Divergence of zebrafish and mouse lymphatic cell fate specification pathways

Lymphatic vessels play pivotal roles in tissue fluid homeostasis, immune cell trafficking and the uptake of dietary fats in the small intestine. A failure in lymphatic vessel development (lymphangiogenesis) or in lymphatic function is causative for several inherited or acquired pathological conditions that lead to tissue swelling by accumulation of extravasated fluids (Tammela and Alitalo, 2010; Schulte-Merker et al., 2011).

In mice, combined activity of the transcription factors Coup-TFII (Nr2f2 - Mouse Genome Informatics) (Srinivasan et al., 2010) and Sox18 (François et al., 2008) leads to the polarized expression of Prox1 in a subset of endothelial cells (ECs) within the cardinal vein at embryonic day (E) 9.5. Shortly thereafter, Prox1-positive ECs leave the cardinal vein in a dorsal direction, mediated by Vegfc- and Flt4-driven processes of polarized sprouting and migration, resulting in the formation of the first lymphatic structures in the embryo (Karkkainen et al., 2004; Hägerling et al., 2013). Prox1 expression in lymphatic precursor cells is essential for the initiation of a lymphatic gene expression program, and Prox1 knockout mice lack all lymphatic structures (Wigle et al., 2002). Forced expression of Prox1 is sufficient to confer lymphatic identity to blood ECs, demonstrating the pivotal role of the gene for lymphatic specification (Hong et al., 2002; Petrova et al., 2002). Continuous expression of Prox1 in lymphatic ECs is also indispensable for the maintenance of lymphatic cell fate during later stages of development (Johnson et al., 2008), thus firmly establishing murine Prox1 as the central determining factor of lymphatic identity.

In the zebrafish trunk, the process of vasculogenesis establishes an initial primitive circulatory loop consisting of the dorsal aorta (DA) and the posterior cardinal vein (PCV). Then, in a wave of angiogenic sprouting from the DA (primary or arterial sprouting), a set of ∼30 arterial intersegmental vessels (ISVs) is formed on each side of the embryo. Shortly thereafter, at about 36 hours post-fertilization (hpf), another group of ∼30 sprouts emerges from each side of the PCV. These venous (or secondary) sprouts also migrate dorsally, and about half of them make a stable connection to pre-existing arterial ISVs, thereby remodeling them into intersegmental veins. Venous sprouts that fail to connect to arteries migrate further dorsally towards the midline of the embryo where they populate the region of the horizontal myoseptum as parachordal lymphangioblasts (PLs) (Hogan et al., 2009a), which constitute a pool of lymphatic precursors in the embryonic trunk. These PLs will subsequently migrate away from the horizontal myoseptum [at 2.5 days post-fertilization (dpf)] using arterial ISVs as migration routes to populate the different regions of the trunk, eventually giving rise to the thoracic duct (TD; situated between the DA and PCV), a number of intersegmental lymphatic vessels (ISLVs) in close proximity to arterial ISVs, and the dorsal longitudinal lymphatic vessel (DLLV) (Bussmann et al., 2010).

Previous work has indicated a strong conservation of the genes controlling lymphangiogenesis between zebrafish and mammals. In all organisms examined, mutations in the transmembrane receptor Flt4, its secreted ligand Vegfc, or the more recently discovered gene ccbe1 lead to a block of lymphangiogenesis already at the level of sprouting from the venous endothelium (Schulte-Merker et al., 2011; Koltowska et al., 2013). Although several publications have suggested that Prox1 function in lymphatic specification might be conserved both in amphibians (Ny et al., 2005) and fish, the evidence in the case of the latter has remained open to interpretation, complicated by the existence of duplicated prox1 genes in zebrafish (Del Giacco et al., 2010; Tao et al., 2011). Although expression of prox1a within lymphatic structures has been reported (Yaniv et al., 2006), it remains unclear whether this expression consistently marks all lymphatic structures during different stages of lymphangiogenesis. No mutant allele of prox1a has previously been described and its morpholino-mediated knockdown results in severely malformed embryos, making a conclusive assessment of its requirement for lymphatic development impossible (Küchler et al., 2006). Thus, although mutations in prox1b do not interfere with normal lymphatic development (Tao et al., 2011), the possibility remains that prox1a could indeed be required during lymphatic specification in fish.

Using a novel transgenic reporter line, we show here that prox1a exhibits a dynamic expression pattern in different endothelial compartments during early vascular development. In contrast to the situation in mice, we found that expression of this reporter gene only becomes a specific and reliable marker for lymphatic ECs at later stages of lymphangiogenesis, arguing against a lymphatic specification function during the onset of venous sprouting. In line with this, using a novel targeted allele of prox1a in combination with the previously described prox1b mutation, we show that lymphangiogenesis can proceed in the complete absence of Prox1 in zebrafish. In addition, we show that the functionally related transcription factors Nr2f2 (referred to as Coup-TFII in the following) and Sox18 are also dispensable during lymphangiogenesis. These results indicate that the Sox18/Coup-TFII/Prox1 lymphatic specification code is not conserved in fish, suggesting an alternative mode of lymphatic commitment in this vertebrate class.

To obtain a more precise and sensitive readout for lymphatic structures during development, we created a bacterial artificial chromosome (BAC) reporter line for flt4 [Tg(flt4^BAC:mCitrine), flt4:mCit hereafter]. Similar to previous observations in mice (Kaipainen et al., 1995), we found that flt4 is initially expressed in all blood ECs but becomes progressively restricted to venous and lymphatic endothelial cells (LECs) after 26 hpf (supplementary material Fig. S1A,B) and 36 hpf (supplementary material Fig. S1C-F), respectively. In contrast to Flt4 expression in mice (Hägerling et al., 2013), flt4:mCit expression is not lost in venous structures, making it a lymphatic-enriched, but not lymphatic-specific reporter at 5 dpf. In pan-endothelial transgenic lines such as fli1a:eGFP, the TD is the only lymphatic structure that is readily distinguishable from blood vessels. By comparison, the flt4 reporter line also reliably highlights more delicate lymphatic structures such as ISLVs or the DLLV in the trunk (supplementary material Fig. S1G,H) as well as the facial lymphatic system (Fig. 1A,B). Combined with the arterial-specific flt1^enh:tdTom reporter (Bussmann et al., 2010), all lymphatic, venous and arterial structures in the embryo can be easily distinguished (see below).

To assess whether prox1a might represent the functional Prox1 ortholog in zebrafish and would therefore be expressed specifically in LECs, we generated different prox1a BAC reporter lines that were analyzed in combination with the aforementioned flt4:mCit line. As simple fluorophore expression under the control of the prox1a^BAC promoter resulted in extremely weak signals (not shown), we took advantage of the enhancing effect of an optimized Gal4 variant [KalTA4 (Distel et al., 2009)] in a prox1a^BAC:KalTA4, UAS:tagRFP expression cassette. Consistent with previously published data (Glasgow and Tomarev, 1998; Thisse and Thisse, 2005; Pistocchi et al., 2009), the transgene marked a wide range of tissues comprising the lens, retina, liver, neuromasts and myotome (Fig. 1A). To identify expression domains masked by the myotomal expression in full z-projections, partial z-projections of the trunk were analyzed, revealing expression in all lymphatic structures within the head and trunk, where the signal colocalized with flt4:mCit expression at 5 dpf (Fig. 1B-D). In contrast to previously published lymphatic markers such as stabilin1:YFP or lyve1:DsRed2 (Hogan et al., 2009a; Okuda et al., 2012) or the flt4:mCit line, at 5 dpf the prox1a reporter exhibits no additional expression domains in other endothelial compartments, such as the PCV, and hence reflects the first truly LEC-specific marker at this stage of development in zebrafish. Taking advantage of this feature, we found that the anterior part of the TD, which connects the head lymphatics with the TD in the trunk, comprises a bilateral structure above the developing swim bladder. The TD splits up near the sixth intersegmental vessel pair and runs in close proximity to the correspondingly bilateral PCVs (Fig. 1E,F; supplementary material Fig. S2A), finally connecting to the facial lymphatic network (Okuda et al., 2012) and ultimately the common cardinal veins (Fig. 1G,H).

Next, we investigated whether zebrafish prox1a expression would, like its murine counterpart, mark LECs at earlier stages, possibly even within the venous endothelium of the PCV at a time point prior to venous (secondary) sprouting. At 32 hpf, shortly before the emergence of the first sprouts from the PCV, prox1a could indeed be detected in a subset of venous cells (Fig. 1I,J) and subsequently in sprouting venous ECs (Fig. 1K). To determine whether prox1a might exclusively mark secondary sprouts that will give rise to lymphangioblasts, expression was analyzed at 44 hpf, a stage when PL cells have extended to the horizontal myoseptum and venous destined sprouts have already established a connection to the respective ISV. We found prox1a reporter expression only in a proportion of PLs (Fig. 1L; 17/78 PLs), indicating that not all lymphatic precursor cells are positive for prox1a at this stage. Equally important, the transgene also transiently marked individual venous ISVs (in 12/21 imaged regions) (Fig. 1L,O,R), suggesting no clear correlation between prox1a expression and the fate of secondary sprouts.

Earlier during development, prox1a reporter activity could also be detected in ECs of developing arterial ISVs (Fig. 1M,N) and in particular of the DA (Fig. 1P,Q), as well as in undefined cells within the axial vessels before the onset of circulation (supplementary material Fig. S2B-E). In addition, transgene expression highlighted a subpopulation of ECs in the caudal vein region already at 26 hpf, an expression domain that could be detected up to day 3 (supplementary material Fig. S2D-G). Probably as a consequence of this early and relatively wide expression within the caudal vein, prox1a-positive venous ISVs and PL sprouts were more frequently detected in this area at 48 hpf (supplementary material Fig. S2G).

Because prox1a is initially expressed within different endothelial subpopulations, we wanted to establish the earliest stage when prox1a expression would reliably and exclusively mark the lymphatic part of the vasculature. We found that, at least at 3.5 dpf, when LECs leave the horizontal myoseptum region and form the different lymphatic structures within the trunk, all LECs were highlighted by expression of the reporter, which was also the case in the head region (Fig. 1S-U). However, because the strong myotomal signal did not allow a conclusive assessment of prox1a expression in lymphangioblasts at the level of the horizontal myoseptum, it cannot be excluded that the gene was already expressed in all LECs while still aligning with the horizontal myoseptum.

Taken together, expression of prox1a becomes a reliable and very useful maker for LECs only at later stages of lymphangiogenesis. Therefore, the expression data presented here would be in line with a prox1a function in specifying PLs at the level of secondary sprouts but not at the level of the vein, as is the case in mice.

To assess the role of Prox1a in lymphangiogenesis, we generated a loss-of-function allele for prox1a using zinc-finger nucleases (supplementary material Fig. S3A). The prox1aⁱ²⁷⁸ allele harbors a 10 bp deletion in the first coding exon, resulting in a frame-shift after amino acid 149, and a premature stop codon after additional four amino acids (Fig. 2A). Homozygous mutants are devoid of full-length Prox1a protein as assessed by immunostaining of slow muscle fibers at 30 hpf (Fig. 2B,C). The overall appearance of homozygous prox1aⁱ²⁷⁸ mutants was normal until between days 4 and 5, at which stage mutant larvae started to develop severe edema around the gut and the eye (Fig. 2D,E). At 5.5-6 dpf, most mutants had impaired blood circulation and showed signs of tissue necrosis. Importantly, timing as well as severity of edema was different from ccbe1, vegfc and flt4 mutant scenarios: although completely lacking lymphatic structures, mutants of each of those genes only develop mild edema starting from 5-6 dpf. The later onset of edema in those lymphatic mutants presumably results from the lymphatic system beginning to perform its drainage function at early day 5 (Karpanen and Schulte-Merker, 2011); thus, atypical tissue fluid accumulations can only arise from this time point. It follows that the edema observed in prox1a mutants does not indicate impaired lymphatic drainage function.

In Prox1-null mutant mice, the failure of LEC progenitors to leave the cardinal vein results in the absence of lymphatic structures (Yang et al., 2012) and even heterozygous embryos die shortly after birth with dysfunctional lymphatics and a lack of lymphovenous valves (Harvey et al., 2005; Srinivasan and Oliver, 2011). If prox1a was equally important in specifying lymphatic cell fate in zebrafish, homozygous mutants should be deficient in lymphatic structures. However, when analyzing the appearance of PLs at the horizontal myoseptum at 2 dpf, we found that prox1aⁱ²⁷⁸ mutants show only a marginal reduction in the number of PLs (Fig. 2F-H). As it is possible that this initially mild defect results in more dramatic effects at later stages, we evaluated whether the formation of the TD was impaired. At 5 dpf, no defects in heterozygous embryos and only minor defects in a proportion of homozygous mutants were evident (Fig. 2I-K). On average, homozygous mutants lacked the TD only in 1.4 of the first ten somites, a mild phenotype that could also reflect a secondary defect caused by the early edema formation and overall impaired appearance of mutants at 5 dpf (Fig. 2L). Hence, expression of prox1a in the venous and lymphatic endothelium as revealed by the reporter gene described above is not essential for the specification of lymphatic structures. To confirm that the loss of prox1a did not affect secondary sprouts that are committed to a venous fate, the ratio of arteries and veins as well as the total number of ISVs were quantified. Overall, no defects on the blood vasculature could be observed and the arterio-venous ratio was also unaffected in mutants (Fig. 2I-K,M-O). We also found that expression of the prox1a reporter line was present in homozygous prox1aⁱ²⁷⁸ mutants, indicating that, in contrast to the situation in mice (Srinivasan et al., 2010), Prox1a protein is not required for maintaining its own lymphatic expression (supplementary material Fig. S4A-D). Furthermore, considering that the prox1a reporter represents a specific lymphatic marker at 5 dpf, the unaltered expression of both flt4:mCit and prox1a:KalTA4,UAS:tagRFP in the TD of homozygous prox1a mutants argues for correct lymphatic specification of LECs even in the absence of functional Prox1a protein. Given the specific zygotic expression pattern of prox1a in different subsets of ECs, our finding that Prox1a protein was undetectable in slow muscle fibers of prox1aⁱ²⁷⁸ mutants prior to secondary sprouting as well as the fact that maternally expressed prox1a transcripts have not been identified (Harvey et al., 2013), it seems unlikely that the lack of zygotic phenotype can be attributed to maternal rescue. We therefore conclude that prox1a function is not essential for any aspect of lympho-venous sprouting and that the gene is dispensable for lymphatic cell fate determination in zebrafish.

Although neither of the two prox1 orthologs appears to be essential for zebrafish lymphatic development (Tao et al., 2011) (this study), it remains possible that they function redundantly. We therefore generated double mutants for prox1aⁱ²⁷⁸ and the previously characterized loss-of-function allele prox1b^SA0035 (Tao et al., 2011). In all allelic combinations, including double mutant embryos, PLs were observed at the horizontal myoseptum (Fig. 3A-C), indicating that even the absence of all Prox1 function fails to block the appearance of lymphatic-fated secondary sprouts. Subsequent analysis of double mutants for the formation of TD at 5 dpf demonstrated that a moderate but significant reduction in the length of the TD within the first ten segments was evident (Fig. 3D-H); whether this is a specific effect of prox1 deficiency, or whether this phenotype represents a secondary consequence of more general defects within the embryo (see Fig. 2E) is difficult to assess. The presence of PL cells, facial lymphatics (supplementary material Fig. S5A-D) and TD tissue in double mutants, however, clearly shows that lymphatic commitment is not governed by zygotic expression of the Prox1a and Prox1b transcription factors. It follows that lymphatic specification in zebrafish must be achieved in a different way, possibly through the broadened activity of another functionally related transcription factor.

The principal factor that promotes venous cell identity in many vertebrate embryos is the transcription factor COUP-TFII (NR2F2), which suppresses expression of arterial genes in venous ECs (You et al., 2005). As lymphatics are venous-derived, loss of endothelial Coup-TFII expression in mice not only causes venous specification defects but also results in a lack of LECs (Srinivasan et al., 2007). Coup-TFII has been reported to be required for the initiation of Prox1 expression within the cardinal vein by direct binding to its promoter region (Srinivasan et al., 2010) and subsequently also for the maintenance of Prox1 expression in future LECs during early lymphangiogenesis. The latter function is thought to be dependent on a direct physical interaction between Coup-TFII and Prox1, and heterodimers have been reported to act as co-regulators of several lymphatic lineage specific genes such as FLT4 in cultured LECs (Lee et al., 2009). More recently, vascular defects including a reduction in PL number and aberrant TD formation has been reported to result from morpholino knockdown of zebrafish coup-TFII (Aranguren et al., 2011). We therefore wondered whether lymphatic lineage specification in fish might be entirely regulated by coup-TFII instead of prox1 genes. To explore this possibility, we employed TALEN constructs targeting the first exon of coup-TFII and generated a 1 bp insertion allele nr2f2^hu10330 (coup-TFII^hu10330 hereafter). Sequencing of cDNA from coup-TFII^hu10330 mutant embryos did not reveal alternative transcripts lacking the 1 bp insertion (data not shown). This suggests that the allele represents a loss-of-function situation because the insertion results in a frame-shift and premature stop codon 5′ to both the nuclear receptor-DNA-binding domain and the ligand-binding domain (Fig. 4A).

Homozygous coup-TFII^hu10330 mutants were viable beyond 6 dpf (Fig. 4B,C) and developed a normal blood vasculature without any sign of arterio-venous identity defects in the main trunk vessels. In addition, when scoring TD formation at 5 dpf, we observed only negligible defects and the majority of mutants were indistinguishable from siblings (Fig. 4D-F). As the ratio of intersegmental arteries and veins was also unaffected in coup-TFII mutant embryos (Fig. 4G), these results demonstrate that zygotically expressed coup-TFII is not essential for venous specification and lymphatic development.

Another transcription factor that has been implicated in lymphatic development in mice and humans is Sox18 (Irrthum et al., 2003). Mice lacking functional Sox18 develop edema in certain genetic backgrounds (François et al., 2008) because of a failure in LEC development. Sox18 is expressed in a subset of cardinal vein cells that subsequently initiate Prox1 expression. In vitro experiments further showed that Sox18 can activate Prox1 expression by direct binding to its promoter region, placing Sox18 upstream of Prox1 during lymphatic specification (François et al., 2008). More recent morpholino knockdown studies as well as overexpression experiments employing a dominant-negative mouse Sox18 variant in zebrafish suggested a specific requirement for sox18 during sprouting of lymphatic-fated secondary sprouts from the PCV. In contrast to mouse, however, zebrafish sox18 is expressed in both the DA and PCV in a non-polarized fashion during venous sprouting (Cermenati et al., 2013), which is counterintuitive in the context of lymphatic specification.

We generated a mutant with a 1 bp insertion 5′ to the HMG-box encoding sequence (Fig. 4H). This allele is predicted to encode a truncated protein that contains neither the essential HMG nor the trans-activation domains (reviewed by Downes and Koopman, 2001), and as no alternative transcripts could be detected in mutants (data not shown) we consider it to represent a loss-of-function allele. Loss of Sox18 function did not lead to any obvious differences in the formation of arterial and venous ISVs, which is in line with previous morpholino knockdown data (Cermenati et al., 2008; Herpers et al., 2008; Pendeville et al., 2008; Cermenati et al., 2013). At 5 dpf, the overall appearance of homozygous sox18^hu10320 mutants was indistinguishable from siblings (Fig. 4I,J) and when scored for the presence of the TD, no lymphatic or blood vascular phenotypes were detectable (Fig. 4K-M). Accordingly, the quantification of PLs at the horizontal myoseptum at 48 hpf did not reveal any significant differences (Fig. 4N), demonstrating that Sox18 is dispensable for venous sprouting and lymphatic cell fate determination.

It has been reported that overexpression of Prox1 in blood ECs in vitro is sufficient to shift their gene expression profile towards an LEC phenotype (Hong et al., 2002; Petrova et al., 2002). Similar results were also obtained from mouse models, where the forced expression of Prox1 in blood ECs led to an LEC-like gene expression pattern, severe edema formation, and embryonic lethality at E13.5 (Kim et al., 2010). Interestingly, the ability of Prox1 to transform blood into lymphatic ECs seems to be restricted to the venous endothelium in vivo (Kim et al., 2013), indicating that the additional expression of only one transcription factor is sufficient to differentiate a venous EC into an LEC. Although we showed here that mutations in the zebrafish orthologs of Prox1 and two other central genes that govern and maintain lymphatic commitment in mammals did not block lymphatic specification, we wondered whether prox1a expression might nevertheless be sufficient to force blood ECs into a lymphatic cell fate. Therefore, we aimed to mis-express prox1a in blood ECs and injected an UAS:prox1a construct into a flt4:Gal4FF driver-line (Fig. 5A) which drives expression from a 3.8 kb flt4 promoter fragment from Medaka (Deguchi et al., 2012). Upon transient injection of the UAS-construct, overexpression of prox1a transcripts was verified by whole-mount in situ hybridization (ISH). As expected, embryos injected with the UAS-construct expressed prox1a in a mosaic pattern in venous and arterial ECs at considerably higher levels than the endogenous endothelial expression in uninjected siblings at 32 hpf (Fig. 5B-E). We checked at 38 hpf (data not shown) and 48 hpf whether this prox1a mis-expression would have an influence on the behavior of positive ECs (marked by the co-expression of mRFP) during venous sprouting. However, neither the overall number of secondary sprouts nor the positioning of RFP-positive cells within the vasculature indicated a differential cell behavior upon prox1a overexpression (Fig. 5F,G,J,K). Furthermore, and in contrast to the mouse experiments, continuous mis-expression of prox1a did also not result in any vascular or lymphatic defects at later stages of lymphangiogenesis. In fact, at 5 dpf no differences in vascular morphology or in the expression pattern of the flt4:Gal4FF;UAS:GFP marker could be appreciated and cells expressing the construct ended up in all endothelial compartments (Fig. 5H,I,L,M; data not shown). Together, these results indicate that zebrafish prox1a is not only dispensable for lymphatic specification, but is also unable to commit ECs to a lymphatic fate.

Specification of the lymphatic lineage in mice is intimately linked to the restricted expression of the transcription factor Prox1 in a subpopulation of cardinal vein cells, and Prox1 function is essential for future LECs to migrate out of the cardinal vein (Yang et al., 2012). In the absence of mutants for Prox1 orthologs in other vertebrate species, a stringent assessment of conserved Prox1 function during lymphangiogenesis has been hampered. Nevertheless, expression studies in, for example, Xenopus (Ny et al., 2005) and zebrafish (Yaniv et al., 2006) suggest that there is evolutionary conservation.

Previous morpholino-knockdown strategies for zebrafish prox1a did not provide a fully conclusive picture and we had previously pointed out that different morpholinos directed against prox1a result in developmental anomalies, which we interpreted as being unspecific (Küchler et al., 2006). Hence, we generated a prox1a loss-of-function model to address conclusively the role of prox1-like genes for lymphangiogenesis in fish. Our results demonstrate that prox1a expression is not essential and also not sufficient for specification of LECs. Even the combined loss of prox1a and prox1b does not inhibit lymphatic development.

In combination with our finding that coup-TFII and sox18 are also not essential for lymphangiogenesis to occur, this suggests that lymphatic specification must be achieved in a different way and that the Coup-TFII/Sox18/Prox1 function has been evolving only in higher vertebrates or has been eliminated in the zebrafish/teleost lineage. How can one explain the strict requirement for Prox1 in mice, while in fish the Coup-TFII/Sox18/Prox1 code seems dispensable?

In mice, angiogenesis and lymphangiogenesis are two temporally separated processes. Initially, starting at E8.0, a network of intersegmental arteries and veins is formed by angiogenesis (Fig. 6A1) (Walls et al., 2008). In a second step, a subpopulation of cardinal vein cells is specified to the lymphatic lineage by Prox1 expression at E9.5 (Fig. 6A2), making only those cells responsive to the VegfC signal. These lymphatic precursors become motile then and leave the epithelial layer of the vein to form the first lymphatic structures (Fig. 6A3,4).

In fish, however, the timing of events is different: angiogenesis begins with a first wave of VegfA- and Vegfr2-governed sprouting from the DA, giving rise to a set of intersegmental arteries (Fig. 6B1). Subsequently, a second set of sprouts arises exclusively from the PCV (Fig. 6B2). The emergence of these venous sprouts that eventually will give rise to both intersegmental veins and lymphatic precursors, depends on the Vegfc/Flt4 pathway and on Ccbe1, indicating that in zebrafish the mechanisms that drive the initial migration of both subpopulations is identical. This might also explain why genes that at some point specifically mark LECs in mice (such as Flt4 or Lyve1) are expressed within the whole venous endothelium in zebrafish. It is only slightly later, when secondary sprouts reach the level of the arterial ISVs that both populations display a differential behavior, suggesting that only from this point in time are lymphatic and venous fates separating (Fig. 6B3,4). It is therefore plausible that it is not an intrinsic mechanism but rather external cues, possibly provided by the ISVs themselves, that control which sprout will connect to an ISV and which will migrate further to the horizontal myoseptum. In fact, the only known signaling pathway so far that has been shown to affect the cell fate decision made by secondary sprouts is Dll4/Notch. Knockdown of dll4 leads to a dramatically enhanced number of venous ISVs and only a small number of PL cells, suggesting that the fate of secondary sprouts is strongly shifted towards venous identity (Geudens et al., 2010). Whether this requirement for Notch signaling is artery intrinsic (dll4 and notch1b are both expressed in arterial ECs) or reflects a signaling interaction between ISV and secondary sprouts requires further analysis, but in either case arterial ISVs directly or indirectly influence the fate decisions made by venous sprouts.

In summary, the scheme depicted in Fig. 6 offers an explanation as to why the Coup-TFII/Sox18/Prox1 signaling axis is required in mice, but dispensable in zebrafish: in mice (and probably other vertebrate classes) Prox1 function specifies LEC fate within a subset of venous ECs within the cardinal vein. In zebrafish (and probably all teleosts), this specification step is not required within the endothelial epithelium of the cardinal vein. Our findings therefore reveal an astonishing difference in the earliest step of lymphangiogenesis between vertebrates. In the absence of a full understanding of the exact point in time when LECs become specified in zebrafish, it remains difficult to appreciate fully the degree of similarity between mice and other vertebrates at the level of genetic control and cellular behavior. However, other steps of lymphangiogenesis, such as the migration of future LECs away from the major vein, are conserved in vertebrates, as evidenced by the conserved functions of vegfc (Karkkainen et al., 2004; Ny et al., 2008; Villefranc et al., 2013), flt4 (Jeltsch et al., 1997; Ny et al., 2008; Hogan et al., 2009b) and ccbe1 (Hogan et al., 2009a; Bos et al., 2011).

Strains were maintained under standard husbandry conditions. Animal experiments have been performed according to the rules of the Animal Experimentation Committee (DEC) of the KNAW and the A^*STAR Biological Resource Centre Institutional Animal Care and Use Committee (IACUC Project #110638). The following published transgenic lines have been employed in this study: Tg(flt1^enh:tdTomato) (Bussmann et al., 2010), Tg(fli1a:eGFP)^y1 (Lawson and Weinstein, 2002), Tg(kdrl:HRAS-mCherry)^s916 (Hogan et al., 2009a).

The Tg(flt4^BAC:mCitrine)^hu7135 line was generated from BAC DKEY-58G10 and the Tg(prox1a^BAC:KalTA4-4xUAS-E1b:uncTagRFP)^nim5 from BAC DKEY-5J3 following standard BAC recombineering procedures (Bussmann and Schulte-Merker, 2011). For the generation of the flt4 promoter construct Tg(flt4:Gal4FF)^hu9236, a previously reported 3.8-kb promoter fragment (Deguchi et al., 2012) was amplified from Medaka genomic DNA (supplementary material Table S1) and cloned into the miniTol2 vector 5′ of the Gal4FF coding sequence. In case of the UAS:prox1a mis-expression construct, the prox1a cDNA was placed behind a 5xUAS cassette in the pT2A vector followed by an IRES sequence and a mTurquoise-NLS cassette including PolyA sequences. Because mTurq-NLS expression behind the IRES turned out to be very low, an additional 5xUAS:mRFP cassette was inserted into the plasmid to identify Gal4FF-positive cells harboring the prox1a overexpression construct after injection. BAC DNA (100 pg/embryo) or plasmids (25 pg/embryo) were co-injected with Tol2 transposase mRNA (25 pg/embryo) into one-cell-stage embryos and the progeny was screened for germline transmission.

For the generation of prox1a mutants, plasmids encoding zinc-finger nucleases targeting the locus were obtained from ToolGen (Korea). The zinc-finger target sites in the first exon of prox1a were: 5′-TGAGATGGAGAG-3′ and 5′-GGTCATGGAGGG-3′ (supplementary material Fig. S3). TALEN-mediated genome editing for the generation of coup-TFII and sox18 mutants was performed as described (Cermak et al., 2011; Bedell et al., 2012). The TALEN binding sites in sox18 exon1 were: TAL1, 5′-TGCCTGGGTCTGGAAC-3′; TAL2, 5′-TGGCCTCCGCTGCTGTT-3′. For coup-TFII, the TALEN recognition sites were TAL1, 5′-TCCGACCCCTCAGACACCCGT-3′ and TAL2, 5′-AACAATAACAACACACAGTCA-3′.

The prox1aⁱ²⁷⁸, sox18^hu10320 as well as the coup-TFII allele nr2f2^hu10330 were genotyped by KASPAR using the primers indicated in supplementary material Table S1. KASPAR genotyping of the prox1b^SA0035 allele was performed as described (Tao et al., 2011).

Antibody staining using the following antibodies was performed as described (Elworthy et al., 2008): mAb F59 [anti-slow myosin heavy chain-1, Developmental Studies Hybridoma Bank (DSHB); 1:100], rabbit anti-Prox1 (AB5475, Chemicon, USA; 1:5000), goat anti-rabbit Alexa488 and goat anti-mouse Alexa546 (Invitrogen; 1:1000). ISH was carried out as described previously (Schulte-Merker, 2002).

Confocal imaging was performed on living embryos embedded laterally (unless otherwise stated) in 0.5-1% low melting point agarose (Invitrogen) on Leica SPE and SP8 microscopes. Brightfield images were taken on an Olympus SZX16 Stereomicroscope. Images were processed using Adobe Photoshop CS5.1 and Fiji (http://fiji.sc/Fiji). Stitching of composite pictures was performed using Leica LAS AF software or Adobe Illustrator CS5.1.

We thank M. Witte and Audrey Iyu for technical assistance; D. Schulte, S. van de Pavert and F. Kiefer for critical comments on the manuscript; and A. de Graaff and the Hubrecht Imaging Centre for supporting the imaging.