The zebrafish common cardinal veins develop by a novel mechanism: lumen ensheathment

To supply the body with oxygen and nutrients, a complex vascular network of arteries, veins and capillaries is formed, with individual blood vessels being specifically adapted to their respective functions. Differences between these individual blood vessel types are reflected in their specific cellular architecture and physiological function, as well as in their differential development and by marked differences in gene expression profiles (reviewed by Dyer and Patterson, 2010; Atkins et al., 2011).

Formation of the zebrafish vasculature (reviewed by Ellertsdóttir et al., 2010) commences at 12 hours post-fertilization with the specification of angioblasts within the lateral plate mesoderm. These migrate to the embryonic midline, where they coalesce into a vascular cord (Jin et al., 2005), giving rise to the dorsal aorta (DA) and the posterior cardinal vein (PCV) (Torres-Vázquez et al., 2003). Recently, it was proposed that in the caudal region of the zebrafish embryo individual venous-fated endothelial cells (ECs) leave the initial vascular cord and migrate ventrally to form the vein, while the remaining ECs differentiate into the DA (Herbert et al., 2009). However, a detailed analysis of cardinal vein formation in zebrafish or mice has yet to be carried out.

To form definitive lumenized vessels, ‘budding’, ‘cord hollowing’, ‘cell hollowing’ and ‘cell membrane invagination’ have all been described as mechanisms of vascular lumen formation (Baer et al., 2009; Iruela-Arispe and Davis, 2009; Ellertsdóttir et al., 2010; Herwig et al., 2011). During DA formation in both zebrafish and mice, ECs form a lumen by a process called cord hollowing, which is a mechanism by which a set of compact cells undergoes shape changes to create a central lumen (Jin et al., 2005; Strilić et al., 2009). It is generally believed that the cardinal veins form their lumen by the same mechanism.

Further angiogenic growth depends on EC migration and proliferation. Proteins of the vascular endothelial growth factor (Vegf) family, especially Vegfa, have been implicated in regulating different steps of EC development, including specification, migration and proliferation (reviewed by Veikkola and Alitalo, 1999; Carmeliet, 2000). Although Vegfa is important for artery formation and specification, it appears not to be involved in venous development (Lawson and Weinstein, 2002b; Wiley et al., 2011). By contrast, Vegfc signaling, mainly via Vegf receptor 3 (Flt4) (Joukov et al., 1996; Lee et al., 1996; Hogan et al., 2009b), plays a major role in regulating the development of venous sprouts from the PCV and of the lymphatic vasculature (Karkkainen et al., 2004; Küchler et al., 2006; Yaniv et al., 2006).

In the present study we have examined the development of the common cardinal veins (CCVs). The CCVs are located in the anterior trunk of the embryo, where they collect all the venous blood and transport it to the heart. Whereas in the posterior region of the zebrafish the DA and the PCV are established as single tubes in the midline below the notochord, in the anterior part of the embryo these vessels split into two arteries [the lateral dorsal aortae (LDA)] and two cardinal veins, located on both sides of the midline (Fig. 1A). Emerging from these bilateral cardinal veins, the CCVs grow around either side of the yolk cell and converge cranioventrally at the sinus venosus of the heart, where they connect with the endocardial ECs (Isogai et al., 2001).

We have analyzed the development of the CCVs by in vivo time-lapse studies and cell tracing, as well as detailed gene expression analysis and loss-of-function studies. We found that the development of the CCVs represents a completely novel mechanism of vessel formation, during which a distinct angioblast population forms the CCVs by aligning along the inner surface of an existing luminal space. Subsequently, the open CCV tubes extend by Cadherin 5-dependent EC migration and eventually enclose the entire lumen. Furthermore, we show for the first time that EC proliferation is controlled by Vegfc, which is provided by circulating blood cells.

Zebrafish (Danio rerio) were maintained in a recirculating aquaculture system under standard laboratory conditions (Westerfield, 1993). All zebrafish experiments comply with the institutional and national animal welfare laws, guidelines and policies. Embryos were staged by hours post-fertilization (hpf) at 28.5°C (Kimmel et al., 1995).

The previously described zebrafish lines were Tg(kdrl:EGFP)^s843 (Jin et al., 2005), Tg(fli1a:EGFP)^y1 (Lawson and Weinstein, 2002a), Tg(gata1:dsRed)^sd2 (Traver et al., 2003), Tg(kdrl:HsHRAS-mCherry)^s896 (Chi et al., 2008), Tg(fli1ep:GAL4FF)^ubs4 (Herwig et al., 2011), Tg(UAS:KAEDE)^rk8 (Hatta et al., 2006), Tg(etsrp:EGFP)^ci1 (Proulx et al., 2010) and Tg(krt4:EGFP-caax)^e112 (Krens et al., 2011). cdh5^ubs8 will be described in detail elsewhere (H.-G.B. and M.A., unpublished). The vegfc^hu6410 allele encodes a stop codon at amino acid position 107 (L107Stop) and will be described in detail elsewhere (T.K. and S.S.-M., unpublished).

pT2UAS:lifeact-GFP was generated by cloning the coding sequence of lifeact-GFP (Riedl et al., 2008) into pT2UAS (Asakawa et al., 2008). pT2UAS:lifeact-GFP plasmid DNA (50 pg) was injected together with tol2 transposase mRNA (250 pg). Three founder fish were identified and that with the brightest GFP expression was used in this study.

Microinjections of mRNA [synthesized using the Sp6 mMessage mMachine Kit (Ambion)] or morpholinos (MOs) were performed as previously described (Nasevicius and Ekker, 2000). vegfc mRNA (Hogan et al., 2009a), H2B-cherry mRNA (Santoro et al., 2007) or sFLT4 mRNA (Mäkinen et al., 2001) were injected at 200 pg.

MOs targeting tnnt2a (Sehnert et al., 2002), mitfa (Robu et al., 2007), sox32 (Chung and Stainier, 2008), gata1 (Galloway et al., 2005) and pu.1 (spi1b - Zebrafish Information Network) (Rhodes et al., 2005) have been described previously. mini-Ruby (tetramethylrhodamine, Invitrogen) was injected (0.1 ng) into the YSL as described previously (Sakaguchi et al., 2001).

Single in situ hybridizations were performed as described (Schulte-Merker, 2002; Thisse and Thisse, 2008); after staining, fixed embryos were further washed twice in methanol and cleared using a 2:1 mixture of benzylbenzoate:benzylalcohol. Double fluorescent in situ hybridizations were performed as described (Schoenebeck et al., 2007).

The following probes were synthesized as published: flt4 (Lawson et al., 2001), flt1 (Bussmann et al., 2007), cdh5 (Larson et al., 2004) and vegfc (Hogan et al., 2009a).

etsrp (also known as etv2) was amplified from cDNA of 19-somite stage embryos using the following primers (5′-3′): etsrp forward, GAAAGGCCCAAGTCACAGAG; etsrp reverse, GAGTTCACTATCTGATGTCAAACCA. The T7 promoter was added to the 5′-end of the reverse primer in a second round of amplification: T7-etsrp reverse, GTAATACGACTCACTATAGGGAGTTCACTATCTGATGTCAAACCA.

Embryos were manually dechorionated and mounted either in a 1% agarose embryo array (Megason, 2009) or in 0.3% agarose (with subsequent removal of agarose from the head/tail region). Medium and agarose were supplemented with a final concentration of 19.2 mg/l Tricaine and 30 mg/l phenylthiourea, respectively. For time-lapse imaging, embryos were kept in a 28.5°C heated chamber surrounding the microscope stage. All fluorescent images were acquired using an upright Leica Sp5 DM 6000, an inverse Leica Sp5 DM 6000, or a Zeiss LSM 780 confocal microscope. Confocal stacks and confocal movies were assembled using Imaris software (Bitplane). Color coding of images was performed using ImageJ (NIH).

For photoconversion, Kaede-positive double-transgenic embryos [Tg(fli1ep:GAL4FF)^ubs4;Tg(UAS:KAEDE)^rk8] were photoconverted by scanning the anterior region of the embryo with a 405 nm diode laser in 3 μm steps using a Leica Sp5 DM 6000 inverse confocal microscope with a 10× objective. Photoconversion was monitored by the change in the Kaede emission wavelength from 518 nm (green channel) to 582 nm (red channel).

Embryos were incubated in 10 mM 5-bromo-2-deoxuridine (BrdU) for 30 minutes on ice and then grown at 28.5°C until the indicated time points and BrdU incorporation was detected as described (Herzog et al., 2009). The vasculature of the Tg(kdrl:egfp)^s843 embryos was visualized using an anti-GFP Alexa 488 antibody (Invitrogen; 1:500).

The following antibodies were used: rabbit anti-zebrafish Cdh5 (Blum et al., 2008), mouse anti-human ZO-1 (TJP1) (Zymed), rabbit anti-zebrafish Podxl2 (Herwig et al., 2011) and Alexa 546 goat anti-rabbit, Alexa 568 goat anti-rabbit and Alexa 633 goat anti-mouse (Invitrogen). Immunohistochemistry was performed as described (Blum et al., 2008).

EC numbers were expressed relative to the average number counted in the CCVs of control embryos (set at 100%). The number of proliferating cells was quantified as the percentage of BrdU-positive among total ECs in the CCV. The length of the CCV was measured using the Measurement Point tool in Imaris. Using the manual Spot Track function in Imaris, the track length was calculated as the total length of displacements within the track, the migration speed was calculated as the track length divided by the time between the first and last object recorded, and the directionality was calculated as the ratio of the distance between the first and the last position and the track length.

Standard error of the mean (s.e.m.) and P-values using a two-tailed t-test were calculated using Microsoft Excel.

Tg(gata1:dsRed)^sd2 zebrafish embryos were grown in the presence of 30 mg/l phenylthiourea. The yolk was removed by pipetting in PBS and centrifugation at 100 g for 3 minutes. A single-cell suspension was prepared by incubating with 0.25% trypsin-EDTA at 28.5°C for 30 minutes, with subsequent inactivation of trypsin by addition of 1% fetal bovine serum and filtration through a 70 μm cell strainer. dsRed⁺ and dsRed^- cells were sorted on a FACSAria II (Becton Dickinson) cell sorter and collected separately.

RNA was isolated from sorted cells using the PicoPure RNA isolation kit (Arcturus). RNA was reverse transcribed using Superscript II reverse transcriptase (Invitrogen). The cDNA was analyzed using the following primers (5′-3′, forward and reverse): β-actin, AATCCCAAAGCCAACAGAGA and CTTCTGCATACGGTCAGCAA; gata1, ATTATTCCACCAGCGTCCAG and CTGGCCGTTCATCTTATGGT; vegfc, CAGAAACAAGGTCGGAGGAG and TTCGGCATTCACACTGACAC.

Anesthetized zebrafish embryos embedded in 1% agarose in gold-plated 200 μm deep specimen carriers were cryofixed by high-pressure freezing using a Leica HPM100. The samples were kept in liquid nitrogen until further processing.

For freeze substitution the samples were transferred to 1% OsO₄, 0.2% uranyl acetate, 5% water in acetone at -90°C. For 72 hours the sample was dehydrated and slowly fixed at low temperatures increasing to -30°C. The samples were then transferred to 4°C for 1 hour and then further processed by embedding using Epon (Schieber et al., 2010). Within areas of interest, 60 nm ultrathin sections were cut on an ultramicrotome (Leica UC6). Sections were counterstained with uranyl acetate and lead, and analyzed using a transmission electron microscope (Tecnai-12-biotwin, FEI, Eindhoven, The Netherlands). Images were documented on image plates (Ditabis, Pforzheim, Germany).

To analyze the development of the CCVs, we used in vivo confocal time-lapse imaging of the developing zebrafish trunk vasculature. We visualized individual angioblasts by transgenic GFP expression [Tg(fli1a:EGFP)^y1], which labels all ECs from the angioblast stage onwards.

We observed that the angioblasts in the trunk region of the zebrafish embryo are specified from posterior to anterior. Newly specified angioblasts are round in shape (arrowheads in Fig. 1B,D,F), before they change to an elongated form at the start of migration (arrows in Fig. 1B,E,F, Fig. 3A). At 14 hpf angioblasts are arranged in two parallel stripes within the lateral plate mesoderm (LPM). This first angioblast population migrates in towards the midline (arrows in Fig. 1B-D) and reaches its final position forming the LDA at 17 hpf (arrows in Fig. 1D, asterisks in 1E). At the same time, a second population of newly specified GFP-positive angioblasts with the characteristic round morphology (arrowheads in Fig. 1D-F) appears within the LPM. Some of these late-specified angioblasts start to migrate outwards to form the CCVs (arrows in Fig. 1F,G, Fig. 3A), whereas the other venous angioblasts form the bilateral cardinal veins (supplementary material Movie 1).

The time difference in the formation of the first and second angioblast populations is reflected by the temporal expression pattern of the early angioblast marker ets related protein (etsrp). Whereas angioblasts of the first population express etsrp at 12.5 hpf while they are located laterally within the LPM, this etsrp expression is lost after their migration to the midline (supplementary material Fig. S2). At 18 hpf the ECs forming the LDA are negative for etsrp (supplementary material Fig. S2D), but express other endothelial-specific genes such as cadherin 5 (vascular endothelial cadherin, cdh5) (supplementary material Fig. S2E). At this time point, the later-specified angioblasts forming the anterior cardinal veins and the CCVs strongly express etsrp (supplementary material Fig. S2D), indicating that they are a distinct population of newly specified angioblasts. etsrp expression in the bilateral cardinal veins and CCVs is downregulated by 24 hpf (supplementary material Fig. S2F). In our confocal time-lapse observations of angioblast specification and movement, we do not see any mingling of these two angioblast populations (supplementary material Movie 1).

To analyze when the ECs of the forming CCVs and the bilateral cardinal veins acquire venous identity, we performed in situ hybridizations for arterial and venous-specific genes. Although it is known that arterial marker genes, such as notch3 (Lawson et al., 2001) and ephrin B2a (Jin et al., 2005), only become expressed after angioblasts have migrated to their final position, we detected expression of flt1 and flt4 in all angioblasts of the forming trunk vessels (Fig. 2B,C,E,F,J-L, compare with cdh5 in 2A,D,G). At later developmental stages (from 21 hpf onwards), flt1 expression becomes restricted to the arteries (Fig. 2H) and flt4 expression becomes restricted to the veins (Fig. 2I).

We therefore found no indication that arterial or venous identity is conveyed by gene expression at the individual angioblast level. Instead, the difference we observed is in the timing of angioblast specification. This difference in timing correlated with a difference in the migration of the angioblasts to their final anatomical location. Additionally, the differentiation into arterial or venous ECs only occurred after the cells had reached the position of the final vessels.

To directly test whether the time of angioblast specification determines their migration behavior, we used transgenic endothelial-specific expression of Kaede, a photoconvertible fluorophore (Ando et al., 2002; Hatta et al., 2006). Photoconversion of transgenic fish was performed at 17 hpf, before the second population of angioblasts turned Kaede positive (Fig. 1H). At 26 hpf all photoconverted red-fluorescing ECs were located in the LDA, whereas all cells of the CCVs expressed only newly synthesized, unconverted Kaede (visualized in green, Fig. 1I; supplementary material Fig. S1).

Hence, we provide evidence that the precursors of the CCV undergo a heterochronic development compared with the angioblasts that later form the LDA, as they become specified at different time points of development and subsequently adopt different migratory routes, guiding them to separate final positions.

We continued to analyze the next steps in CCV development by confocal time-lapse microscopy. After the CCV angioblasts are specified within the second population, they migrate away from the midline between the yolk syncytial layer (YSL) and the epidermis and immediately start to align as a monolayer on top of the YSL (Fig. 3A-C′, Fig. 5A). Next, single angioblasts delaminate from the monolayer at the YSL side to the epidermal side (asterisks in Fig. 3E, Fig. 5B,C) to form a lumen before the onset of circulation (Fig. 3D-D′, Fig. 5D).

After delamination from the cellular sheet on top of the YSL (Fig. 3, Fig. 5B,C), the ECs forming the CCVs now align around a lumen and form open-ended tubes, which are sandwiched between the proximal YSL (Fig. 4B) and the distal epidermis (Fig. 4C). At the onset of blood flow, the CCVs are connected to the bilateral cardinal veins (midline side of the CCVs) and open towards the ventrolateral side, out of which the blood gushes into a space, which is free of ECs (Fig. 3D,D′, Fig. 4A,D; supplementary material Movie 2). The blood re-enters an EC-lined structure at the heart inflow tract (Fig. 4F,F′).

During further development, the ECs underneath the epidermis migrate collectively as a sheet (Fig. 3D′, Fig. 4C,J,K, Fig. 5E; supplementary material Movie 2) towards the endocardial ECs at the heart inflow tract (Fig. 4F,F′; supplementary material Movie 3). To analyze whether this rapid movement represents an active migration, we generated transgenic fish expressing lifeact-GFP (Riedl et al., 2008) under the control of a UAS promoter [Tg(UAS:lifeact-GFP)^mu271] and mated these with fish expressing Gal4 specifically in the vasculature [Tg(fli1ep:GAL4FF)^ubs4] (Herwig et al., 2011). Analysis of the double-transgenic offspring showed that the ECs forming the CCV indeed migrate actively, as indicated by the formation of actin-rich lamellipodia at the migration front (Fig. 4E,E′; supplementary material Movie 4).

To form the lumen, the ECs at the outer edges of the sheet start to fold inwards to engulf the blood stream (Fig. 4I,N, Fig. 5F). Consecutively, they fuse on the YSL side (Fig. 4L,L′; supplementary material Movie 5) in a dorsal to ventral sequence.

Interestingly, we could not detect apical EC polarization [analyzed by Podocalyxin-like 2 expression; supplementary material Fig. S4], which represents another significant difference to the previously described mechanisms of vascular lumen formation.

Our data demonstrate that in zebrafish the CCVs form via a previously undescribed mode of lumen formation, which we term ‘lumen ensheathment’ (Fig. 5).

Using electron microscopy and morpholino (MO)-based knockdown of gene function we analyzed the surrounding ultrastructures (Fig. 4M,N), the influence of blood flow (by tnnt2a MO; supplementary material Fig. S5B), possible co-migration with melanocytes (by mitfa MO; supplementary material Fig. S5D) and the role of the endoderm (by sox32 MO; supplementary material Fig. S5F) in the migration of CCV cells. None of these was found to influence EC migration in the CCV.

Studies in mice and zebrafish have revealed the indispensable roles of Cdh5 and blood flow during vascular lumen formation (Strilić et al., 2009; Wang et al., 2010; Herwig et al., 2011). We investigated the role of these factors during delamination and lumenization in the CCV. At 32 hpf homozygous cdh5^ubs8 mutant embryos showed no obvious morphological growth defects, but lacked circulation. Careful analysis revealed that in cdh5^ubs8 mutant embryos the ECs forming the CCV failed to delaminate from the migrating monolayer to form a lumen (Fig. 6B′).

When we examined the length of the CCVs in cdh5^ubs8 mutant embryos, we observed a reduction equivalent to the length of just one cell (Fig. 6B,E), indicating that at this developmental step EC migration is only minimally affected by Cdh5. Additionally, morphological cell adhesion was not impaired by lack of Cdh5 in the area of delamination (not shown). Our results indicate that Cdh5 is specifically required for the delamination that occurs during initial CCV formation.

To analyze the role of blood flow in lumen formation, we injected MOs targeting cardiac troponin T2a (tnnt2a) (Sehnert et al., 2002). Interestingly, the lack of blood flow does not influence the process of delamination or lumen formation in the CCV (Fig. 6D,D′).

Cadherin proteins have also been shown to be important during collective cell migration in epithelial sheets (Desai et al., 2009; Dupin et al., 2009). Therefore, we analyzed the expression of Cdh5 in the migrating CCV EC sheet. The ECs of the migrating sheet are connected by adherens and tight junctions (as shown by Cdh5 and ZO-1 staining; supplementary material Fig. S3).

Up to 55 hpf, the CCV has not connected to the endocardial cells of the heart inflow tract (data not shown) in cdh5^ubs8 mutant embryos [but has in the wild type (WT); supplementary material Movie 3]. Therefore, we analyzed the behavior of single ECs in the migrating sheet of WT and cdh5^ubs8 mutant embryos. Whereas in WT embryos the ECs in the collective sheet migrate in a highly directional manner towards the endocardium (Fig. 6F,H), this directionality is lost in cdh5^ubs8 mutant embryos (Fig. 6G,H), where cells changed their direction during migration frequently, although their velocity was unaffected (Fig. 6H).

To identify other molecular factors regulating CCV development, we analyzed the expression of various candidate genes. Of these, we detected domains of vegfc expression in the heart region and, with substantial variation between individual embryos, also along the CCV migration route (Fig. 7G, Fig. 8A). Vegfc signals via Flt4 (Vegfr3), which is strongly expressed in the developing veins (Fig. 7F). We therefore analyzed CCV development in embryos with altered Vegfc levels: either a vegfc mutant (vegfc^hu6410) or mRNA encoding a soluble version of human FLT4 (sFLT4) (Mäkinen et al., 2001) were used to lower Vegfc levels. Vegfc mRNA injection was employed to increase the level of Vegfc. Although loss of Vegfc did not influence EC migration (Fig. 7B), Vegfc appeared to regulate cell numbers in the developing CCVs. When Vegfc levels were elevated by vegfc mRNA injection, the CCVs were found to contain significantly more ECs than the CCVs of control embryos injected with H2B-cherry mRNA (Fig. 7C-E). Homozygous vegfc^hu6410 mutant embryos displayed significantly reduced numbers of ECs compared with their WT or heterozygous mutant siblings (Fig. 7A,B,E; supplementary material Fig. S6A,B). In line with an increase in EC number in the CCVs of embryos injected with vegfc mRNA, we found more proliferating ECs (Fig. 7E; supplementary material Fig. S6D). By contrast, loss or reduction of Vegfc led to a reduced percentage of proliferating ECs (Fig. 7E; supplementary material Fig. S6E).

We observed patchy vegfc expression and a large variation in the expression domains between individual embryos (compare Fig. 7G with Fig. 8A). We hypothesized that this variation could result from vegfc expression by circulating blood cells. Blood cells are commonly found unevenly distributed in the circulatory system upon fixation, with the exception of the heart, where blood cells are usually ‘trapped’ within the heart chambers. In order to identify which cell population is the source of Vegfc, we injected MOs to ablate either macrophages (pu.1 MO) (Rhodes et al., 2005) or erythrocytes (gata1 MO) (Galloway et al., 2005). In situ hybridization showed that vegfc expression was unaffected in embryos injected with control MO or pu.1 MO (arrow in Fig. 8B). However, ablation of erythrocytes by gata1 MO injection resulted in loss of the vegfc expression domain in the heart region (arrow in Fig. 8C). In addition, we isolated erythrocytes at 24 hpf and 48 hpf from Tg(gata1:dsRed)^sd2 transgenic embryos by FACS and performed RT-PCR analysis for vegfc on these isolated cell populations. This analysis confirmed that erythrocytes express vegfc at 48 hpf, whereas vegfc expression by the gata1⁺ cell population was not, or barely, detectable at 24 hpf (Fig. 8J).

Consequently, we examined whether vegfc expression by erythrocytes constitutes the physiological source of Vegfc for regulating EC numbers in the CCVs. We again ablated either macrophages or erythrocytes by MO injection and counted EC numbers in the developing CCVs. We observed a significant reduction in the number of ECs in gata1 MO-injected embryos (Fig. 8D), thereby identifying erythrocytes as a relevant source of Vegfc for CCV development. In agreement with these results, we also showed that the lack of blood flow, and therefore the complete lack of erythrocytes in the CCVs, results in a significant decrease in EC number in the CCVs (Fig. 8E).

Since the CCV is the largest vessel in the zebrafish embryo at this developmental time point (Fig. 8F-H), it is reasonable to speculate that the ECs in the CCV are exposed to much higher levels of erythrocyte-derived Vegfc than in other developing vessels. In addition, we noted that the onset of proliferation in ECs of the CCV coincides with the onset of circulation and therefore with the exposure to blood (Fig. 8I). At 24 hpf, ECs in the CCV downregulate the angioblast marker etsrp (supplementary material Fig. S2F) and ECs in the CCV start to proliferate (as analyzed by BrdU incorporation, Fig. 8I). This timing supports our notion that erythrocyte-derived Vegfc regulates EC proliferation and thereby EC numbers in the growing CCVs.

We have analyzed the development of the CCV, starting from the earliest stages of angioblast specification up to the generation of a lumenized and functional vessel. We show that this particular component of the venous system develops by a completely novel mechanism, which we have termed lumen ensheathment.

Previous reports have suggested that angioblasts are prespecified within the LPM into venous and arterial precursors (Zhong et al., 2001). In our marker expression analysis we found no difference in the expression of flt1 or flt4 in angioblasts forming the bilateral cardinal veins, the CCVs or the LDA. Instead, we observed that flt1 expression was downregulated during venous differentiation, whereas flt4 [as shown previously (Thompson et al., 1998)] was downregulated in arteries. Similarly, arterial expression of such as notch3, ephrin B2 (Lawson et al., 2001; Jin et al., 2005) or dll4 (data not shown) is only observed once the ECs start differentiating at the final position of the DA.

However, we detected a difference in the timing of angioblast specification by analysis of etsrp expression as well as by time-lapse analysis and photoconversion experiments. We therefore propose that the temporal separation of the two populations forming the anterior arteries and veins could result in their spatial separation and constitute the underlying mechanism for the differential expression of arterial and venous marker genes.

Whereas other reports have focused on arterial and venous development in the tail of the zebrafish embryo (Torres-Vázquez et al., 2003; Jin et al., 2005; Herbert et al., 2009), we have focused on venous development and CCV formation in the anterior trunk. Although it was recently suggested that in the caudal tail venous ECs segregate from the ECs of the DA (Herbert et al., 2009), we found no common precursor structure for the arterial and venous ECs in the trunk.

Within the trunk region of the zebrafish embryo the arterial and venous morphology, with two bilateral DA and two bilateral cardinal veins, closely resembles the initial murine morphology (Swift and Weinstein, 2009). Therefore, the principles described here might also apply during mammalian development.

Different mechanisms of lumen formation have been discussed (Lubarsky and Krasnow, 2003) and recent reports have highlighted the differences in lumen formation with respect to the various vessels (Kamei et al., 2006; Blum et al., 2008; Iruela-Arispe and Davis, 2009; Strilić et al., 2009; Herwig et al., 2011). However, all previously described mechanisms rely on angiogenic vascular growth while maintaining an EC-lined, closed luminal space at all times. Surprisingly, we found that the CCV forms without respecting this requirement. Instead, it grows by directly assembling a tube from migrating ECs, not through a cord hollowing step but by Cdh5-dependent delamination of individual ECs and their subsequent alignment around the future lumen. Although the role of delamination has been extensively studied for many developmental processes, this is the first report to show that delamination plays an important role during vascular lumen formation.

By this process, the ECs initially create an open-ended tube, allowing for blood to exit the closed vascular network, re-entering it again at the sinus venosus. Consecutively, the CCV becomes extended by active migration of the ECs located underneath the epidermis towards the ECs of the heart inflow tract. Our results show that Cdh5 is uniquely required for the directionality of the collectively migrating ECs in the CCV. To enclose the lumen, ECs at the edges of the migrating sheet bend inwards, folding around the luminal space and therefore the blood stream. This morphogenetic process is very similar to the closure of the neural tube (Colas and Schoenwolf, 2001; Pyrgaki et al., 2010), suggesting a conserved role of this mechanism in lumen formation.

Many invertebrates have an open cardiovascular system, with hemolymph being pumped from the heart through blood vessels into the open hemocoel (Strilić et al., 2010). One might therefore speculate that the initially open CCV reflects an evolutionarily primitive stage of cardiovascular development. However, even in the human placenta maternal blood leaves the closed vessel system and enters the intervillous space. This indicates that within the cardiovascular system there are highly specialized vessels or areas, and that understanding the formation and the specializations of the individual vessels will reveal additional principles that might provide insight into the various misregulations that occur during disease.

We also demonstrate for the first time that circulating erythrocytes can regulate EC proliferation by the secretion of growth factors, namely Vegfc. Vegfc is mainly known for its role in lymphangiogenesis and regulation of lymphatic vessel size (Rissanen et al., 2003; Goldman et al., 2005), but few reports have addressed the role of Vegfc during blood vessel angiogenesis (Covassin et al., 2006; Tammela et al., 2011). Of the hematopoietic cell lineages, previously only macrophages have been shown to aid vessel formation by Vegfc secretion (Tammela et al., 2011) or to regulate jugular lymph sac size (Gordon et al., 2010).

By contrast, we show for the first time that zebrafish erythrocytes can regulate EC proliferation. Zebrafish erythrocytes, like most vertebrate erythrocytes, contain transcriptionally active nuclei and are therefore able to produce and secrete growth factors. Similarly, mammalian erythrocytes, generated during primitive hematopoiesis, initially have nuclei. Although it has been proposed that they undergo enucleation in the circulation (Kingsley et al., 2004), this seems to occur long after they have entered circulation (McGrath et al., 2008). However, induction of vascular proliferation through growth factor secretion from erythrocytes might also play a role during early mammalian development. We show that deletion of the Pu.1 lineages, including all macrophages, leads to an increase in EC number of the CCVs (Fig. 8D), similar to the increase that results from vegfc overexpression. In zebrafish, deletion of the myeloid lineages leads to an increase in the erythroid lineage and vice versa (Galloway et al., 2005; Rhodes et al., 2005). We conclude that the higher EC number in the CCVs of embryos lacking macrophages is caused by an increase in erythrocytes and therefore higher levels of secreted Vegfc. Of note, there have been reports of other growth factors secreted into the blood stream, such as human BMP9, which was found in blood serum and shown to influence vascular quiescence (David et al., 2008).

In summary, our findings illustrate that angioblasts forming the CCVs initially share flt1 and flt4 expression with the angioblasts of the LDA, but differ in their time point of specification and consequently in the onset of migration. Additionally, we identify a novel mode of lumen formation, in which ECs of the future CCVs migrate, delaminate in a Cdh5-dependent manner and then align around an existing luminal space. Next, they migrate along the dorsal periphery of this lumen and thereby extend the CCV, at a time when circulating blood cells are already passing through. During this process, Cdh5 is important for the directionality of the collective EC migration. Subsequently, the CCV is completed by dorsal to ventral lumen enclosure. Although this process of CCV formation is flow independent, the proliferation of ECs within the CCV is regulated by erythrocytes that provide Vegfc.

We thank Aurelien Courtois and Jeroen Bussman for helpful discussion and suggestions; Ralf Adams and Friedemann Kiefer for critical comments on the manuscript; Stefania Nicoli for advice regarding the FACS sorting; Martin Stehling for FACS sorting; Reinhild Bußmann for fish husbandry; Karina Mildner for technical support for the electron microscopy analysis; Erez Raz for the lifeact-GFP plasmid; and Arndt Siekmann for providing the gata1 MO.