Tg(-0.8flt1:RFP)hu5333 (Bussmann et al., 2010), Tg(kdrl:HRAS-mCherry-CAAX)s916 (Hogan et al., 2009a), Tg(flk1:EGFP-NLS)zf109 (Zygmunt et al., 2011), Tg(dll4in3:GFP)lcr1 (Sacilotto et al., 2013), Tg (hs:gal4); Tg(5xUAS-E1b:6xMYC-notch1a) (Scheer and Campos-Ortega, 1999), foxc1ash356 and foxc1bsh408.
Bioinformatic analysis of foxc1a and foxc1b synteny
Orthology information for the gene of interest, plus the nearest neighbouring genes were mined from the Ensembl API (accessed October 2015). If no known orthologous gene was present in a species of interest then BLAST was used to identify the closest three genome hits with an e-value less than 1E-10. If a BLAST hit was not within a known gene model then the likelihood of an unannotated gene being present was manually analysed using EST, RNA-Seq sequences and GenScan data. However, in the majority of cases an Ensembl gene was identified and therefore orthology was determined based on synteny.
Generation, selection and genotyping of foxc1a and foxc1b mutant allelesfoxc1ash356 allele
Zinc finger nucleases (ZFN) specific for foxc1a (ENSDARG00000091481) were generated via context dependent assembly (Sander et al., 2011) targeting the following sequence 5’-gTACCCCGCCAGCATGGCGAGGGCa-3’. For the left and right subunits, zinc fingers were added by PCR using pCS2ta3LFok1 and pCS2ta3RFok1 respectively (Ben et al., 2011) as templates, and primers listed in Supplementary Table 1. PCR products were digested with AgeI and self-ligated to generate pCS2foxc1a5-1L and pCS2foxc1a5-1R. Capped mRNA from each plasmid was generated by in vitro transcription and 800-1600pg mRNA injected per embryo such that embryos had an appropriate 30% rate of deformity at 24hpf. To detect potential somatic mutations, genomic DNA extracted from non-deformed embryos was amplified by PCR using foxc1a5-1F and foxc1a5-1R genotyping primers (Supplementary Table 1). Roche Titanium 454 amplicon sequencing identified 90/871 (10%) amplicon molecules included insertions or deletions at the target site. G0 adults derived from embryos injected with ZFN capped mRNA were in-crossed and G1 progeny genotyped by PCR. The sh356 allele contains a 4bp insertion which generates a unique NsiI restriction site (Figure S3). Genotyping was performed as described previously (Wilkinson et al., 2013) followed by restriction fragment length polymorphism (RFLP) analysis. Following digestion with NsiI, the foxc1ash356 allele generates fragment sizes of 119bp and 71bp in comparison to 186bp for WT foxc1a allele.
foxc1bsh408 allele
TALENs specific for foxc1b (ENSDARG00000055398) were designed against the following sequence 5’-ctcgcgCATATGggccgg-3’ to target an NdeI restriction site upstream of the conserved forkhead DNA binding domain. TALENs were assembled using the Golden Gate TALEN and TAL Effector Kit (Addgene, MA, USA) (Cermak et al., 2011) to generate the pFoxc1bTal1L and pFoxc1bTal1R plasmids. Following linearisation of pFoxc1bTal1L/R with NotI, capped mRNA was generated by in vitro transcription and 1500pg TALEN mRNA was injected per embryo. Individual G0 embryos were tested by PCR and RFLP analysis using foxc1bTAL1F and foxc1bTAL1R (Supplementary Table 1) to identify somatic mutations which destroyed the NdeI restriction site. The progeny of TALEN injected G0 adults were incrossed and genotyped to confirm the presence of the SH408 allele, which consists of a 13bp deletion which destroys a unique NdeI restriction site (Figure S3). Following digestion with NdeI, WT foxc1b allele generates fragment sizes of 130bp and 98bp, whereas the foxc1bsh408 allele generates an undigested 215bp fragment.
Plasmid construction and full length mRNA synthesis
foxc1a coding sequence from Danio rerio foxc1a cDNA clone IMAGE: 6789584 was cloned into pCS2+ using EcoRI and XhoI. pCS2-sflt1 was kindly provided by Ferdinand le Noble (Krueger et al., 2011). Capped mRNA was generated using mMessage Machine SP6 Kit (Ambion).
In situ probes
foxc1a and foxc1b in situ probes were generated by PCR amplification using primer sets foxc1aT3F/T7R, foxc1bT3F/T7R and IMAGE: 6789584 (foxc1a) or IMAGE: 5601888 (foxc1b) as templates. PCR products were subsequently transcribed with RNA polymerase. xirp2a was kindly provided by Salim Seyfried (Otten et al., 2012).
Whole-mount in situ hybridisation
Whole mount colorimetric and fluorescent in situ hybridisation was performed as previously described (Wilkinson et al., 2012) (Thambyrajah et al., 2016). GFP expression was detected using Anti-GFP antibody (TP401; amsbio; 1:1000) alongside anti-DIG-POD antibody. Embryos were incubated with AlexaFluor-488 secondary antibody (Invitrogen; 1:500) and imaged.
Microinjection
0.4ng vegfc ATG morpholino 5′-GAAAATCCAAATAAGTGCATTTTAG-3′ (Genetools) (Hogan et al., 2009a) or standard control morpholino (5’-CCTCTTACCTCAGTTACAATTTATA-3’) (Genetools) (Lee et al., 2002) were injected into one cell stage embryos.
Heat shock induction of UAS-NICD
Heat shock was performed at 18s by incubating embryos with pre-warmed E3 at 37°C for 30mins. Embryos were maintained at 28.5°C following heatshock.
Microscopy and image processing
Confocal images were collected using Perkin Elmer Ultraview Vox microscope and lightsheet images were performed using a ZEISS Lightsheet Z.1 microscope. Spinning disk confocal images were analysed using Volocity (PerkinElmer) and Lightsheet images were analysed with ZEN software. Images of embryos following in situ hybridisation was taken using a Leica M165FC and Leica DFC and imaged using Leica Application Suite software (LAS v4.3.0). Image analysis was performed using ImageJ.
Statistical Analysis
All statistical analysis used two-tailed tests and was performed using GraphPad Prism 7. All error bars display the mean and standard deviation. P values, unless exact value is listed, are as follows: *=<0.05, **=<0.01, ***=<0.001, ****=<0.0001.
Acknowledgements
We thank the University of Sheffield aquarium team for excellent care of zebrafish, Michael Moorhouse for technical assistance with amplicon alignments; Cecille Otten, Salim Abdelilah-Seyfried, Ferdinand le Noble and Martin Gering for probes; Yvonne Padberg, Stefan-Schulte-Merker and Wilson Clements for sharing unpublished data. This work was supported by a JG Graves Medical Research Fellowship awarded to R.N.W, Royal Society Research Grant RG120564 awarded to R.N.W, University of Sheffield Faculty Scholarship awarded to Z.J, BHF Infrastructure Grant IG/15/1/31328 awarded to T.J.A.C and R.N.W, and MRC grant MR/N020979/1 supporting T.E and M.L. The authors declare no competing interests.